The setup, in plain English

Medicare wrote the manual, then forgot to translate it

Contract heterogeneity, or why one provider’s “VBC contract” is actually fifteen

Quality measure overload and the normalization problem

Risk adjustment vs. MA, different sport, same jersey

Specialty is where the money actually lives

Employers turned into active buyers, not passive bag-holders

Where the tooling gap turns into a real business

The bigger pattern, and why this is finally investable

Abstract

Core thesis: Commercial value-based care has crossed the line from theoretical to operational, but the infrastructure required to run it at scale is roughly a decade behind Medicare. Primary sources from the largest commercial payers now show actual scale, real dollars, and explicit specialty expansion, which means the next generation of healthcare infrastructure companies will be built to operate inside this fragmentation, not around it.

Key data anchors used throughout:

- Cigna Collaborative Care: 230+ primary care arrangements across 32 states, 2.65M+ commercial customers, 144,000+ contracted physicians (81K PCPs, 63K specialists)

- Cigna’s stated specialty focus: cardiology, GI, OB-GYN, oncology, orthopedics, representing 57% of medical spend

- Elevance: value-based arrangements >60% of medical expense (2021 reporting)

- Blue Shield of California: 56% of cost-of-care spend in pay-for-value, $31M net 2023 ACO savings, 7.4% cost reduction in Primary Care Reimagined, 7.4% savings in orthopedic episodes

- Blue Cross NC: Episodic Bundle Payment Plan live as of Jan 1, 2025 for knee, hip, and shoulder replacements with a single member copay

- Highmark behavioral health: 24.7% PMPM cost reduction commercial, 45.9% PMPM Medicare, in a value-based collaboration with Value Network IPA

- BCBSMA: pay-for-equity layered onto the Alternative Quality Contract

- Highmark + Stellar Health: VBC enablement spanning Commercial and MA lines

Categories of opportunity examined:

- Multi-payer contract intelligence and management

- Quality measure normalization

- Specialty bundle administration (orthopedics, oncology, cardiology, OB, GI, behavioral)

- Attribution and patient-panel engines for commercial books

- Risk and stop-loss hybrids

- Case-rate billing infrastructure

- Performance analytics for independent specialty groups

The setup, in plain English

For roughly fifteen years, the entire conversation about value-based care in this country has effectively been a Medicare conversation dressed up in different outfits. ACOs, MSSP, Pioneer, Next Gen, Direct Contracting, ACO REACH, Medicare Advantage capitation, the various bundled payment demos out of CMMI: all government, all anchored in CMS rulemaking, all measured against a Medicare fee schedule that everyone in the industry treats as the gravitational center of payment design. The commercial side, by contrast, has spent most of that time being characterized as a slow follower, a payer-by-payer mishmash of pay-for-performance bonuses and half-hearted ACO knockoffs that nobody on a board deck took particularly seriously. That framing is now obsolete. The published material from the largest commercial insurers in the country tells a different story, and it is worth taking that story at face value rather than continuing to describe commercial VBC as a Medicare spillover.

Run through the receipts and the picture changes pretty fast. UnitedHealth Group’s October 2025 white paper on advancing value-based care explicitly extends the program logic to commercial health plans and employers and frames an all-payer alternative payment model context. Cigna has more than 230 Collaborative Care arrangements with primary care groups across 32 states, covering more than 2.65 million commercial customers and contracts with over 144,000 doctors, including roughly 63,000 specialists. Elevance has stated that value-based arrangements represented more than 60% of medical expense in 2021. Blue Shield of California has publicly stated that 56% of its cost-of-care spend now flows through pay-for-value models, with a stated ambition to push that number to 90%. Blue Cross NC went live January 1, 2025 with the Episodic Bundle Payment Plan for knee, hip, and shoulder replacements at a single member copay. BCBSMA has layered equity-linked incentive payments on top of its Alternative Quality Contract. Highmark’s behavioral health value-based arrangement with Value Network IPA has produced a 24.7% PMPM reduction for commercial members and a 45.9% PMPM reduction for Medicare. None of this is a press release fantasy. These are operational programs with attribution engines, contract clauses, settlement timelines, and dollar flows attached to them.

So commercial VBC is real. The interesting question is no longer whether it is real, but what kind of market structure it actually creates. And the short version of the answer is that it creates a fundamentally different beast from Medicare, one that is much messier, much more fragmented, much harder to operate inside, and therefore much more interesting for anyone trying to build infrastructure businesses on top of it.

Medicare wrote the manual, then forgot to translate it

Read more

- Thesis: PA is shifting from labor-arbitrage cost containment to API-driven infrastructure; value is migrating from BPO and in-house staff to software, data, and middleware

- Regulatory stack: CMS-0057-F (Jan 2024 final, Jan 2027 compliance, four FHIR APIs), CMS-0062-P (Apr 2026 proposed, Oct 2027 proposed compliance, drug PA), ONC HTI-1 (EHR certification), AHIP/BCBSA June 23, 2025 commitments covering 257M Americans, state laws in WA, NE, TX, AR, MD, CA, VA, IN, AK, MI, IA, IL, VT, CO

- Scale of dysfunction: 39 PAs/week per physician, 13 hrs/week of physician+staff time, 40% of practices have dedicated PA staff, 29% of docs report a serious adverse event from PA delay, only 35% of PAs processed electronically (CAQH), only 9% of surveyed orgs could support an ePA API by 2027 (CAQH), nearly half of PA requests still submitted by fax or phone per BCBSA Jan 2026, $515M annual CAQH savings estimate, $15B CMS 10-yr savings estimate

- Critical regulatory unlock: CMS granted enforcement discretion in 2024 allowing an all-FHIR PA workflow in place of the legacy X12 278, effectively waiving the HIPAA X12 mandate for PA and opening the door to pure-FHIR infrastructure

- Four CMS-0057-F APIs: Patient Access, Provider Access, Payer-to-Payer (the mechanism for 90-day continuity carryover, shipping up to 5 years of claims, encounters, USCDI data, and PA history between plans on patient opt-in), Prior Authorization

- Technical stack: HL7 Da Vinci PA suite (CRD for discovery, DTR for documentation, PAS for submission/adjudication), FHIR-to-X12 bridge available but no longer required, CDS Hooks cards in EHR workflow

- AI boundaries: state laws (MD HB0820, CA SB 1120, NE LB 77, CA SB 363) require individualized basis, mandate human decision-making, require quarterly audits, impose up to $1M per case fines for high appeal-overturn rates; nH Predict (UnitedHealth) and PXDX (Cigna) as cautionary tales

- Opportunity map: provider-side orchestration, payer-side FHIR middleware (“PA as a service”), PA portability/Payer-to-Payer layers, AI audit and compliance tooling, PA-aware clinical decision support, patient-facing PA transparency and appeal assistance, specialty benefit manager modernization, and monetization of the public-reporting PA dataset

- Named players: provider-side (Cohere, Rhyme, Infinitus, Myndshft, Infinx, Anterior, Itiliti Health); payer infra (Smile Digital Health, Firely, HealthEdge, CareEvolution); specialty benefit managers (Evolent, eviCore, Carelon, Lumeris) as fragmented modernization targets

- Honest risks: payer foot-dragging via thin FHIR veneers, dual-stack necessity during transition, gold card complexity varying across states, consolidation pressure from clearinghouses and EHRs, utilization rebound as PA becomes frictionless

Table of Contents

How Bad Is It Actually

The Adoption Gap Nobody Wants to Talk About

The Federal Rails and the Four APIs of CMS-0057-F

The Quiet Unlock: CMS Waives X12 for PA

CMS-0062-P and the Drug Rule

The June 2025 AHIP Announcement and Why 257M Is the Real Number

States Are Writing the SLA

Gold Carding as a Data Problem in Disguise

The 90-Day Carryover Is a Payer-to-Payer API Problem

AI Guardrails and Where the Lawsuits Are Pointing

The Da Vinci Stack in Plain English

Where the Plumbing Actually Breaks

What AI Can and Cannot Do Here

Provider-Side Orchestration as a Business

Payer-Side FHIR Middleware

The Portability Layer Nobody Owns Yet

Compliance Tooling for the Algorithmic Audit Era

PA-Aware Clinical Decision Support

Patient-Facing PA Transparency and Appeal Tools

Specialty Benefit Manager Modernization

The Dataset Is the Real Prize

Incumbents, New Entrants, and Where the Whitespace Is

Honest Risks and What Could Go Sideways

Closing Thoughts

How Bad Is It Actually

Before getting into the opportunity, it helps to sit with how genuinely broken the current state is, because the scale of the dysfunction is what makes the rebuild economically obvious. The AMA’s 2024 Prior Authorization Physician Survey, which pulled responses from 1,000 practicing docs, reads like a clinical fever chart. Physicians are knocking out an average of 39 PA requests per week, burning roughly 13 hours of combined physician and staff time to do it. Forty percent of practices have dedicated PA staff who do literally nothing else. About one in three physicians say their requests get denied often or always, and 75 percent say the denial volume has climbed over the past five years.

Then there’s the part that should make any policymaker queasy. Ninety-three percent of physicians report that PA delays necessary care, 82 percent say it leads to treatment abandonment, and 29 percent, which is more than one in four, report that PA has caused a serious adverse event for a patient under their care. Hospitalizations in 23 percent of cases. Life-threatening events in 18 percent. Permanent disability or death in 8 percent. For a process whose entire justification is cost containment, the morbidity tail is unusually loud.

The money side is just as ugly. The CAQH 2024 Index pegs the savings from full ePA adoption at $515 million annually, with 14 minutes of staff time shaved per authorization compared to the current mixed bag of portals, faxes, and the occasional API. CMS has projected $15 billion in savings over ten years from its interoperability rule alone. Those numbers are the market size. They explain why every incumbent clearinghouse, EHR vendor, UM platform, and health tech seed company is suddenly interested in a transaction type they mostly treated as infrastructure plumbing five years ago.

The Adoption Gap Nobody Wants to Talk About

The most damning numbers in the CAQH 2024 Index are not the savings figures. They are the adoption figures. Only 35 percent of prior authorizations are currently processed electronically at all. That is not “not fully automated.” That is “roughly two-thirds of all PA transactions in American healthcare still move on paper, fax, portal, or phone in 2024.” Every other major administrative transaction in healthcare (eligibility, claims, remittance) has adoption rates above 90 percent. PA is the last unreconstructed workflow in the whole X12 catalog.

The second number is worse. Only 9 percent of surveyed organizations reported that they could currently support an ePA API of the kind CMS-0057-F mandates by January 1, 2027. Nine percent. That is not a compliance gap, that is a compliance chasm, and it is the single most important statistic in the entire PA economy. It says the industry has roughly eight months of actual build time from the publication of this essay to stand up infrastructure that 91 percent of surveyed orgs cannot yet support. Either the rule gets watered down or delayed, which is possible but not assumed, or the next 14 months are going to be a historic scramble for FHIR-native middleware, systems integrators, and anyone who can plausibly claim to ship production-grade FHIR endpoints at payer scale. Market-sizing exercises that ignore the 9 percent number are missing the scale of the urgency.

BCBSA in a January 2026 article confirmed the provider-side picture bluntly: nearly half of PA requests are still submitted by fax or phone. Half. In 2026. At payer plans that are publicly committed to 80 percent real-time ePA by January 2027. The gap between the aspirational commitment and the operational reality is roughly the size of the opportunity.

The Federal Rails and the Four APIs of CMS-0057-F

The piece that actually forces the rebuild is CMS-0057-F, the Interoperability and Prior Authorization Final Rule that CMS dropped in January 2024. It applies to Medicare Advantage organizations, state Medicaid and CHIP fee-for-service programs, Medicaid managed care plans, CHIP managed care entities, and Qualified Health Plan issuers on the federally-facilitated exchanges. The operative word is breadth. These plans collectively cover a huge slice of the insured population, and by January 1, 2027 they all have to implement a Prior Authorization API built on HL7 FHIR standards. In addition, payers must send PA decisions within seven days for standard requests and 72 hours for urgent ones, and provide a specific reason for any denial.

What tends to get lost in casual coverage is that CMS-0057-F actually requires four distinct FHIR APIs, not one. The Prior Authorization API gets the headlines because it is the operational showpiece, but it is only one layer of the stack. The other three are just as important to the economics of the new PA economy, and each of them underpins a different business model category.

The Patient Access API requires payers to make claims, encounter data, formulary information, and (by 2027) PA information available to members through third-party apps on request. This is the rail that lets patients see their own PA status, approvals, denials, and the reasons for those decisions in real time through consumer-grade apps. The Provider Access API requires payers to share claims, encounters, and PA data with in-network providers for attributed members, which gives clinicians longitudinal visibility into prior care and prior PA history when they are making a decision. The Payer-to-Payer API is the most operationally consequential for the portability business model: when a member opts in, the previous payer must ship up to five years of claims, encounter data, USCDI-aligned clinical data, and PA history to the new payer. That is the technical mechanism by which the 90-day continuity of care carryover actually happens. And the Prior Authorization API is the real-time decisioning surface.

These four APIs together constitute the FHIR rail layer. Every business model in the new PA economy sits on top of one or more of them. Treating “the PA API” as the whole story is like treating the checkout step as the whole story of e-commerce. The sleeper provision in CMS-0057-F, separate from the APIs themselves, is the public reporting mandate. Payers have to annually report approval rates, denial rates, average decision timeframes, and the percentage of requests approved only after appeal. This is the first time in American health insurance history that PA behavior becomes a standardized, comparable, public dataset. The implications for analytics, benchmarking, and market access products are enormous and mostly unpriced.

ONC’s HTI-1 final rule closes the loop on the EHR side by updating the Health IT Certification Program to require certified EHRs to support FHIR-based data exchange. You can have all the payer APIs in the world, and if the EHR cannot consume them cleanly, the value never reaches the clinician. HTI-1 fixes that on the certification side, though actual vendor implementation quality is a separate and more cynical conversation.

The Quiet Unlock: CMS Waives X12 for PA

This is the piece of the regulatory package that does not get nearly enough attention outside the standards community, and it is arguably the single most important enabler of the new infrastructure category. Under HIPAA, electronic PA transactions have historically been required to use the X12 278 standard. That requirement is what kept the clearinghouse ecosystem in the middle of every PA, because clearinghouses are the plumbing layer for X12 EDI. Any company trying to build a pure-FHIR PA product was either violating HIPAA or had to maintain a FHIR-to-X12 bridge for the electronic transaction of record.

In 2024, alongside CMS-0057-F, CMS granted enforcement discretion allowing payers and providers to use an all-FHIR PA workflow in place of the X12 278. In effect, this waives the HIPAA mandate for PA specifically. The regulatory permission slip to skip the clearinghouse entirely is now issued. Any pure-FHIR infrastructure company that stands up a Da Vinci-compliant PAS pipeline can now legally operate as the full transaction rail for PA without needing to round-trip through X12 or a traditional clearinghouse. That is not an incremental improvement. That is the moment a whole category of legacy intermediaries becomes bypassable by regulation rather than by ambition.

This is why the Da Vinci PAS guide defines a FHIR-to-278 translation as optional for backwards compatibility rather than as a required bridge. Companies can run pure FHIR end to end. The clearinghouses know this, which is why their FHIR investments are accelerating. But by the time most of them ship FHIR-native product, several of their largest transaction categories will have already migrated.

CMS-0062-P and the Drug Rule

In April 2026 the Trump administration followed up with CMS-0062-P, a proposed rule that extends the same framework to prescription drugs. Impacted payers would have to support electronic PA for medications, adopt updated FHIR and NCPDP SCRIPT standards for drug transactions, and report their interoperability API endpoints and usage metrics back to CMS. The deadlines on the drug side are tighter. Medicaid plans would get 24 hours to respond to drug PA requests. ACA marketplace plans would get 72 hours for standard and 24 for expedited. The comment period runs until June 15, 2026, with proposed compliance in October 2027. Assuming it survives comment substantively intact, the drug rule locks in what the medical rule established: PA is now a federally regulated, API-delivered, time-bound transaction, and pharmacy is next.

The June 2025 AHIP Announcement and Why 257M Is the Real Number

The commercial sector, which covers most working-age Americans, is largely outside the direct scope of CMS-0057-F. The industry knows that if it does not move voluntarily, Congress or a future administration eventually will. On June 23, 2025, AHIP and the Blue Cross Blue Shield Association announced a joint voluntary commitment, built in partnership with HHS and CMS, designed to streamline PA across the commercial sector.

The headline number is the one that matters: the participating plans collectively cover roughly 257 million Americans. That includes UnitedHealthcare, Cigna, Aetna, Elevance Health, Humana, Centene, Kaiser Permanente, and the full Blue Cross Blue Shield federation. That roster is effectively every payer in the country that matters at scale. When a voluntary commitment covers 257 million lives, it stops being a PR gesture and becomes a de facto industry standard. It is also a flare to regulators and Congress that further rulemaking is unnecessary, which is the political function of such commitments generally.

The announcement laid out six specific commitments with staggered 2026 and 2027 effective dates. Honoring existing approvals for at least 90 days when a patient switches plans. Reducing the volume of PA requirements by identified service categories. Providing clear communication and transparency on denials and appeal rights. Expanding real-time responses across plans, culminating in 80 percent of electronic PA requests being addressed in real time by January 1, 2027. Ensuring continuity of care during transitions. And committing to specific measurement and public reporting of performance.

The April 2026 progress report was the first real read on whether the commitment was being executed. Participating plans reported eliminating 11 percent of prior authorizations across various medical services, which translated to 6.5 million fewer requests for patients. In Medicare Advantage the reduction was over 15 percent. BCBSA CEO Kim Keck re-committed the Blues specifically to the 80 percent real-time ePA target by January 1, 2027. Real time is the key phrase. Real time at the point of care is not a workflow improvement. It is an infrastructure requirement. It rules out batch, portal scraping, and any architecture that cannot guarantee sub-second response for the common cases. Hitting 80 percent real time on ePA forces FHIR-native backends, and a 257-million-life cohort of commercial payers is now collectively on the hook for that build by January 2027.

States Are Writing the SLA

While the federal and industry layers are setting the direction, state legislatures are the ones actually prescribing the operational constraints. The December 2025 NAIC Prior Authorization White Paper grouped state activity into four buckets: turnaround time requirements, gold carding, continuity of care protections, and AI guardrails. Each of them turns a discretionary administrative tool into a regulated SLA product, which is the exact moment when a workflow becomes a software market.

Washington State’s ESSHB 1357, passed in 2023, is the most technically prescriptive. It required health plans to build and maintain a FHIR-based PA API by January 1, 2025, with 72 hours for urgent and 7 days for standard turnaround. Miss the deadline and the request is automatically approved. Auto-approval on SLA miss is a bright line that effectively converts the SLA from aspirational into a liability-bearing legal obligation, because now a missed deadline has a measurable dollar cost. Nebraska’s LB 77 in 2025 did something similar, setting 72 hours for urgent and 7 days for non-urgent with auto-approval on miss, tightening urgent down to 48 hours starting in 2028.

Other states have piled on. Indiana’s SB 0480 requires 48 hours for urgent and 5 business days for non-urgent, and mandates electronic receipt of PA requests. Alaska’s HB 0144 requires 72 hours for standard and 24 hours for expedited. Michigan imposes 72 hours for urgent and 7 days for standard, with automatic approval on a missed deadline, using the same liability mechanism as Washington. Iowa’s HF 303 sets 48 hours for urgent and 10 days for non-urgent, and requires annual reporting to the state insurance commissioner. Illinois and Vermont have adopted 90-day continuity windows for plan switches, tracking the AHIP/BCBSA commitment. Colorado has passed rules allowing multi-year authorization validity for stable chronic regimens, which is operationally significant for specialty drug access. And California’s SB 363 goes in a different direction entirely by imposing denial-rate disclosure requirements and fines of up to $1 million per case where more than 50 percent of appeals are overturned. That last provision is effectively a statutory penalty for overly aggressive algorithmic denial, and it is going to be cited in every investor deck for an AI audit company for the next three years.

Gold Carding as a Data Problem in Disguise

Texas kicked off the gold carding trend with HB 3459 in 2021, which exempts physicians from PA for specific services if their approval rate over the previous 12 months is 90 percent or higher. The 2025 amendment, HB 3812, extended the evaluation window from six months to a full year and forced insurers to submit annual gold-card data reports to the Texas Department of Insurance. Arkansas followed with HB 1301 in 2025, applying gold carding at the group practice level.

Gold carding is usually sold as a patient-access mechanism, which it is, but operationally it is a data problem. To run a compliant gold card program at scale, a payer has to track provider-level approval rates by service category across rolling 12-month windows, with enough granularity to identify when a provider crosses the threshold and enough auditability to defend the calculation to a state regulator. That is the definition of a structured longitudinal dataset. Once a payer has that, the same infrastructure can be resold back to providers as a benchmarking tool, to employer groups as a network-performance metric, and to life sciences as a provider-segmentation input. Gold carding accidentally creates a provider-performance database as a regulatory byproduct.

The 90-Day Carryover Is a Payer-to-Payer API Problem

State laws are increasingly protecting patients from care disruptions when they switch plans. The AHIP and BCBSA commitments include a 90-day transition period where an existing PA must be honored for benefit-equivalent, in-network services. Virginia’s HB 736 from 2026 goes further, requiring that initial PAs remain in effect for at least six months and continued requests for at least 12 months. Nebraska’s LB 77 provides that approved PAs are valid for one year in most cases and can follow a patient for 60 days after a plan switch. Illinois and Vermont both now have statutory 90-day continuity windows. Colorado allows multi-year authorization validity for stable chronic regimens.

The mechanism that makes all of this work is the Payer-to-Payer API in CMS-0057-F. The statutory text of the carryover commitments presumes that an existing PA and its clinical documentation can physically move between payers in a timely, reliable, machine-readable way. That does not happen on its own. The Payer-to-Payer API is the rail: when a member opts in, the old payer ships up to five years of claims, encounters, USCDI clinical data, and PA history to the new payer in FHIR. Without that data flow, the 90-day honoring commitment reduces to the new payer taking the patient’s word for what was previously approved, which no payer will do at scale.

Think about what this actually requires in code. A PA approval is a state object. It has attributes: service, duration, clinical justification, approving clinician, expiration date, and associated documentation. It has to be portable across competing payers. It has to survive plan switches, network changes, and PBM transitions without losing integrity. Building the rails to securely transfer that state object between payers that are actively fighting for the same members is not a feature. It is a neutral intermediary layer, and no incumbent wants to be the one who builds it for the other side. That is a greenfield opportunity disguised as a compliance burden, with the Payer-to-Payer API as the underlying transport.

AI Guardrails and Where the Lawsuits Are Pointing

For anyone building AI for utilization management, the most important state-level trend is the wave of laws explicitly regulating algorithmic decision-making in PA. At least five states have enacted specific guardrails. Maryland’s HB0820, effective October 1, 2025, closely tracks California’s SB 1120 from 2024. Both require carriers, PBMs, and private review agents to ensure that any AI tool used in utilization management bases its coverage decisions on the enrollee’s individual medical and clinical history, not on group or demographic statistics. The Maryland law mandates at least quarterly reviews of any AI used in UM, requires carriers to report metrics on AI use in adverse decisions, and makes AI tools available for audit by the insurance commissioner. Violations can trigger misdemeanor charges, monetary penalties, and revocation of certificates. Nebraska’s LB 77 goes a step further and prohibits AI as the sole basis of a denial, requiring that adverse decisions be made by a physician. California’s SB 363 adds denial-rate disclosure and the million-dollar-per-case fine structure for high appeal-overturn rates.

These laws did not emerge in a vacuum. UnitedHealthcare’s nH Predict algorithm, used to set appropriate lengths of post-acute care for Medicare Advantage patients, has been the subject of class action litigation and congressional attention. Denial rates allegedly climbed from 10.9 percent in 2020 to 22.7 percent in 2022, a period that coincided with UnitedHealth’s acquisition of naviHealth, which built the tool. Plaintiffs claim the algorithm has a 90 percent error rate and routinely overrides physician recommendations. A federal judge in April 2026 ordered UnitedHealth to produce AI claim denial documents in discovery, which is going to be interesting to watch.

Cigna’s PXDX system, exposed in a 2023 ProPublica investigation, allegedly allowed medical directors to deny hundreds of thousands of claims per month without reviewing individual patient files by auto-flagging claims that did not match a list of pre-approved condition-procedure pairs. Reporting suggested a single medical director could push through more than 50 denials in seconds. Whether you view these tools as efficient triage or algorithmic rubber-stamping, the regulatory consensus has crystallized: AI can assist, but cannot be the decider. Any architecture that assumes otherwise is betting against the direction of every relevant legislature and most relevant plaintiffs’ firms.

The Da Vinci Stack in Plain English

The technical foundation for all of this is the HL7 Da Vinci Project, a private-sector standards initiative that brought payers, providers, and EHR vendors together to build FHIR-based implementation guides for value-based care workflows. CMS formally endorsed the Da Vinci PA guides as the basis for CMS-0057-F compliance, which means the Da Vinci PA suite is now effectively regulation by reference.

The ePA stack has three pieces. The first is Coverage Requirements Discovery, or CRD. When a clinician initiates an order or referral inside the EHR, a CRD call fires in the background to the patient’s payer and asks, in effect, does this service require PA for this member on this plan. The payer replies with a CDS Hooks card that drops into the clinician’s workflow and tells them before the order is locked in. Catching the PA requirement at the point of decision is the single highest-leverage intervention in the whole workflow, because it is the moment at which alternatives, documentation, or a different service choice are still possible.

The second piece is Documentation Templates and Rules, or DTR. Once PA is confirmed as needed, the DTR API pulls the payer’s specific clinical documentation requirements and business rules and presents them to the provider as a SMART on FHIR app or a structured questionnaire. DTR can also pre-populate answers by pulling from the patient’s existing clinical data in the EHR, which is where AI documentation synthesis has its biggest near-term footprint.

The third piece is Prior Authorization Support, or PAS. This is the submission layer. Once DTR has assembled the packet, PAS bundles the clinical data with the PA request using FHIR resources, submits it to the payer, and returns either a real-time decision or a tracking ID for pending cases. Historically the PAS guide included a FHIR-to-X12 278 mapping for backwards compatibility, but with the 2024 CMS enforcement discretion that mapping is optional rather than required. The response comes back as a structured FHIR resource, which means the EHR can auto-update the record and the workflow without human re-entry. When it works end to end, the whole thing looks less like prior authorization and more like a payment authorization on a credit card network, which is broadly the right mental model.

Where the Plumbing Actually Breaks

The architecture is clean on paper. Real life is messier. Even with the CMS waiver on X12, the X12 278 transaction still dominates current production UM environments and most clearinghouses are still overwhelmingly X12-based. The transitional reality is multi-hop architectures, latency budgets blown on translation, and error surfaces at every mapping boundary. Every translation is a place where bugs live.

The AMA numbers on the provider side are ugly. Only 23 percent of physicians say their EHR offers ePA for prescriptions, and 30 percent say the PA requirement information in their EHR is rarely or never accurate. Which means even where payers do stand up FHIR endpoints, the provider-side tooling to consume them is frequently absent, broken, or stale. Layer on BCBSA’s January 2026 admission that nearly half of PA requests are still submitted by fax or phone, and the picture is that fax machines and portal scraping are persistent because the alternative has not reliably arrived at the clinician’s desktop. The 2024 CAQH Index’s 35 percent electronic processing figure says the same thing from a different angle. And the 9 percent of surveyed orgs that can support a 2027-compliant ePA API says it from a third.

The clearinghouse layer is the other structural drag. Clearinghouses sit between providers and payers, routing transactions and translating formats. Most of the majors are optimizing their X12 infrastructure rather than rebuilding FHIR-native. That creates a real risk that the new FHIR rails get routed through legacy clearinghouse plumbing that neutralizes the whole latency premise of real-time adjudication. The CMS enforcement discretion theoretically lets anyone skip this layer entirely for PA. The question is who actually builds a credible alternative at payer scale fast enough to matter.

What AI Can and Cannot Do Here

AI is both the most interesting and the most dangerous layer in the new PA economy. On the provider side, AI genuinely shines in orchestration and synthesis. Give it a patient’s unstructured clinical notes, a payer’s DTR rule set, and a target service, and modern models can reliably identify the criteria, map documentation to them, and auto-populate the questionnaire. The clinician is still in the loop; the AI is doing the tedious evidence-assembly that a human has been doing with a highlighter and a prior fax for 20 years. That work is low-controversy, high-value, and straightforward to productize.

On the payer side, AI can route incoming requests intelligently, flag missing documentation, identify clear auto-approvals based on objective guideline adherence, and surface patterns that warrant clinician review. CMS itself has begun incorporating AI into review workflows, with an explicit requirement that licensed clinicians sign off on final decisions.

Where AI breaks is where vendors try to collapse clinical judgment into a model. The nH Predict and PXDX cases are not abstract: they are the templates for the next decade of litigation. The AMA survey found 61 percent of physicians are worried AI will increase denial rates, and there is no regulatory trajectory anywhere in the country in which AI gets to be the sole decider on an adverse determination. The durable business model is AI as triage and documentation, clinicians as deciders, with clean audit trails that prove which role each played. Anyone architecting the opposite will eventually be a discovery exhibit.

Provider-Side Orchestration as a Business

On the provider side, the immediate pain is unambiguous. Practices are spending $20 to $30 per PA transaction, and 40 percent of them have dedicated PA staff. The market for end-to-end orchestration is huge, underserved, and not yet consolidated.

The shape of the winning product is by now pretty clear. Sit on top of the EHR via SMART on FHIR or a direct integration. Use AI to ingest clinical notes and map them against payer-specific rules, using CRD and DTR where available and falling back to portal scraping or direct payer connectivity where they are not. Auto-package and submit the PAS request. Track state in real time. Close the loop back into the EHR with a structured result. The moat is coverage breadth. Any orchestration tool that only handles FHIR-native payers is a toy, because roughly half of PA volume is still fax or phone per BCBSA. The one that handles all payers and all service types, drugs included now that CMS-0062-P is bringing pharmacy PA into the electronic framework, is the one that wins the practice.

The competitive field on provider-side orchestration is already busy. Cohere Health, Rhyme, and Infinitus are among the better-known names. Myndshft, Infinx, Anterior, and Itiliti Health are each executing on meaningful slices of the workflow, with different specialty depths and different integration footprints. The category is not yet consolidated and specialty-specific orchestration (oncology, advanced imaging, MSK, high-cost drugs) still has whitespace where the generic horizontal tools tend to underperform.

The interesting second-order revenue model here is gold carding. An orchestration platform that consistently gets its providers to 90 percent approval rates across the service categories that matter is effectively shrinking the total addressable volume of PAs for those providers, which is an unusual and potentially disruptive product promise. It also makes the state-by-state complexity of gold card programs a moat, because a national tool that can track and administer gold card status across Texas, Arkansas, and any future state that enacts a program is meaningfully harder to build than it looks.

Payer-Side FHIR Middleware

Payers have a hard deadline and a hard problem. CMS-0057-F requires four FHIR APIs by January 1, 2027. Most legacy UM stacks were built on mainframes and X12 EDI and physically cannot hit sub-second response times. A 2026 Becker’s Payer survey found 28 percent of payers estimated spending $1 million to $5 million just on API implementation, which probably undercounts the true total cost of ownership once clinical policy digitization, integration testing, and ongoing SLA monitoring are priced in. The CAQH 9 percent readiness figure puts a number on the gap: the overwhelming majority of payers cannot actually ship to the rule as currently written.

That gap is the market for white-labeled FHIR middleware, or, as the pitch decks are now calling it, Prior Auth as a Service. The product ingests inbound PAS requests, runs them against digitized clinical guidelines (InterQual, MCG, internal policies), auto-approves the clear cases, and intelligently routes the gray ones to medical directors. The payer keeps control of clinical policy. The vendor handles the FHIR compliance, the latency engineering, and the public-reporting data layer. This is fundamentally a distributed systems problem, not a clinical one, which is why cloud-native engineering teams will likely outcompete legacy health IT vendors on this layer. The AHIP and BCBSA 80 percent real-time commitment is functionally a performance spec for the middleware category.

The infrastructure layer already has credible players, though none that has yet consolidated the market. Smile Digital Health (formerly Smile CDR) and Firely are the most prominent pure-play FHIR infrastructure vendors, with deep HL7 engineering benches and broad payer deployments. HealthEdge and CareEvolution are each building middleware-adjacent products aimed at the same 2027 deadline. None of them is an obvious default yet, which is unusual for a market with a hard deadline 14 months out.

The Portability Layer Nobody Owns Yet

The 90-day carryover from AHIP and BCBSA, combined with state rules like Nebraska’s 60-day portability, Virginia’s minimum duration requirements, and the Illinois and Vermont 90-day windows, creates a structural need for a neutral PA portability layer. The Payer-to-Payer API is the plumbing, but the plumbing needs an operator. When a member switches plans, the new payer has to receive the existing approvals and the clinical documentation that justified them in a machine-readable, standardized format, and the mechanism has to work across every combination of incumbent and recipient payer. The two payers are competitors. They have no existing data-sharing relationship. Neither wants to be the one to build the exchange that honors the other’s decisions at scale.

That is the definition of a neutral intermediary opportunity. The analogy is not perfect, but credit bureaus are instructive: a shared infrastructure that holds a history that any party in the ecosystem needs but none wants to operate individually, with a per-transaction fee model and a SaaS layer for access. A FHIR-native PA portability service that brokers Payer-to-Payer API calls, holds PA state objects, enforces data provenance, and exposes them to payers and providers on request would solve a compliance headache for the entire industry and would benefit from a strong regulatory flywheel as more states mandate continuity of care. The fact that nobody has obviously won this space yet, 14 months out from the 2027 deadlines, is mildly remarkable.

Compliance Tooling for the Algorithmic Audit Era

The Maryland and California AI laws, the parallel moves in Nebraska, and California’s SB 363 fine structure create a compliance surface that incumbent UM vendors are not structurally equipped to handle. Maryland alone requires quarterly audits of any AI used in UM, written policies and procedures for AI use, metrics reporting on AI involvement in adverse decisions, and the ability to make AI tools available for commissioner inspection. California layers on denial-rate disclosure and up to $1 million per case where appeals are overturned at rates above 50 percent. Any payer that has deployed AI into its UM stack now has a running compliance obligation that needs tooling.

There is a nascent market here for independent AI audit and compliance platforms. The product tests payer algorithms against clinical guidelines, monitors denial patterns for demographic disparity (the “unfair discrimination” standard in Maryland’s law), generates the reporting required by state insurance commissioners, and maintains the documentation needed to defend an algorithm under regulatory review. ONC’s HTI-1 algorithm transparency requirements, which require certified health IT developers to disclose information on predictive algorithms and assess them for fairness, appropriateness, validity, effectiveness, and safety, extend this demand to the vendor side.

The analog is SOC 2 or HITRUST, but for clinical algorithms. Payers will eventually want to be able to point at an external audit report the way they currently point at a SOC 2 Type 2. Someone is going to build that playbook and sell it to every UM vendor in the country.

PA-Aware Clinical Decision Support

One business model category that is hiding in plain sight inside the Da Vinci stack is PA-aware clinical decision support. The CRD API, by design, tells a clinician at order entry whether a service will require PA for a specific member on a specific plan. Flip that around and it is the raw material for a different kind of CDS: a tool that shows the clinician, in real time, which clinically equivalent alternative would be auto-approved, which would trigger PA, and which would likely be denied, before the order is committed.

This is a subtle reversal of the usual CDS flow. Traditional CDS surfaces clinical appropriateness. PA-aware CDS surfaces approvability as a first-class property of the order, alongside appropriateness. For service categories where multiple clinically equivalent options exist (imaging modalities, drug classes, step therapy alternatives, DME configurations), this tooling can eliminate a huge share of PA volume at the point of order entry by quietly steering clinicians toward the approvable option. The payer likes it because utilization conforms to guidelines without a denial fight. The provider likes it because the administrative burden disappears. The patient likes it because there is no delay. The CRD output is the input. Building this well requires deep integration with the EHR, strong payer rule coverage, and a UX that clinicians actually tolerate, which is a high bar, but the economic alignment across all three parties is unusually strong.

Patient-Facing PA Transparency and Appeal Tools

The Patient Access API in CMS-0057-F extends to PA status by 2027. That is a quietly important development for the patient financial experience category. Patients will be able to see, in real time through consumer-grade apps, which of their PAs are approved, which are denied, what the reasons for denial are, and where they stand in the appeal process. Historically that information has lived in payer portals that patients never log into, in letters that arrive after the fact, and in occasional calls to member services.

Once that data is accessible through a consumer-facing API, a new product category becomes possible: patient-facing PA transparency and appeal assistance tooling. Think of it as a mashup of patient financial experience software and legal tech. The product ingests a patient’s PA history through Patient Access, explains denials in plain language, automatically generates appeal letters with the clinical justification populated from the patient’s own clinical data, and tracks the appeal through to resolution. For high-cost services where PA denials are most consequential (specialty drugs, cancer care, rare disease therapies, complex surgical procedures), this is a genuinely novel category. It is also a natural adjunct to existing patient advocacy, price transparency, and financial navigation products. The CMS rule is the enabler; the market does not yet have a clear leader.

Specialty Benefit Manager Modernization

One area that is conspicuously absent from most PA economy analyses is the specialty benefit manager segment. Companies like Evolent, eviCore, Carelon, and Lumeris operate carved-out PA workflows for specific benefit categories (oncology, cardiology, radiology, musculoskeletal, post-acute) under delegated utilization management arrangements with payers. These pipelines are structurally fragmented, often built on bespoke legacy technology inherited from a decade of acquisitions, and not generally FHIR-native.

CMS-0057-F treats delegated UM vendors the same as the underlying payer for compliance purposes, which means the specialty benefit managers have the same January 2027 deadline as their payer clients. That creates two distinct business opportunities. One is a modernization play: building FHIR-native UM infrastructure specifically for the specialty benefit manager segment, which is operationally different from a horizontal payer middleware product because the rules, the specialties, and the clinical guidelines are narrower and deeper. The other is a consolidation play: a well-capitalized specialty benefit manager that modernizes its stack and then rolls up weaker competitors could meaningfully reshape the segment. Either way, this is a category that has been quietly overlooked relative to its PA transaction volume, and it deserves more attention than it gets.

The Dataset Is the Real Prize

Of all the layers in the PA economy, the one most chronically underpriced by entrepreneurs is the data layer. Once PA fully digitizes on FHIR rails, the exhaust is extraordinary. The CMS-0057-F public reporting requirements will produce, for the first time, a standardized, comparable, longitudinal dataset on payer PA behavior across Medicare Advantage, Medicaid, and ACA marketplace plans. CMS-0062-P extends the same framework to drugs. California’s SB 363 denial-rate disclosure adds state-level granularity. Over time this dataset captures payer decisioning patterns by service category and geography, denial rates by provider and specialty, provider performance against guidelines (the basis of gold carding), drug access patterns by formulary tier, appeal overturn rates at the level California fines on, and the correlation between PA denial rates and downstream clinical outcomes.

That dataset is monetizable in a dozen directions. Life sciences companies pay serious money to track market access for new drugs. Providers will pay to benchmark and optimize their gold card status. Payers will pay to benchmark their own UM efficiency against peers. Employers will pay to evaluate health plan performance on behalf of their beneficiaries. Even the plaintiffs’ bar will pay for pattern analysis, particularly in jurisdictions with California-style fine structures. The regulatory framework here is not just creating transparency. It is creating a structured public dataset that becomes the foundation for a new analytics category. The companies that aggregate, clean, and enrich this dataset first will have durable advantages that are hard to unwind once public reporting is established.

Incumbents, New Entrants, and Where the Whitespace Is

The current competitive map sorts roughly into five buckets. Legacy clearinghouses (Change Healthcare now under Optum, Availity, Waystar) own the X12 278 transaction layer and are slow on FHIR, partly because FHIR cannibalizes some of their existing economics and CMS just waived the regulatory requirement that kept them in the middle. EHR vendors (Epic, Oracle Health) are building CRD, DTR, and PAS into their platforms, but they are constrained by enterprise sales cycles and the sheer complexity of supporting thousands of payer-specific configurations inside a single product.

Specialty PA vendors on the provider side (Cohere Health, Rhyme, Infinitus, Myndshft, Infinx, Anterior, Itiliti Health) are executing well on AI orchestration but are not in the infrastructure layer. Payer infrastructure vendors (Smile Digital Health, Firely, HealthEdge, CareEvolution) are in the middleware race but have not consolidated the market. UM platform incumbents and specialty benefit managers (eviCore, Carelon, Evolent, Lumeris) hold payer relationships but tend to run on legacy stacks and have the most to lose from genuine disintermediation.

The whitespace is almost entirely in the infrastructure plane and the adjacent data and compliance layers. FHIR middleware for payer-side compliance. A neutral portability layer brokering Payer-to-Payer API transactions. Algorithmic audit and compliance tooling. PA-aware CDS. Patient-facing PA transparency and appeal tooling. Specialty benefit manager modernization. A standardized data products layer sitting on top of the CMS public reporting data. These are all platform businesses with genuine network effects, deep switching costs once the integrations are built, and a regulatory tailwind that is effectively a customer acquisition subsidy.

The other underappreciated angle is the drug side. CMS-0062-P specifically targets pharmacy PA, which is where patient harm from delay is often most acute (specialty drugs, oncology, rare disease therapies) and where the existing tooling is arguably even further behind the medical side. Surescripts dominates ePrescribing, but the ePA flow for drugs has been historically messy. A clean FHIR and NCPDP SCRIPT-native drug PA orchestration layer, tuned for specialty pharmacy workflows, is a highly defensible wedge if the rule finalizes on something like the proposed timeline.

Honest Risks and What Could Go Sideways

Any serious entrepreneur or investor in this space needs to hold five risks clearly in mind, because each of them has meaningful probability and each kills specific theses.

The first is payer foot-dragging. The history of payer compliance with federal interoperability rules is one of minimum viable effort. The plausible 2027 outcome is that a large share of payers ship thin FHIR veneers bolted on top of legacy UM stacks, technically satisfying the rule while delivering almost none of the real-time, sub-second experience the rule was designed to produce. Products that presume deep, high-quality FHIR implementation on day one will underperform. Products that assume a multi-year maturation curve for payer endpoints will be better positioned.

The second is dual-stack necessity. During the transition, which will last years, any provider-side product that only handles FHIR-native payers is non-viable, because half of PA volume is still fax or phone per BCBSA and 65 percent is still non-electronic per CAQH. The winning products will run FHIR where available and degrade gracefully to portal automation, fax, and phone where FHIR is absent. Architecting for dual-stack is a meaningful engineering investment that pure-FHIR purists will skip, to their commercial detriment.

The third is gold card complexity. Every state that enacts a gold card program writes its own rules: different thresholds, different evaluation windows, different service categories, different reporting obligations. A national tool that cannot handle the full compliance matrix across every gold-card state will get outcompeted by state-specific tools in its weakest jurisdictions. This is a classic compliance-as-moat dynamic, but only if the product genuinely gets it right.

The fourth is consolidation pressure. Clearinghouses and EHR vendors have strong incentives to buy FHIR middleware, audit tooling, and orchestration startups rather than build them. This is a positive outcome for founders but a material risk for investors who are underwriting independent platform trajectories. Strategic acquirers will pay for distribution and regulatory clocks; the timing of those deals will shape returns more than any individual product win.

The fifth is utilization rebound. This one is the most philosophically uncomfortable. If PA becomes frictionless (real-time, auto-approved for most requests, low administrative burden), payers may rationally expand the scope of services subject to PA rather than shrink it, because the cost of administering PA on a new service category drops to near zero. The end state may be more services subject to PA, not fewer, with the volume flowing through the new API rails. That is still a better patient experience than the status quo, but it is worth being honest: PA is being rebuilt, not eliminated. The rails are the market, and the rails are getting more capacity, not less. Anyone whose thesis depends on PA shrinking in aggregate is probably going to be disappointed.

Closing Thoughts

The Prior Authorization API Economy is not speculative. It is under active construction. Federal rules are laying the FHIR rails across four APIs. State laws are writing the SLA parameters and drawing the AI guardrails across at least a dozen jurisdictions. A voluntary commitment covering 257 million Americans is pulling the commercial sector onto the same standards without waiting for Congress. CMS has already waived the X12 mandate for PA, which is the quiet legal unlock that makes pure-FHIR infrastructure viable. And the documented dysfunction of the status quo (65 percent non-electronic, 9 percent of orgs actually ready, half by fax or phone, $15 billion of projected savings, 13 hours per physician per week, adverse events affecting nearly one in three physicians’ patient panels) makes the economic case for rebuild unambiguous.

The simple framing for entrepreneurs and investors is this. PA is becoming a programmable transaction, and programmable transactions always generate platforms. The companies that build the orchestration, the middleware, the portability, the audit tooling, the PA-aware CDS, the patient-facing transparency tools, the specialty benefit manager modernization, and the data products on top of the new FHIR rails will capture the value that currently evaporates in fax machines, peer-to-peer phone tag, and administrative overhead. Regulatory pressure is not the headwind in this market. It is the forcing function creating the market. Programmable medical necessity is a new category of health technology infrastructure, and 2027 is the year the rails get pressure-tested. The practical question is not whether to build here but which layer to build at: rails, trains, or stations. Each has a different capital profile, a different moat, and a different buyer, and each is wide open.​​​​​​​​​​​​​​​​

- Focus: Late 2025 and early 2026 earnings calls across major payers and hospital operators

- Core idea: margins are not disappearing, they are moving across segments

- Key tensions: utilization vs pricing, insurance vs services, volume vs acuity

- Critical signals: Medicare Advantage pullback, PBM revenue quality, hospital revenue cycle strain -

- Who is winning: diversified platforms with services leverage

- Who is exposed: pure-play MA insurers and lower-scale hospital operators

Table of contents

- The setup: utilization did not normalize, it reset

- Insurance margins: the illusion of recovery

- Medicare Advantage: from growth engine to controlled burn

- Services businesses: where the margin actually went

- Hospitals: strong volumes, stubborn economics

- Revenue cycle: the silent margin battlefield

- Where narratives conflict across the stack

- What management teams are signaling vs what they are avoiding

- What this means for operators, builders, and investors

The setup: utilization did not normalize, it reset

The single most important thing buried across these calls, and it really is buried because nobody wants to say it out loud, is that utilization did not revert. It stepped up and stayed there. That sounds obvious in retrospect, except the entire industry spent roughly eighteen months telling itself a different story. The official narrative through most of 2023 and well into 2024 was that the senior utilization spike was a transitory unwind of deferred care, a kind of pig moving through the python, and once it cleared things would settle back into the old trend line. That narrative is gone now. It got quietly retired somewhere between the mid 2024 guidance cuts and the late 2025 calls, and nobody held a funeral.

What replaced it is messier and a lot less comforting for anyone who priced risk based on the old assumptions. The senior population is older, sicker, and more engaged with the system than the actuarial models expected. Chronic disease burden is not receding. Cardiovascular procedures, orthopedic interventions across both inpatient and outpatient settings, and specialty drug utilization are all running at levels that make 2019 through 2022 pricing decisions look like they were made on a different planet. Part of this is demographic, with the leading edge of the boomer cohort now well into the high utilization zone. Part of it is behavioral, with supplemental benefit designs from the Medicare Advantage boom years having trained a generation of seniors to actually use their coverage. And part of it is pharmaceutical, with GLP-1 adjacent prescribing patterns and specialty pipeline launches pulling spend forward in ways that underwriting did not fully anticipate.

The earnings language tracks this shift if you read it with the right ear. The word normalization has mostly disappeared. What you hear now is calibration, recalibration, repricing, discipline, and the ever popular operating rigor. That is not accidental. Calibration implies the system is adjusting to a new steady state. Normalization implied the old steady state was coming back. Those are very different concepts and the word swap is doing real work.

The practical implication is that every business model in the stack that was sized for 2019 utilization economics has been quietly rebuilt, or is in the process of being rebuilt. That rebuild is the actual story of the last four quarters, and most of what follows is really just variations on it.

Insurance margins: the illusion of recovery

At a headline level the payer numbers have stopped getting worse. MLRs are no longer expanding at the same pace. Some carriers are guiding to modest year over year improvement. If you squint, it looks like stabilization, maybe even the early innings of recovery. That is the surface.

Underneath it is not really recovery. It is a repricing exercise catching up to a cost base that shifted two years ago, combined with benefit design trims and network tightening that took out some of the worst unit economics. The underlying cost structure did not get better. The pricing got closer to the cost.

The cleanest illustration of where this leaves things sits inside CVS, where the insurance segment has produced quarterly losses even while the enterprise as a whole still generates billions in operating income. That is the whole story in one line. The underwriting engine is not the profit engine right now. It is being carried by everything around it. The insurance book is still functional, still servicing millions of lives, still generating massive revenue, but the margin contribution has effectively been redirected. Management can frame that as strategic integration, and they do, but the accounting reality is that one segment is subsidizing another.

Cigna shows a variation on the same pattern with different geometry. The stub healthcare segment, once adjustments for the Medicare divestiture are baked in, is stable to modestly growing. The scale and the growth both sit in Evernorth. But inside Evernorth the split between top line and bottom line is where the interesting tension lives. Revenue is expanding very quickly, driven by pharmacy volume, specialty pipeline, and care services. Operating income is growing too, but not at the same rate in the PBM core. That gap is meaningful. It tells you that some of the growth is lower quality than the revenue line suggests. Spread compression in traditional PBM economics is real and ongoing, even as specialty and care services do the heavy lifting on incremental margin.

UnitedHealth looks cleaner on the surface mostly because Optum is enormous and the segment reporting structure allows the services strength to mask the insurance pressure. The insurance pressure is still there. It is just being absorbed inside a larger system. The 2024 medical cost pressure stories, the 2025 guidance revisions, and the ongoing commentary around the individual senior lives with higher than expected cost trend all point to the same underlying dynamic the rest of the sector is dealing with. Optum is not making that dynamic go away. It is providing enough offsetting profit contribution that the consolidated numbers stay investor friendly.

Elevance sits in a somewhat similar place, with Carelon doing more of the work each quarter. Humana, because the services arm is smaller relative to the insurance book, shows the pressure more rawly. The variance across carriers in how painful the current environment looks is largely a function of how much services leverage is sitting next to the insurance book. That is not a coincidence. That is the structural story.

The industry level takeaway is simple to state and slightly awkward to live with. Insurance margins did not magically recover on the back of some endogenous improvement in the business. They stabilized because pricing caught up, benefits were trimmed, networks were tightened, acquisition costs were pulled back, and supplemental design was rationalized. The cost structure underneath is not fundamentally different than it was when MLRs were blowing out. It just has a bigger revenue number on top of it.

Medicare Advantage: from growth engine to controlled burn

For most of the last decade Medicare Advantage was the easiest growth story anywhere in healthcare. Enrollment compounded at rates that would make a SaaS founder jealous, risk adjustment methodology supported margin, supplemental benefit escalation drove market share wars, and public markets rewarded whoever could show the fastest lives growth. That era is visibly over. Not winding down. Over.

The earnings language now leans into discipline, which is the polite industry word for walking away from members. Plans are exiting geographies, trimming or eliminating supplemental benefits that drove prior enrollment gains, pulling back on broker channel aggression, and being much more selective about which segments of the senior population they want to grow in. The 2026 plan year filings, read in aggregate, show benefit reductions that would have been unthinkable three years ago. Dental, vision, OTC allowances, transportation, grocery cards, the whole supplemental stack that defined the competitive dynamic in 2019 through 2022 is being scaled back or eliminated across large swaths of the market.

Humana is the most transparent about this because it has to be. As the most MA concentrated of the major publicly traded insurers, the pressure shows up in its numbers without anywhere to hide. The messaging centers on recalibration and return to margin, but the actions are sharper than the words. Benefit designs are being cut. Growth is being traded explicitly for per member profitability. The company is basically telling the market it would rather shrink into a more profitable book than grow into a less profitable one. That is the right strategy. It is also a complete inversion of the playbook that drove valuation for a decade.

The variance across players in how they can absorb this shift is where the two speed market really shows up. A company with a large services arm can tolerate compressed insurance margin because it is monetizing the same lives through pharmacy, specialty distribution, care delivery, behavioral, and analytics. The insurance book becomes a customer acquisition channel for a broader platform. A pure play MA insurer does not have that option. Its insurance margin is its margin. When that compresses, everything compresses. That is why the smaller and more concentrated MA names have had such a brutal stretch relative to the diversified platforms, even when the underlying medical cost trend they are fighting is roughly similar.

Risk adjustment methodology is the other overhang. The V28 transition phase in, combined with increased audit activity and the ongoing debate about coding intensity, is putting downward pressure on risk scores at exactly the moment cost trend is elevated. Plans that built their margin on aggressive coding are getting squeezed from both sides, with costs going up and risk adjusted revenue going down. That is not a temporary mismatch. That is a structural reset of the economics of the product.

The part none of the calls really address head on is what this means for the supplemental benefit arms race as a competitive dynamic. If everyone is trimming benefits at roughly the same time, the relative competitive position does not change that much. The senior who was getting a three hundred dollar OTC card is now getting a hundred and fifty, but so is everyone else in the market. The question is whether enrollment behavior responds to the absolute level of benefits or the relative level. Early signals suggest absolute matters more than the industry wants to admit, which would imply disenrollment pressure as seniors shop around looking for the benefit levels they got used to. That shows up in the 2026 AEP data, and it is a tension that is going to define the next twelve to eighteen months.

Services businesses: where the margin actually went

If there is one through line across every major payer call, it is that services are doing the work. Optum, Evernorth, and the CVS Health Services segment are not diversification anymore. They are the core earnings engines. The margin that used to sit in insurance has been relocated one layer up in the stack, into the services businesses that sit adjacent to or inside the same enterprise.

Evernorth is worth walking through because its segment reporting is relatively clean and the dynamics are visible. Top line growth is being driven by pharmacy volume, with specialty continuing to be the fastest growing component, and by care services including the Express Scripts and Evernorth Care businesses. Operating income growth is positive but materially slower than revenue growth in the PBM core. That gap is spread compression. Pass through pricing models, transparent rebate arrangements, and ongoing employer pressure on PBM economics are all grinding at the traditional margin structure. Specialty pharmacy and care services are carrying a disproportionate share of incremental operating income, which is where the higher quality growth actually is.

CVS Health Services shows a similar shape with more noise. The segment is highly profitable on an adjusted basis. It is also where the significant goodwill impairment charges tied to care delivery assets have landed, most visibly on the Oak Street side. That is the part that makes the vertical integration narrative complicated. The strategic logic, which is that owning primary care for seniors gives the insurance arm more control over cost and quality, is sound in theory. The economics of acquiring those assets at 2021 valuations and integrating them into a much larger system have been mixed. Some of the acquired capital is performing. Some of it is being written down. When you read the adjusted numbers you see the performing part. When you read the GAAP numbers and the impairment disclosures, you see both.

The distinction matters because it changes how to think about the returns on vertical integration strategy at the sector level. If the services margin that is replacing insurance margin is partially financed by writing down the assets that were supposed to generate it, the net economics are softer than the adjusted numbers suggest. That does not make the strategy wrong. It makes the strategy more expensive than the adjusted framing implies, and it raises the bar for whether the promised long term synergy benefits materialize.

Optum is the benchmark here because it has operated at scale the longest. Optum Health, Optum Rx, and Optum Insight each show different margin profiles and different growth trajectories. Optum Rx looks a lot like Evernorth on the PBM side, with pharmacy volume growth outpacing operating income growth. Optum Health is a mixed bag, with value based care arrangements performing unevenly across geographies and populations. Optum Insight is the quietly interesting one, because the technology and revenue cycle component has relatively clean high margin growth that does not carry the same cyclical exposure as the insurance or care delivery components. That is where the infrastructure tax on the rest of the system shows up, and it is showing up reliably.

The industry level point is that the services wrapper around the insurance book is not a hedge. It is where the margin is. But not all services are equal. PBM core economics are compressing. Specialty pharmacy economics are holding up but under ongoing pressure. Care delivery economics are mixed and carry real integration risk. Data and technology services are the highest quality component, with the caveat that they are smaller in absolute dollars. Reading the services story as a monolith misses the texture. The texture is what matters.

Hospitals: strong volumes, stubborn economics

On the provider side the first read of the data looks positive. Volumes are up. Acuity is up. Revenue per adjusted admission is up. Same facility revenue growth is running at rates that would historically drive meaningful margin expansion. The margin expansion has not really shown up, or has shown up much less than the top line would suggest.

The reason is a cost base that reset in 2021 and 2022 and has not reset back. Contract labor is well off its peak, which gets called out on every call as a tailwind, but base wages stepped up permanently. Nursing compensation, tech compensation, and the compensation structure for the scarce clinical roles are all at levels that are not reverting. Supply costs, particularly for implants and specialty pharmaceuticals used in the inpatient and outpatient settings, are elevated and sticky. Professional fees for hospital based physician services, especially anesthesia and emergency medicine, have stepped up as the staffing companies in those spaces work through their own economic issues.

HCA, because it is the largest and most operationally sophisticated of the for profit hospital operators, shows the cleanest version of this dynamic. Same facility revenue growth is strong, acuity is favorable, and management commentary is generally constructive. Margins are improving but the rate of improvement is slower than the top line growth would historically predict. That gap is the cost base.

Tenet shows a slightly different picture because of USPI. The ambulatory surgery center portfolio is genuinely high margin and growing, and it pulls the consolidated margin profile up in a way that masks some of the hospital side pressure. Stripping USPI out, the pure hospital margin story looks a lot more like Community Health Systems and HCA.

Community Health Systems is worth calling out because the numbers are unvarnished in a way that is useful. Same store revenue growth has been positive, but admissions have been flat to slightly down in several recent quarters. That tells you that pricing, case mix, and supplemental payment programs are doing most of the work. It is not bad. It is lower quality than volume driven growth, and it depends on the durability of the supplemental payment structures, which is not guaranteed.

The supplemental payment dynamic is its own topic and does not get enough attention in the calls. State directed payment programs, Medicaid supplemental arrangements, and various 1115 waiver structures are material contributors to reported margin for a lot of hospital systems, particularly those with higher Medicaid and uninsured exposure. Those programs get renewed or restructured on state level timelines, and the federal posture toward them has been shifting. A material change in the supplemental payment environment would move provider margin noticeably. The calls mostly treat this as background noise, which it is not.

Payer mix is the other thing quietly eating into the volume story. The continued shift from commercial to Medicare, driven by demographics and by employer plan changes, is compressing average reimbursement per case even as acuity and volume both trend up. Medicare Advantage now represents a majority of Medicare eligible lives in many markets, which means hospital reimbursement is increasingly set by MA contract rates rather than traditional Medicare. Those rates vary by plan and by contract vintage, and the tension between hospitals and MA plans over reimbursement, prior authorization, and claims adjudication is visible in the contracting commentary across calls. Several large health systems have publicly exited MA contracts with specific plans, which is a new dynamic and worth watching.

The hospital margin story right now is an operational grind. Incremental improvement is possible and is happening. The idea of snapping back to 2019 margins at the sector level is not supported by anything in these calls.

Revenue cycle: the silent margin battlefield

Almost every provider call touches revenue cycle, but usually as a side note rather than a central theme. That framing is wrong. Revenue cycle is where a large share of the unresolved margin battle is actually being fought.

Denial rates are up, and the composition of denials is shifting toward more clinically complex reasons that take longer to overturn. Prior authorization requirements are expanding across both commercial and Medicare Advantage books, and the administrative friction is rising in ways that are not fully captured in any single line item. Collections are getting harder as high deductible plan penetration continues and patient out of pocket obligation grows. Average days in accounts receivable have drifted up across most of the large publicly traded systems, even as they have invested heavily in revenue cycle technology and offshoring.

The actuarial reality is that the cost of getting paid has risen faster than the cost of providing care in many systems. That is a striking statement and the data in these calls roughly supports it. Hospital systems are seeing more complex patients, generating more revenue per case, and spending more to capture that revenue. The net margin benefit is muted because the administrative cost layer is growing.

This is also where the disconnect between payer and provider narratives is most visible. Payers talk about medical cost trend management, which on the earnings call sounds like pricing discipline and utilization management. Providers experience the same thing as more denials, more documentation requirements, more prior authorization loops, and more back end friction in getting paid for care that has already been delivered. Both narratives are describing the same underlying activity. They are framed in opposite moral terms, and both framings contain real information.

For anyone building in this space, and several of the calls mention deals or investments in revenue cycle technology and services without ever quite using the phrase, this is one of the most structurally durable problem areas in healthcare. It is not a transient issue waiting for a policy fix. It is getting more complex each year as payer rules proliferate, as more care moves to risk based and value based contracting structures that require new kinds of documentation, and as the underlying clinical acuity of the patient population increases. The vendor landscape around it is fragmented, which is both the opportunity and the challenge. Whoever consolidates durable workflow across the denial management, prior authorization, and patient financial experience stack captures an enormous amount of value, and the large incumbents know it, which is why the strategic activity in this category has been quietly intense.

Noteworthy that Optum Insight, Waystar, R1, and the private equity owned middle layer of revenue cycle services have all been in various stages of repositioning over the last twelve months. Some of that is public, some of it is not. The point is that the capital is flowing where the friction is, and the friction is here.

Where narratives conflict across the stack

Reading the calls side by side, rather than in isolation, is where the most useful insight comes out. The interesting moments are not where the stories agree. They are where the stories should agree and do not.

Providers describe payer mix as favorable and pricing as constructive. Payers describe medical cost trend as elevated and rate increases to providers as moderate. Both of those statements cannot be fully true at the same time. If providers are getting paid meaningfully more per case and payers are not spending meaningfully more per case, the math does not close. The resolution is that both are partially true. Case mix acuity is shifting, so payers are spending more per case partly because the cases are more complex, not only because unit rates are higher. And providers are capturing a mix of rate, acuity, and supplemental payment that reads as favorable on their side without being purely a unit price story. But there is still real tension embedded in the gap, and that tension is where the 2026 and 2027 contract negotiation cycles are going to be painful.

The services story contains a similar tension. Payer calls emphasize growth in pharmacy and care services. The profitability of that growth varies widely by component. High revenue growth in PBM core with lower operating income growth means that growth is not as high quality as the top line suggests. Specialty pharmacy is better, but it is also where employer and regulatory scrutiny is most intense. Care delivery is the highest potential margin and the most capital intensive, and is where the integration risk has been most visible. If you read the services growth narrative as uniformly positive across the components, you miss the real picture. The real picture is a mix of durable margin, compressing margin, and potential margin that is still being built out with uneven execution.

The timing mismatch is the other recurring conflict. Providers are benefiting from higher acuity now, in current period results. Payers are repricing for it with a lag, through bid cycles, contract renewals, and benefit redesigns that take one to two years to fully manifest. During the lag, provider results can look strong while payer results look weak, even when the underlying economic reality is that both sides are converging toward a new equilibrium. The sector level earnings split between the two sides over the last six quarters reflects this lag effect as much as it reflects any real divergence in underlying performance.

The other conflict worth naming is between the vertical integration narrative and the segment level financial reality. The public narrative from the large diversified platforms is that vertical integration is working and delivering the long promised synergies. The segment data shows a more complicated picture, with some components performing well, some being actively restructured, and some being written down. That does not mean the strategy is failing. It means the strategy is expensive and uneven, and investors who take the adjusted narrative at face value without digging into the segment reporting are missing the texture.

Conflicts like these are where the alpha is in reading these calls. Clean consistent narratives across the stack are rare in healthcare. Misalignment is the signal, and the misalignments pointing to the biggest structural shifts are the ones that show up repeatedly across multiple companies, not just in one.

What management teams are signaling vs what they are avoiding

Every management team shapes the narrative, and in healthcare that shaping is especially pronounced because the businesses are complex enough that the framing choices matter a lot.

When a management team talks about adjusted operating income without giving equivalent weight to impairment charges, that is signaling. When revenue growth gets prominent treatment but the associated margin compression only shows up in footnotes and analyst day decks, that is signaling. When the word transitory gets used more than once per call, that is almost always signaling.

Cigna is relatively good at structured disclosure. The segment breakouts are usable and the commentary generally lets you follow the story without heroic reconstruction. That is a compliment, not a setup for a criticism. When management teams structure their reporting to allow real analysis, it is easier for long term investors to get comfortable, and it is harder for short term narrative shifts to distort valuation.

CVS provides a lot of data but relies heavily on adjusted frameworks that remove some of the messier components of recent history. The company is not unusual in this, and management is navigating a genuinely difficult strategic position. But interpreting CVS requires more work than interpreting Cigna, and the adjusted to GAAP bridge is where a meaningful share of the real story sits.

UnitedHealth is a specific case, partly because of the scale and partly because of the current environment. The disclosures have become more structured under pressure, and the commentary around medical cost trend has become more specific over the last several quarters. That is investor demand forcing clarity. It is also an implicit acknowledgment that the prior framing was not sufficient given what was happening underneath.

Hospital operators tend to emphasize same facility metrics, which is the right frame and is genuinely helpful. They often spend less time on the sustainability of supplemental payment programs, the specifics of payer contracting disputes, and the long term implications of payer mix shifts, all of which are material to the forward margin profile. The language around labor is usually specific enough to be useful. The language around revenue cycle and administrative burden is usually general enough to be vague.

Across the entire set, the topic that gets the least structured discussion relative to its actual economic importance is administrative complexity and revenue cycle friction. Every participant is experiencing it. Very few are quantifying it in any useful way on the call. That gap between what is being lived operationally and what is being surfaced narratively is a useful place to focus attention, both for investment analysis and for anyone trying to build a business that addresses the problem.

What this means for operators, builders, and investors

Step back from the specifics and the picture is coherent. The industry is not losing margin. It is redistributing margin. The absolute pool of profit across the healthcare stack has not materially shrunk. It has moved, and it is continuing to move, and the direction of travel is reasonably clear if you squint past the quarterly noise.

Insurance is becoming less reliable as a standalone profit engine. Not unprofitable. Less reliable. The cyclicality of medical cost trend, the structural shift in the senior population, the compression of Medicare Advantage economics, and the ongoing pressure on risk adjustment methodology all point to an insurance business that is more volatile and more capital intensive than the market assumed during the growth years. That does not make it a bad business. It makes it a different business than the one that was being valued in 2019 through 2022.

Services are becoming more central, but the services category is heterogeneous. Pharmacy benefit management in its traditional form is under long term margin pressure from transparency demands, employer sophistication, and regulatory scrutiny. Specialty pharmacy is a better position, with higher growth and more durable economics, though not immune to the same pressures. Care delivery is the highest potential return and the highest integration risk, and the experience of the last three years suggests that execution quality matters more than strategic positioning. Data, analytics, and revenue cycle technology services are the cleanest margin profile in the category, with the caveat that they are smaller in absolute dollars and require real operational scale to generate durable advantage.

Providers are recovering on volume and acuity, but constrained by a cost base that has reset and by an administrative environment that is getting more complex, not less. The good operators will keep grinding incremental margin improvement. The idea of a snap back to pre pandemic sector margins is not supported by anything visible in the current cycle. The weaker operators, particularly the smaller systems without the scale to invest in revenue cycle, labor productivity, and service line rationalization, are going to continue consolidating into larger platforms. That consolidation is itself a margin redistribution event, with value accruing to the acquirers who can integrate efficiently.

For operators inside the system, the implication is that diversification across the value chain is not optional if meaningful scale is the goal. A pure play insurance business, or a pure play hospital system, or a pure play PBM, is a harder business to run in the current environment than it was ten years ago. That does not mean every company should try to be vertically integrated. It means companies that are not vertically integrated need to have a clear answer to why their single segment positioning is durable, and that answer has to account for where the sector level margin is actually accumulating.

For builders, and this is where it gets interesting for anyone working in the early stage and growth stage ends of the market, the friction points in the stack are where durable businesses get made. Revenue cycle and administrative workflow sit at the top of that list. Specialty pharmacy infrastructure, particularly around patient support, benefits investigation, and prior authorization, is another. Care delivery enablement for the new value based contracting structures is a third, though the execution bar is high. Data infrastructure that actually connects clinical, financial, and utilization data in usable form is a fourth, and the existing incumbents have not solved it cleanly despite significant capital deployed.

For investors, the implication that gets most consistently missed is that segment level analysis matters more than consolidated analysis, and consolidated analysis matters more than narrative. The same company can have a weak insurance segment, a compressing PBM segment, a struggling care delivery investment, and a genuinely high quality data and technology business, all at the same time, and the consolidated adjusted earnings number will obscure most of that texture. Valuation that rests on consolidated adjusted numbers without a view on segment level durability is valuation that is going to be surprised by segment level transitions. There have been a lot of segment level transitions in healthcare recently. There are more coming.

The easy narratives have mostly broken. Growth is still there. Profit is still there. Both are just not sitting where they used to sit, and the disconnect between where they used to sit and where they sit now is the single most important thing to internalize about reading these calls. Everything else flows from that.​​​​​​​​​​​​​​​​

What just happened on April 10

The size of the prize, in actual dollars and procedures

Why the old CJR playbook is only partially useful here

Builder opportunity one: the post-acute steerage and PT layer

Builder opportunity two: PROMs as a revenue cycle problem

Builder opportunity three: the outpatient migration toolkit

Builder opportunity four: risk-adjusted target price intelligence

Builder opportunity five: the convener and gainsharing back office

Builder opportunity six: device, implant, and supply rationalization

Builder opportunity seven: rural and safety net co-pilot

Builder opportunity eight: the patient navigation and demand-side play

Channel and GTM notes for anyone selling into this

Closing thoughts on timing, exits, and what could break the thesis

Abstract

Quick framing for anyone skimming:

- CMS dropped the FY27 IPPS proposed rule on April 10, 2026, and tucked inside is CJR-X, the first ever mandatory, nationwide, episode-based payment model in US history.

- Start date: October 1, 2027. Five performance years through Sept 30, 2032. Roughly 3,000+ IPPS hospitals in scope, minus the ~700 already in TEAM and minus Maryland.

- Episode = 90 days post-discharge from inpatient or outpatient hospital LEJR. Triggers: MS-DRG 469, 470, 521, 522, plus HCPCS 27447 (TKA) and 27130 (THA), plus total ankle.

- Risk adjustment jumps from 3 levers in CJR to 29 in CJR-X (same engine as TEAM). 5% stop-loss for safety net, dual-eligible heavy, geographically rural, MDH, and SCH hospitals.

- Quality first: 5 measures including the THA/TKA PRO-PM (CMIT 1618), which CMS plans to weight heavier than TEAM does. PROMs are now a P&L input.

- TAM context: ~1M+ Medicare TKAs and ~600K Medicare THAs annually as of 2022, growing at 5.9% and 7.6% CAGR respectively, with surgeon reimbursement down ~55% inflation-adjusted since 2000. Hospitals are squeezed on the front end and now risk-bearing on the back end.

- Net savings from the original CJR’s PY6 and PY7 (2021 to 2023): $112.7M, with quality flat. CMS has a green light to scale.

- Eight buildable categories laid out in the body, plus channel notes for selling into hospitals, ortho groups, ASCs, SNFs, and home health.

What just happened on April 10

The April 10 release of the FY 2027 IPPS proposed rule, all 1,500-ish pages of it, contained the regulatory equivalent of a starter pistol for healthcare services builders. Buried inside that doorstop is CJR-X, which CMS would like to begin on October 1, 2027 as the first nationwide test of a mandatory episode-based payment model. Public comments are open through June 9, 2026. CMS Administrator Mehmet Oz framed it in the press release as aligning incentives with outcomes and protecting taxpayer dollars, which is the standard phrasing for “we are putting hospitals on the hook for 90 days of post-op spend whether they like it or not.”

The mechanics are familiar to anyone who lived through the original CJR (2016 through 2024) or who has been reading the TEAM tea leaves. Acute care hospitals get a target price covering all Part A and Part B services in a 90 day episode beginning with the procedure. They keep getting paid FFS during the year. After year-end, CMS reconciles actual spend against the target price, applies a quality adjustment based on the composite quality score, and either cuts a check or sends an invoice. Stop-loss is set at 20% for most participants, 5% for the protected categories. The quality first principle still applies, meaning if a hospital fails to clear the minimum composite quality score, no reconciliation payment, full stop.

What is meaningfully different from CJR is the scope. The original ran in 67 metros at peak, then settled into 34 MSAs. CJR-X is every IPPS hospital in the country except Maryland (carved out by the Total Cost of Care waiver) and the roughly 700 hospitals already locked into TEAM, who flip into CJR-X when TEAM expires on December 31, 2030. Total ankle replacement is now in scope. Outpatient hospital procedures are now in scope. The episode definition follows the patient through the HOPD. Critical access hospitals and rural emergency hospitals stay exempt because they sit outside the IPPS/OPPS plumbing.

The other meaningful change is the risk adjustment engine. The old CJR ran with three risk adjusters and got criticized, fairly, for clobbering hospitals that drew from sicker, more complex populations. CJR-X imports the TEAM methodology: 29 adjusters at the episode level (age, HCC count, dual-eligibility, procedure type, disability as the original reason for Medicare enrollment, prior PAC use, plus 21 specific HCCs) and two at the participant level (bed count and dual-eligible share). For builders, this matters more than it sounds. A more sophisticated risk model means hospitals can no longer just “select better patients” to game the model. The acuity is priced in. The only way to win is to actually run a better episode.

The size of the prize, in actual dollars and procedures

Read more

What it is: The ACCESS (Advancing Chronic Care with Effective, Scalable Solutions) Model is a 10-year, voluntary, nationwide CMS Innovation Center model launching July 5, 2026, testing an Outcome-Aligned Payment (OAP) methodology for technology-enabled chronic care across four clinical tracks: eCKM (early cardiometabolic), CKM (full cardiometabolic/diabetes/CKD), MSK (musculoskeletal pain), and BH (behavioral health: depression and anxiety).

Key payment mechanics: Monthly fixed per-patient payments with a 50% withhold reconciled at end of 12-month care periods. Full payment contingent on achieving OAP Measure targets across a patient panel. The Outcome Attainment Threshold (OAT) starts at 50% in the first 18 months. A Substitute Spend Adjustment (SST starting at 90%) penalizes participants whose patients receive duplicative FFS services for the same condition from other providers. Only the larger of the two adjustments is applied per reconciliation period, capped at 50% and 25% respectively.

The payer context: 14 major payers representing 165 million covered lives have signed the ACCESS Payer Pledge, committing to align commercial, MA, and Medicaid payment to the model’s structure by January 1, 2028. This includes UnitedHealthcare, Humana, Cigna, CVS/Aetna, Centene, Devoted Health, and several BCBS plans.

The participant field: 150-plus organizations accepted as of April 2026, spanning virtual-first digital health companies, traditional brick-and-mortar specialists, pharmacies, care management vendors, and a handful of entries that defy easy categorization.

Key questions addressed: Who has the business model alignment, patient panel sophistication, and outcome track record to actually make money here? Who will make the biggest clinical dent in Medicare’s chronic disease burden? Who is absent from the list who probably should be there? And who are we looking at going, really?

How to Think About This List Before Getting Into Specifics

There’s a temptation to treat the ACCESS accepted applicant list as a flat roster, an alphabetical spreadsheet of companies that all said yes to the same thing. That would be a mistake. The gap between the top tier and the bottom tier of this group, in terms of actual readiness to thrive in an outcome-aligned payment model, is enormous. The business model differences are even starker than the clinical ones. Some of these organizations have been building toward exactly this moment for years, running commercial book-of-business programs with outcome accountability, investing in remote monitoring infrastructure, and proving out that they can move biomarkers at scale. Others are walking in from adjacent markets with a lot of enthusiasm and not a lot of evidence. A few are genuinely puzzling additions.

To understand who’s positioned well, you need to internalize one thing about how the OAP payment structure actually works in practice. The model pays a fixed monthly per-patient rate (amounts not yet published as of the April 2026 application period), with 100% of payments flowing in months one through six and the back half effectively held until the 12-month mark. Full release of that withhold depends on the participant’s Outcome Attainment Rate, meaning the share of their total patient panel who hit the required clinical targets. The OAT starts at 50% across the first 18 months, which sounds lenient until you realize most of these organizations have never tried to move Medicare FFS patients through an outcome-linked accountability framework before. Medicare populations are older, more complex, and more likely to have comorbidities that complicate biomarker improvement. The organizations that have spent years managing commercially insured populations in their 40s and 50s are going to face a real wake-up call when the panel shifts to original Medicare beneficiaries, many of whom are in their late 60s, 70s, and 80s with entrenched disease patterns and polypharmacy issues.

The Substitute Spend Adjustment is the other thing that makes this genuinely hard. If a participant’s patients start getting physical therapy evaluations, psychiatric evals, DSMT sessions, or cardiac monitoring billed through FFS channels by other providers during the care period, the participant’s Substitute Spend Rate drops and their payment gets dinged. This creates a coordination burden that tech-first, virtual-only companies are particularly bad at managing, because they often don’t have deep enough relationships with a patient’s brick-and-mortar care team to prevent those FFS encounters from happening. That 90% SST in year one is going to be a real problem for anyone who doesn’t have tight care coordination workflows.

With that backdrop, let’s go through the landscape.

The Category That Will Probably Make the Most Money: Scaled Virtual-First CKM and eCKM Specialists

Read more

The CMS CY 2027 Medicare Advantage and Part D Rate Announcement dropped April 6, 2026. It finalizes a net average payment increase of 2.48 percent, or roughly 13 billion dollars in additional MA payments. But the real story for health tech builders and investors is not the topline number. It is the collection of new operational mandates, data architecture changes, and program integrity crackdowns that will force payers to either build or buy entirely new capabilities over the next 12 to 24 months. This essay reads the Rate Announcement through the lens of a venture incubator asking one question: what companies need to exist because of what CMS just did?

Key themes:

Unlinked chart review record exclusion creates a massive encounter data remediation market

Audio-only diagnosis exclusion forces telehealth workflow retooling

Separate PDP and MA-PD normalization opens a wedge for Part D analytics startups

Star Ratings measure set overhaul, including new depression screening, creates HEDIS/quality reporting buildout demand

Skin substitute repricing aftermath continues to scramble risk adjustment economics

The BALANCE Model and GLP-1 utilization create pharmacy benefit uncertainty that favors risk analytics tooling

ESRD payment adequacy concerns open a door for kidney care cost management platforms

Puerto Rico-specific adjustments signal a niche but real market for territory-focused plan services

Table of Contents

The Topline Numbers and Why They Lie

Unlinked Chart Reviews: The Biggest Startup Catalyst in the Document

Audio-Only Telehealth Gets Kneecapped for Risk Adjustment

Separate Worlds: PDP vs MA-PD Normalization and What It Means for Part D Startups

Star Ratings Overhaul: Depression Screening, Measure Removals, and Quality Tech

Skin Substitutes, Risk Model Freeze, and the Coding Economy

GLP-1s, BALANCE, and the Part D Pharmacy Benefit Chaos

ESRD and the Kidney Care Opportunity

Puerto Rico as a Micro-Market

What to Build

The Topline Numbers and Why They Lie

CMS finalized a 2.48 percent net average increase in MA plan payments for CY 2027. That is about 13 billion dollars more than 2026. When the Advance Notice came out in January projecting a 0.09 percent increase, the industry basically had a collective panic attack, so 2.48 percent feels like a relief. But relief is not the same as generosity. When you add in the estimated 2.50 percent risk score trend from underlying coding changes and population shifts, the effective growth is closer to 4.98 percent. Sounds decent until you factor in what the plans themselves say they are seeing in utilization trend. Nearly every large MA organization commented that inpatient utilization, Part B drug spending, post-acute volumes, and behavioral health costs are all running hotter than what CMS projects. The commenters are not subtle about this. The public comment section of the Rate Announcement reads like a 50-page group therapy session for actuaries who feel gaslit by the Office of the Actuary.

What matters for builders is that 2.48 percent is a number designed to keep the lights on for plans, not to let them print money. The margin environment stays compressed. Plans that cannot operationally adapt to the new risk adjustment rules, the new quality measures, and the new encounter data requirements will be the ones who exit markets or cut benefits. And the ones who can adapt will need vendors to help them do it, because the operational changes CMS is requiring are not trivial.

Unlinked Chart Reviews: The Biggest Startup Catalyst in the Document

This is the single most consequential policy change in the Rate Announcement for health tech entrepreneurs, and it barely got any press. CMS is finalizing the exclusion of diagnoses from unlinked chart review records from risk score calculation starting in CY 2027. The only exception is for beneficiaries who switch from one MA organization to another.

To understand why this matters, you need to understand how chart reviews work in the MA risk adjustment ecosystem. MA plans submit encounter data records, which are basically electronic claims documenting the services provided to enrollees. Separately, plans can submit chart review records, which add diagnoses found during retrospective medical record reviews. When a chart review record is “linked,” it is tied to a specific encounter data record and says “we found an additional diagnosis in the chart from this visit.” When it is “unlinked,” it floats free with no connection to any documented service encounter.

CMS found that roughly 85 percent of the 88.8 million unlinked chart review records submitted in 2023 for 2024 payment could not be matched to any encounter data record even when they tried matching on beneficiary, billing provider, and dates of service within three days. That is a staggering number. CMS is essentially saying: if you cannot tie a diagnosis to a real clinical encounter, it does not count for payment.

The average payment impact of this exclusion is negative 1.24 percent. Without the switcher exception, it would be negative 1.78 percent. But the average masks enormous variation. Plans that have relied heavily on in-home health risk assessments, third-party chart review vendors, and retrospective coding operations to generate unlinked chart review records are going to get hit much harder. Some of the largest national MA plans have built entire revenue lines around this practice.

So what needs to be built? First, encounter data remediation platforms. Every MA plan in America now needs to audit its encounter data submission infrastructure and figure out how to make sure every diagnosis that matters is either submitted on an encounter data record or on a chart review record that links to one. That is a workflow and data engineering problem. The plans that have been lazy about encounter data submissions (and many have, because unlinked chart reviews were an easy workaround) need to fix their pipelines fast. Second, prospective risk adjustment tools that capture diagnoses at the point of care rather than retrospectively through chart review. CMS has been pushing this direction for years, and the unlinked chart review exclusion is the final shove. Tools that integrate with EHR workflows to ensure accurate, complete diagnosis capture during clinical visits are now table stakes, not nice-to-haves. Third, chart review linking services. For plans that still do retrospective chart review (and many will, it is not going away), the ability to reliably link chart review findings to existing encounter data records becomes a critical compliance function. This is a data matching and reconciliation problem that screams for a purpose-built SaaS product.

The market here is not small. Risk adjustment revenue for the MA program is hundreds of billions of dollars. A 1 to 2 percent payment impact across 34 million MA enrollees translates to billions of dollars at risk. Even capturing a small slice of the remediation and tooling market around this policy change is a venture-scale opportunity.

Audio-Only Telehealth Gets Kneecapped for Risk Adjustment

CMS is finalizing the exclusion of diagnoses from audio-only telehealth encounters from risk score calculation. If the only service lines on a claim have audio-only modifiers (modifier 93 or FQ), the diagnoses from that encounter are out for risk adjustment purposes. The average payment impact is zero, which tells you something about how few plans were actually submitting these diagnoses. But the policy signal is loud: CMS wants diagnoses tied to face-to-face encounters, period.

For telehealth companies, this is a clarification that has been a long time coming. Audio-only visits were never supposed to generate risk-adjustment-eligible diagnoses under existing CMS guidance, but the operational enforcement was spotty. Now it is explicit. Companies building telehealth platforms for MA plans need to make sure their visit type routing clearly distinguishes between audio-video visits (which do support risk adjustment) and audio-only visits (which do not). This is a feature, not a product, but it matters for any company selling into the MA risk adjustment workflow. Plans will want reporting dashboards that show what percentage of their telehealth volume is audio-only versus audio-video, broken down by provider and condition, so they can coach their networks accordingly.

Separate Worlds: PDP vs MA-PD Normalization and What It Means for Part D Startups

CMS finalized separate normalization factors for standalone prescription drug plans and MA prescription drug plans, plus separate continuing enrollee model segments in the RxHCC risk adjustment model. This is a technical change that has big strategic implications.

The background: MA-PD plans have historically had higher risk scores than standalone PDPs because MA organizations can affect the submission of diagnoses through their Part C encounter data. PDPs cannot. The old single normalization factor basically split the difference, which overpaid MA-PD plans and underpaid PDPs. CMS has been moving toward separate normalization for a couple of years, and the CY 2027 finalization locks it in with a methodology that assumes equal underlying risk between the two sectors.

Why does this matter for startups? The PDP market has been under enormous stress since the IRA Part D redesign. Plan liability went up, the catastrophic phase cost sharing structure changed, and the premium stabilization demonstration is year-to-year. Several large PDP sponsors commented that they need narrowed risk corridors or other stabilization mechanisms to keep offering standalone drug plans. CMS basically punted on all of that, saying it cannot decide whether to continue the premium stabilization demonstration until it sees 2027 bids.

The investment thesis here is that PDPs need better Part D analytics, utilization management, and formulary optimization tools, and they need them from vendors who understand the PDP-specific economics, not just the MA-PD economics. The separate normalization creates a distinct financial environment for PDPs that is different from MA-PDs in terms of how risk is predicted and paid. Any company building Part D risk adjustment, bidding support, or pharmacy benefit analytics tools should be building for both segments now and pricing the products accordingly. There is also an opening for companies that help PDP sponsors improve their diagnosis data. CMS acknowledged that PDPs have a structural disadvantage in diagnosis submission because they do not control the medical encounter. Tools that help PDP sponsors get better diagnosis data from FFS claims, or that help them partner with provider organizations for data sharing, could be valuable in this new normalization environment.

Star Ratings Overhaul: Depression Screening, Measure Removals, and Quality Tech

The Star Ratings changes in the CY 2027 Rate Announcement and the companion final rule are the kind of thing that sounds boring until you realize how much money is on the line. MA quality bonus payments are tied to Star Ratings, and a contract’s Star Rating determines whether it gets a 5 percentage point benchmark bonus, a 3.5 point bonus, or nothing. For a large MA contract, the difference between 3.5 stars and 4 stars can be hundreds of millions of dollars.

CMS is removing 11 measures from the Star Ratings, adding a Depression Screening and Follow-Up measure starting with the 2027 measurement year (for 2029 Star Ratings), keeping the historical reward factor instead of implementing the Health Equity Index reward that the prior administration had developed, and adding or updating four measures for the 2027 Star Ratings cycle including Colorectal Cancer Screening (respecified), Care for Older Adults Functional Status Assessment (returning after spec change), Concurrent Use of Opioids and Benzodiazepines, and Polypharmacy with Multiple Anticholinergic Medications in Older Adults.

The Depression Screening measure is the one to watch from a venture perspective. Behavioral health measurement in MA has been a gap forever. Adding a Part C depression screening measure creates demand for clinical workflow tools, member engagement platforms, and data capture systems specifically designed to ensure that plans can document depression screening at scale across their enrolled populations. This is not a small operational lift. Plans need to screen millions of members, document the screening, and ensure follow-up if the screen is positive. That is a care coordination problem, a data capture problem, and a member engagement problem all wrapped into one quality measure.

The polypharmacy and opioid-benzodiazepine measures create similar demand for medication management platforms. These are clinical pharmacy measures that require plans to have visibility into their members’ full medication profiles and the ability to intervene when dangerous combinations are detected. Companies building medication therapy management tools, pharmacist intervention platforms, or prescriber alert systems are well-positioned here.

The removal of 11 measures focused on administrative processes is also worth noting. CMS is explicitly saying it wants Star Ratings to focus on clinical outcomes and patient experience, not on administrative checkbox compliance. For quality improvement vendors, this means the sales pitch needs to shift from “we help you check boxes” to “we help you move clinical outcomes.” That is a harder product to build but a stickier one.

Skin Substitutes, Risk Model Freeze, and the Coding Economy

CMS decided not to finalize the proposed 2027 CMS-HCC risk adjustment model recalibration and will instead continue using the 2024 model for CY 2027. This was the single biggest change from the Advance Notice to the final Rate Announcement and is the main reason the payment increase jumped from 0.09 percent to 2.48 percent. The proposed model would have used 2023 diagnoses and 2024 expenditure data, which included the anomalous skin substitute spending that inflated certain condition coefficients while deflating others.

The skin substitute story is worth understanding because it illustrates how FFS payment policy changes ripple through MA in unexpected ways. Spending on physician-administered drugs, which includes skin substitutes, went from 9.66 dollars per member per month in 2023 to 22.26 in 2024 to 40.04 in 2025, and then CMS projects it crashes to 1.53 in 2026 because of the CY 2026 Physician Fee Schedule rule that reclassified skin substitutes from biologicals to incident-to supplies. That reclassification cut payment by roughly 90 percent. CMS tested what would happen if they reduced skin substitute prices in the proposed model and found that raw MA risk scores would have been only 0.1 percent lower, meaning the skin substitute distortion was real but small in aggregate. But the condition-level impacts were all over the place. Skin condition relative factors went up 42.9 percent in the proposed model versus the 2024 model, while lung conditions went down 14.3 percent, kidney went down 14.3 percent, and metabolic went down 8 percent.

By freezing the model at the 2024 version, CMS avoided this mess for now. But they have signaled clearly that they will recalibrate in a future year. For startups in the risk adjustment space, this creates a window of relative stability in the HCC coefficient structure. But it also means that when the recalibration does come, it will be a bigger jump than if it had been done incrementally. Companies that help plans model the financial impact of risk adjustment model changes, stress-test their revenue under different coefficient scenarios, and optimize their coding and documentation strategies around the specific HCCs that are most likely to move are going to be in high demand when CMS proposes the next recalibration.

The normalization factors for 2027 are also worth noting. The 2024 CMS-HCC model normalization factor is 1.079, up from where it has been. Normalization is CMS’s mechanism for keeping the average MA risk score anchored to 1.0 in FFS, and the factor trends upward as coding intensity in FFS increases. The 5.90 percent coding pattern difference adjustment, which is the statutory minimum, continues to be a blunt instrument that some commenters argue should be higher and others argue should be lower. For risk adjustment vendors, the normalization and coding pattern dynamics are the bread and butter of their value proposition, and nothing in this Rate Announcement changes that.

GLP-1s, BALANCE, and the Part D Pharmacy Benefit Chaos

The Part D sections of the Rate Announcement are a masterclass in reading between the lines. CMS finalized updated RxHCC models using 2023 diagnoses and 2024 expenditure data, reflecting IRA benefit changes for 2027 including the increased manufacturer discount for specified small manufacturers. The deductible for the defined standard benefit goes from 615 dollars to 700 dollars, and the out-of-pocket threshold goes from 2,100 to 2,400. The annual percentage increase driving these parameter updates is 13.65 percent, which reflects a 9.37 percent annual drug spending trend plus 3.92 percent in prior year revisions. That 9.37 percent trend tells you everything about what is happening in Part D drug spending.

Multiple commenters raised concerns about the BALANCE Model, which is the CMS Innovation Center model focused on GLP-1 receptor agonists scheduled to start in CY 2027. CMS basically said that the design of Innovation Center models is outside the scope of the Rate Announcement. But the anxiety is real. GLP-1 utilization is accelerating, and the Part D benefit redesign has increased plan liability in the catastrophic phase. Plans are simultaneously dealing with higher drug costs, higher plan liability, and uncertainty about whether CMS will negotiate prices on additional drugs.

The startup opportunities in this space are obvious and large. Part D actuarial and bidding support tools that can model the interaction between GLP-1 utilization projections, IRA benefit changes, risk corridor parameters, and premium stabilization scenarios. Specialty pharmacy management platforms that help plans manage high-cost drug categories while maintaining Star Ratings performance on medication adherence measures. Prior authorization and utilization management tools specifically calibrated for the post-IRA Part D benefit structure. And formulary optimization engines that can account for maximum fair prices on negotiated drugs, manufacturer discounts under the redesigned benefit, and the new separate PDP/MA-PD normalization dynamics.

ESRD and the Kidney Care Opportunity

The ESRD payment sections got a lot of comment activity. Several commenters argued that state-level ESRD rate setting masks within-state variation in kidney care costs and that the MOOP limit creates an implicit cross-subsidy from non-ESRD to ESRD enrollees. CMS acknowledged the concerns but did not change the methodology. The agency pointed to prior analyses showing that moving to sub-state (CBSA-level) ESRD rates would actually reduce payments in areas with higher deprivation indices, which is why they have not pulled the trigger.

For kidney care startups, the signal is that CMS is aware of the ESRD payment adequacy problem but is constrained by the statute (Section 1853(a)(1)(H)) from making dramatic changes. Plans are going to have to manage ESRD costs within the existing rate structure, which means they need better tools for kidney care management, transplant coordination, dialysis utilization management, and functioning graft monitoring. The ESRD risk adjustment models (2023 ESRD CMS-HCC models) are separate from the regular HCC model and have their own normalization factors and coefficient structures. Companies that build ESRD-specific analytics and care management tools have a niche but defensible market because the payment mechanics are genuinely different from the rest of MA.

Puerto Rico as a Micro-Market

CMS continues its longstanding adjustments for Puerto Rico, basing MA county rates on beneficiaries with both Part A and Part B (unlike the mainland, where Part A-only and Part B-only beneficiaries are included) and adjusting for the higher proportion of zero-claims beneficiaries. The adjusted FFS costs in Puerto Rico get a 4.4 percent uplift from the zero-claims adjustment. Multiple commenters, including plans operating on the island, argued that rates are still inadequate given the extremely high MA penetration (far higher than any state), the high proportion of dual-eligible beneficiaries, provider infrastructure challenges, and the ongoing movement of both providers and beneficiaries to the mainland.

This is a small market but a real one. Companies that can serve the specific operational needs of MA plans in Puerto Rico, including bilingual member engagement, provider network management in a strained delivery system, and D-SNP operations for the heavily dual-eligible population, have limited competition and genuine demand. The political dynamics also make it likely that CMS will eventually do something more substantial for Puerto Rico rates, whether through demonstration authority or legislative action, which would expand the market further.

What to Build

Reading this Rate Announcement as a venture thesis, the highest-conviction opportunities cluster around a few themes. Encounter data infrastructure and chart review linking tools are the most urgent, because the unlinked chart review exclusion takes effect for CY 2027 and plans need to adapt now. Point-of-care risk adjustment capture tools that reduce dependence on retrospective chart review are the medium-term play. Part D bidding, analytics, and formulary optimization platforms that account for the separate PDP/MA-PD normalization and the IRA benefit redesign are underfunded relative to the complexity of the problem. Depression screening and behavioral health quality measurement tools are a new category created by the Star Ratings changes. Medication management platforms targeting the new polypharmacy and opioid safety measures have a clear regulatory catalyst. And ESRD-specific care management platforms address a payment adequacy problem that CMS has acknowledged but not solved.

The common thread is that CMS is simultaneously tightening the rules on how plans get paid (chart review exclusions, audio-only exclusions, coding pattern adjustments) while increasing the operational complexity of what plans need to do (new quality measures, IRA benefit changes, separate Part D normalization). That gap between tighter payment rules and higher operational demands is where startups live. Plans do not have the internal capability to build all of this themselves, and the timeline is too short for them to try. They need to buy.

For angel investors and seed-stage builders in health tech, the Rate Announcement is basically a procurement forecast disguised as a regulatory document. Every mandate CMS finalizes is a purchase order that has not been written yet. The plans know they need to comply. The question is just who builds the tools they will buy to do it.​​​​​​​​​​​​​​​​

The Timing Is Not a Coincidence

The 2.4% Update and What the Numbers Look Like

The Non-Hospice Spending Crisis Nobody Talks About

Operation Never Say Die: When the Data Becomes a Mugshot

SSVI: CMS Finally Gets a Fraud Scorecard

Making the Election Addendum Mandatory

The Wage Index Problem and the BLS RFI

MAID, Palliative Care, and the Policy Frontier

Quality Reporting Gets a Shaming Mechanism

So What Does This Mean for Investors and Builders

Abstract

CMS published its FY 2027 Hospice Wage Index and Payment Rate Update as a proposed rule (CMS-1851-P) on April 6, 2026, two days after Operation Never Say Die arrests in LA

Operation Never Say Die: 15 defendants charged, 8 arrested, $60M in alleged Medicare fraud across LA County; one nurse had an 85% non-death discharge rate, nearly 5x the national average

More than 700 of approximately 1,800 hospice providers in LA County alone have triggered multiple fraud red flags; CMS Administrator Dr. Oz stated in April 2026 that he believes roughly half of LA’s hospices are fraudulent

CMS has revoked 220 hospice approvals in California in 10 weeks; suspended 221 providers in a single week (up from 70 the prior week)

Estimated $3.5B in fraudulent Medicare reimbursements from LA County alone per House Oversight Committee

Total estimated financial impact of FY 2027 rule: $785M in increased payments to hospices

Proposed payment update: 2.4% (hospital market basket 3.2% minus 0.8% productivity adjustment)

New payment rates: RHC days 1-60 rises from $230.83 to $236.56; RHC days 61+ from $181.94 to $186.53; GIP from $1,199.86 to $1,232.71; hospice cap to $36,210.11

Non-hospice spending during a hospice election hit $2.8B+ in FY 2024 across Parts A, B, and D, up from $1.3B in FY 2020

Skin substitute billing during hospice stays went from $18M to $714M in four years

For-profit hospices averaged 167% higher non-hospice spending per day vs. non-profits in FY 2024, up from 60% in FY 2022

CMS introduces SSVI (Service and Spending Variation Index), a 0-16 score across nine metrics for each hospice, scoring 6,642 to 6,735 providers; published for FY 2024 and FY 2025

Election statement addendum proposed to become mandatory for all elections, not just on request

RFI on BLS-based hospice-specific wage index to replace the IPPS hospital wage index

RFI on MAID, currently legal in 11 states and DC

Care Compare tool to get a non-compliance icon for quality reporting failures, no earlier than FY 2028

Non-compliance with quality reporting was 23.53% of hospices in FY 2025, despite a 4-point APU penalty

Enhanced CMS oversight already active in AZ, CA, NV, TX, GA, and OH; more than 200 Medicare enrollment revocations in the four original states alone

The Timing Is Not a Coincidence

On April 2, 2026, FBI agents in Los Angeles fanned out across Covina, Anaheim, Glendale, and Lakewood before 6 a.m., executing arrest warrants on doctors, nurses, a chiropractor, and a psychologist. By the time the press conference wrapped downtown, federal prosecutors had charged 15 people in a scheme they called Operation Never Say Die, a $60 million Medicare hospice fraud ring spanning nine separate investigations across Southern California.

The same day, CMS published its FY 2027 Hospice Wage Index and Payment Rate Update proposed rule. That timing was not accidental. CMS Administrator Dr. Oz stood at the Operation Never Say Die press conference alongside FBI and DOJ officials and said publicly that he believes roughly half of the hospice providers operating in Los Angeles County are fraudulent. For context, LA County has somewhere around 1,800 hospice providers. Nationwide there are about 6,600. That means roughly one-third of all hospices in the entire country are concentrated in a single county, which Dr. Oz called out by name as immediately suspicious. One building in Van Nuys had 197 registered hospice companies on a single address. Another plaza housed 89.

The proposed rule and the enforcement action are best read as companion documents. The rule codifies the measurement infrastructure and transparency mechanisms. The arrests show what the measurement infrastructure is meant to catch.

The suspects in Operation Never Say Die follow a pattern that is by now disturbingly familiar to anyone tracking hospice fraud enforcement. Recruiters approach people at grocery stores, offering them $300 a month to pose as dying patients. A nurse named Lolita Minerd ran a hospice in Artesia that submitted over $9.1 million in fraudulent Medicare claims, collecting $8.5 million. Her hospice had a survival rate of 85%, which is approximately five times the national average for a population supposed to have a six-month prognosis. A couple named Gladwin and Amelou Gill, both of whom had prior tax evasion convictions that should have barred them from operating a hospice, opened one using their daughter’s name and collected over $4 million in Medicare payments. A 76-year-old woman named Nita Palma allegedly ran three fraudulent hospice operations while simultaneously incarcerated in a federal prison in Seattle and free on bond for a separate prior fraud case. A single dermatologist was associated with 63 different hospice facilities across California and billed Medicare more than $35 million in 2025 alone.

The House Oversight Committee estimates that Medicare was defrauded roughly $3.5 billion in LA County alone from fraudulent hospice reimbursements. JD Vance’s Anti-Fraud Task Force, which led the Operation Never Say Die coordination, has suspended 221 hospice providers in a single week, up from 70 the prior week. Federal authorities have said they expect clusters of similar takedowns every few months as the 2026 enforcement initiative continues to audit high-billing hospice agencies nationwide. CMS has already revoked 220 hospice approvals in California in just 10 weeks, a pace Dr. Oz noted was approaching what the Newsom administration accomplished in four years.

All of that is the fire. The FY 2027 proposed rule is the sprinkler system. Neither one alone is sufficient.

The 2.4% Update and What the Numbers Look Like

The proposed payment update of 2.4% is the hospital inpatient market basket increase of 3.2% reduced by a 0.8 percentage point productivity adjustment. CMS uses IGI’s fourth quarter 2025 forecast and reserves the right to revise if better data arrives before the final rule, which typically lands in August. The total financial impact is estimated at $785 million in increased payments over FY 2026.

Broken out by level of care: RHC days 1 through 60 moves from $230.83 to $236.56. RHC days 61 and above goes from $181.94 to $186.53. Continuous home care at a 24-hour full rate moves from $1,674.29 to $1,728.02, or about $72 per hour. Inpatient respite care goes from $532.48 to $546.46. General inpatient care, which is the acute symptom management level, goes from $1,199.86 to $1,232.71.

Hospices that do not meet quality data submission requirements get the inverse: a -1.6% update instead, which is the 2.4% payment increase minus the 4-point APU penalty. Non-compliant RHC day 1-60 rate drops to $227.32. Non-compliant GIP drops to $1,184.56. This penalty structure was increased from 2 to 4 percentage points beginning in FY 2024, and as discussed later, it has not moved the needle on compliance rates.

The aggregate cap, which limits total annual Medicare payments per patient, is proposed at $36,210.11, up from $35,361.44. The Consolidated Appropriations Act of 2026, signed in February 2026, extended the market basket-based cap update methodology through 2035, meaning the CPI-U approach that originally governed cap increases will not return until at least then.

The Service Intensity Add-On budget neutrality factors for FY 2027 are proposed at 0.9999 for both RHC categories, meaning SIA utilization is essentially flat year over year. SIA pays for direct RN or social worker care in the last seven days of a patient’s life under RHC conditions, capped at four hours per day at the CHC hourly rate. Budget neutrality requires that SIA payments be offset by a slight reduction in the base RHC rate to keep total RHC spending constant.

The Non-Hospice Spending Crisis Nobody Talks About

Buried inside the same proposed rule is a data story that is frankly remarkable in how bluntly CMS tells it. Non-hospice Medicare spending for beneficiaries already enrolled in hospice grew from $790 million in FY 2020 to over $2 billion in Parts A and B alone in FY 2024, a 160% increase in four years. Adding Part D drugs brings the combined FY 2024 total above $2.8 billion. The single-year jump from FY 2023 to FY 2024 was $770 million, or 60 percent growth in one year.

This matters because the hospice per diem is explicitly designed to be all-inclusive. When a patient elects hospice, they waive Medicare payment for services related to their terminal illness outside the hospice benefit. The hospice receives a daily rate covering essentially everything. Every dollar billed outside the hospice benefit during a hospice election is, at minimum, a compliance question and, at worst, a fraud signal.

The driver that stands out most aggressively in the CMS data is carrier and physician supply claims. Those grew 317.5% from FY 2020 to FY 2024. The largest single diagnosis within that category is pressure ulcers, almost entirely from skin substitutes, which went from $18 million to $714 million, a roughly 4,000% increase over four years. CMS made major changes to skin substitute reimbursement effective January 1, 2026, transitioning most products to a national unified rate of approximately $127.14 per square centimeter under a new incident-to classification designed to end the ASP-plus-6-percent billing model that created this runaway incentive. The problem is that the 2020 through 2024 data shows the escalation in full and CMS is not pretending it was accidental.

The breakdown by diagnosis is even more pointed. For patients with neurological and degenerative conditions like Alzheimer’s and Parkinson’s, there was roughly $576 million in non-hospice DME and carrier claims in FY 2024, plus $205 million in Part D. For circulatory and cerebrovascular patients, about $590 million in non-hospice DME and carrier claims plus $276 million in Part D. CMS asks directly in the rule: why are bronchodilators being billed to Part D for a respiratory hospice patient? Why is oxygen being billed separately when the hospice per diem is supposed to cover it? Why are anticoagulants, beta blockers, and vasodilators appearing as outside claims for heart failure patients in hospice?

The for-profit versus non-profit split is the most politically damaging data point. Beneficiaries receiving hospice services from for-profit providers averaged 167% higher non-hospice spending per day compared to patients in non-profit hospice care in FY 2024. That same gap was 60% in FY 2022, meaning it nearly tripled in two years. And 67% of all non-hospice spending occurred after election day 60, meaning this is not a phenomenon concentrated in the enrollment adjustment period. It is chronic and systematic.

Operation Never Say Die: When the Data Becomes a Mugshot

The $2.8 billion non-hospice spending figure and the 167% for-profit premium on outside billing are aggregate statistics. Operation Never Say Die translates that aggregate into individual criminal defendants.

The structural pattern is consistent across the LA cases: open a hospice using someone else’s name or credentials if your own history makes you ineligible, recruit non-dying patients via cash kickbacks paid to referral agents and directly to beneficiaries, submit claims to Medicare for services that are either medically unnecessary or never happened, and collect reimbursements until you get caught or go out of business. The beneficiaries collecting $300 a month from Lolita Minerd are the supply side of a demand driven entirely by Medicare’s per diem payment structure.

The per diem pays whether or not any service is rendered on a given day. Under the routine home care level, which is the dominant billing category, a hospice can collect $236 per day for a patient who is visited once a week for 20 minutes. The incentive to enroll ineligible, relatively healthy patients who do not require intensive services and to keep them enrolled as long as possible is not subtle. An 85% survival rate at one of the defendant hospices, when the eligibility standard is a six-month terminal prognosis, is not a statistical outlier. It is business model documentation.

The clustering patterns CMS and DOJ identified in LA County make this worse. When 197 hospices share a single address in Van Nuys or 89 share a single commercial plaza, what you are looking at is a systematic exploitation of CMS’s enrollment processes. California instituted a moratorium on new hospice licenses in 2021, but the enforcement gap between when fraud begins and when licenses get revoked gave existing fraudulent operators years to run at scale. Authorities found that initial fraudulent activity in the Operation Never Say Die cases goes back as early as 2018, with some cases not reaching arrest stage until April 2026, an eight-year window.

Federal authorities have explicitly said they expect this to become a rolling national enforcement campaign, with clusters of similar takedowns across high-billing hospice markets every few months. CMS has already expanded enhanced oversight beyond the original four states, Arizona, California, Nevada, and Texas, into Georgia and Ohio. The pattern recognition work CMS is doing at the aggregate level in the proposed rule, particularly the SSVI, is designed to create the targeting infrastructure for that campaign nationally.

SSVI: CMS Finally Gets a Fraud Scorecard

The SSVI, or Service and Spending Variation Index, is a new scoring system published for the first time in this proposed rule. Each of the roughly 6,600 active hospices gets a score between 0 and 16, calculated from nine claims-based metrics. CMS is publishing FY 2024 and FY 2025 SSVI data alongside the proposed rule, covering 148 million and 156.5 million hospice days respectively.

The nine metrics are: providing no CHC and no GIP service at all, which gets 1 point; percentage of RHC days delivered in a nursing home or skilled nursing facility at 40% or above, 1 point; percentage of last two RHC days of life with visits at or below the 25th percentile (which was 85.7% in FY 2025), 1 point; live discharge rate at or above the 75th percentile (47.5% in FY 2025), 1 point; percentage of discharges with length of stay over 180 days at or above the 75th percentile (33.2% in FY 2025), 1 point; average skilled nursing minutes per RHC day at or below the 25th percentile (9.8 minutes per day in FY 2025), 1 point; weekend RHC days with a skilled visit at or below the 25th percentile (4.8% in FY 2025), 1 point; live discharges where the beneficiary returns to the same hospice within 7 days at or above the 75th percentile (15% in FY 2025), 1 point; and total non-hospice spending, scored on an 8-tier scale worth up to 8 points.

The non-hospice spending scoring structure is where the index gets most of its discriminatory power. The tiers run from 1 point for any non-hospice spending below $6,352 per year up to 8 points for the highest-spending eighth of hospices, which starts above $517,204. Because this single component can contribute up to 8 of the 16 possible points, SSVI scores are heavily loaded toward the non-hospice spending signal. That design is not accidental. The other metrics capture operational patterns like thin visit volumes, high live discharge rates, and long stays, which are the utilization signatures that federal investigators found in every Operation Never Say Die defendant’s claims history.

The SSVI was clearly not designed as a quality improvement instrument, despite CMS framing it carefully that way. CMS says directly in the rule that a high score signals a hospice that may require additional targeted education or oversight, including medical review, education, and investigations that could result in payment suspension and revocation if fraud, waste, or abuse is identified. That is law enforcement language dressed in program integrity clothing. The SSVI is the data infrastructure that gives CMS a ranked list of providers to hand to enforcement teams for prioritization.

For context, the defendant hospice with an 85% non-death discharge rate would score near the maximum on at least two or three SSVI metrics immediately: the live discharge rate metric, almost certainly the length of stay distribution metric, and potentially the non-hospice spending metric depending on how much outside billing accompanied the phantom care. That kind of multi-metric clustering is exactly what CMS says it is looking for.

For health tech investors, the SSVI is a new public data asset that creates an entire product category. Provider benchmarking, competitor quality intelligence, referral network risk assessment, post-acute analytics, and M&A diligence for hospice operators all now have a publicly available claims-derived 0-to-16 score updated annually. Companies that can contextualize that score alongside payer mix, staffing ratios, length of stay distributions, and operational benchmarks have a product with obvious buyers across payers, IDNs, referral platforms, and PE-backed hospice roll-ups trying to distinguish clean assets from liability-laden ones.

Making the Election Addendum Mandatory

The hospice election statement addendum explains to patients in plain language what the hospice benefit covers, what it does not cover, and what happens to their Medicare coverage for outside services once they elect hospice. Since the FY 2020 rule created it, hospices have only been required to provide it upon patient request. That provision is remarkably toothless given the information asymmetry between a patient enrolling in hospice during a medical crisis and a provider who knows exactly what the benefit covers and what they intend to bill elsewhere.

CMS is now proposing to make it mandatory at the time of every hospice election. This is procedurally straightforward but operationally significant at scale. Every intake workflow across 6,600 providers needs to be updated to include delivery, documentation, and patient or representative acknowledgment of the addendum. The connection to the fraud backdrop is direct: an informed patient who understands that her hospice is supposed to be covering wound care is better positioned to ask uncomfortable questions when that same care suddenly appears as a separate Medicare charge.

For technology vendors operating in hospice intake, documentation, and compliance, this is a clear new workflow requirement. The implementation window before the FY 2027 effective date of October 1, 2026 is short enough that any provider without a streamlined addendum delivery and documentation process is already late.

The Wage Index Problem and the BLS RFI

The hospice wage index has used the pre-floor, pre-reclassified hospital inpatient wage index as its base since the mid-1990s. The problem with this proxy is that hospital labor cost data is built around acute inpatient staffing, primarily physicians, specialists, and inpatient nurses, while hospice labor is predominantly hospice aides, registered nurses working in community settings, and social workers. The occupational mix is fundamentally different, and in many labor markets, especially rural ones, the divergence between hospital wages and community home-based care wages is substantial.

CMS convened a Technical Expert Panel in September 2025 and is now formally soliciting comment on a potential hospice-specific wage index using Bureau of Labor Statistics Occupational Employment and Wage Statistics data. The proposed methodology is detailed: calculate CBSA-level wage estimates for 10 occupational categories, weight them by a national occupational mix derived from freestanding hospice cost report expense data and claims-based minutes of care, divide by the national aggregate to produce an index value, then apply the existing hospice floor and 5% cap on decreases.

The proposed occupational mix from that methodology is revealing. Hospice aides at 38.11% of the mix and registered nurses at 28.46% account for nearly two-thirds of the weighting. That is a very different picture from what drives hospital wage indexes, where physician compensation and specialty nursing play much larger roles. The remaining 10 occupational categories cover nursing administration, physician services, LPNs, medical social services, nurse practitioners, and therapists.

A shift to BLS-based hospice-specific wage indexing would produce winners and losers across markets. Rural areas where hospice labor costs diverge significantly from hospital labor costs stand to see payment rate changes that could run into the millions of dollars annually for larger operators. The transition policy question CMS is explicitly soliciting input on matters enormously for anyone modeling long-run operating economics in the hospice space.

This is a multi-year policy signal rather than a one-year action. But the combination of a completed TEP, a detailed methodology document, and structured public comment questions covering data sources, occupational mix, geographic delineation, and transition design suggests this is moving toward implementation within a few rulemaking cycles. Anyone doing capital allocation in hospice operations, whether as an operator, a PE sponsor, or an early investor, needs this uncertainty in their underwriting assumptions.

MAID, Palliative Care, and the Policy Frontier

Two RFIs in this rule are genuinely forward-looking in ways that are easy to overlook given the fraud enforcement drama surrounding it. The medical aid in dying section and the community palliative care section both signal a CMS that is actively mapping the policy edges of end-of-life care.

On MAID: the Assisted Suicide Funding Restriction Act of 1997 prohibits federal funds from paying for any service intended to cause or assist in causing death. That law has not changed. What has changed is the number of states legalizing MAID, currently 11 states plus Washington DC, with more moving in that direction. The eligibility requirement in most state MAID laws is a six-month terminal prognosis, which is identical to the Medicare hospice eligibility standard. That means a meaningful and growing population of hospice-enrolled patients in MAID-legal states are simultaneously eligible for both programs.

CMS wants to understand what actually happens in practice. Do hospices continue providing clinical care while a patient pursues MAID qualification? Do patients remain on service until natural death or revoke election to pursue MAID? What happens with unused lethal medications? The questions are operational rather than philosophical, and the underlying policy concern is making sure federal hospice payments are not directly or indirectly subsidizing MAID services, which the 1997 law explicitly prohibits. CMS is also asking whether additional oversight mechanisms are needed to enforce that prohibition more reliably.

The community palliative care RFI is the longer-term signal. Medicare does not have a standalone palliative care benefit. Patients who need serious illness support but have not yet elected or are not eligible for hospice fall into a patchwork of Part B evaluation and management codes, advance care planning CPT codes 99497 and 99498, chronic care management codes, and limited home health coverage. CMS is asking whether these existing pathways can be optimized to produce more seamless community palliative care without requiring new legislation or a new benefit category.

For investors and founders building in the serious illness population management, advance care planning, or community-based palliative care delivery spaces, this RFI is a material signal. CMS is explicitly naming the pre-hospice transition zone as a policy priority and asking for detailed input on billing practices, care gaps, documentation burdens, and staffing barriers. The infrastructure built on top of the existing billing code landscape today will be positioned as the natural foundation if a more formal palliative care pathway emerges in future rulemaking.

Quality Reporting Gets a Shaming Mechanism

The HQRP section of this rule continues the transition from the old Hospice Item Set to the newer HOPE assessment tool with the iQIES platform. The novel proposal is the Care Compare icon.

Roughly 20 to 24 percent of hospices have been non-compliant with quality reporting requirements in recent fiscal years. In FY 2023 the rate was 20.07%. In FY 2024, the first year of the 4-point APU penalty, it rose to 22.06%. In FY 2025 it peaked at 23.53% before settling to 20.37% in FY 2026. The doubling of the financial penalty from 2 to 4 percentage points has produced essentially zero improvement, and the non-compliance rate has never dropped below 20%.

The proposed response is a consumer-facing icon on the Medicare.gov Care Compare tool identifying hospices that failed the 90% HOPE submission threshold. The icon would appear on both the provider search page and individual hospice profile pages, similar to how penalty flags appear for nursing homes and hospitals. It is proposed to go live no earlier than FY 2028, based on CY 2026 submission data.

The logic of public shaming outperforming financial penalties in this setting is actually defensible. The APU penalty hits cash flows that a financially distressed or fraudulent operator may not care about. A visible compliance failure flag on the primary consumer-facing comparison platform hits something different: referrals. Hospices depend heavily on physician referrals and on family choices made during acute medical transitions. A red flag on Care Compare, visible to any family member running a quick search, creates reputational exposure that financial penalties do not. For fraudulent operators, CMS non-compliance flags are also input data for enforcement prioritization, which makes the icon a dual-purpose tool.

For health IT vendors in the hospice documentation and quality reporting space, this creates a hard deadline and a clear value proposition. Getting clients above the 90% submission threshold before October 2027 is now a competitive differentiator, because the CY 2026 data used for the FY 2028 icon is being generated right now.

So What Does This Mean for Investors and Builders

The FY 2027 hospice proposed rule and Operation Never Say Die land in the same week for a reason. CMS is simultaneously writing the infrastructure rules and executing the enforcement actions, and both are accelerating. Dr. Oz’s statement that he believes roughly half of LA’s hospices are fraudulent, if anything close to that is accurate nationally even at a fraction of the LA intensity, represents one of the largest active fraud situations in any Medicare sector.

For investors, there are a few distinct opportunity signals here. The SSVI creates a new public data asset for the post-acute analytics market. Provider benchmarking, referral network quality scoring, and M&A diligence for hospice operators now have a claims-derived 0-to-16 annual scorecard that did not exist before. Analytics companies that can layer SSVI data against operational benchmarks, payer mix, staffing ratios, and length of stay distributions have obvious buyers across payers, IDNs, referral platforms, and PE sponsors managing hospice roll-ups.

The compliance urgency around the mandatory election addendum and the Care Compare icon creates near-term workflow and documentation demand. These are administrative problems that affect thousands of sites simultaneously, and a significant share of the hospice market, particularly smaller independent providers, lack the internal infrastructure to handle them cleanly. That is repeatable revenue for documentation platforms, compliance software, and quality reporting tools already embedded in hospice operations.

The non-hospice spending enforcement trajectory and the SSVI together create genuine demand for care coordination technology that helps hospices manage patients comprehensively within the per diem. The 167% higher outside spending rate at for-profit hospices is not sustainable under the new transparency and enforcement environment. Hospices that can demonstrate comprehensive care delivery and low non-hospice spend ratios will have a competitive advantage in both referral networks and enforcement priority ranking. Technology that helps document, coordinate, and optimize within-benefit care delivery is building into a regulatory tailwind that is only getting stronger.

The BLS wage index transition is a multi-year underwriting risk factor. If finalized, it would change payment rates in specific markets by amounts that could be material to operating margins, particularly in rural and high-wage urban markets where the divergence from hospital labor costs is largest. Anyone deploying capital into hospice operations in markets that diverge significantly from their hospital wage index peers should be modeling this exposure explicitly.

The palliative care pre-hospice transition zone is a medium-term market formation signal. CMS is clearly interested in formalizing community palliative care pathways, is asking detailed questions about billing and delivery gaps, and has named this as a policy priority. Infrastructure built on the existing code landscape today, spanning serious illness population identification, ACP documentation, and community-based care coordination, is well-positioned for the eventual formalization of that pathway.

None of this is an argument that hospice is a sector to avoid. End-of-life care is a demographic inevitability. The need is growing, not shrinking. But the regulatory sophistication required to operate and invest in this space just stepped up significantly. CMS is building real measurement capability, publishing provider-level scores, designing consumer-facing enforcement mechanisms, and coordinating with DOJ on criminal prosecutions at a pace and scale the hospice industry has not seen before. The operators and technology vendors who understand that this is not a temporary crackdown but a permanent infrastructure shift will find themselves on the right side of where this market is going.​​​​​​​​​​​​​​​​

Background: How We Got Here

Star Ratings Overhaul: The $18.56B Question

Part D Redesign Codification: Locking In the IRA Changes

Supplemental Benefits and Debit Card Guardrails

Regulatory Rollbacks: The Deregulation Layer

Health Equity in Retreat

Investment Signals and Opportunity Map

Abstract

Published April 2, 2026; effective June 1, 2026; applicable to coverage starting Jan 1, 2027

Covers Medicare Advantage (Part C), Part D, and Medicare Cost Plan regulations

Key moves: 11 Star Rating measures cut, Health Equity Index reward scrapped, IRA Part D redesign codified into permanent reg, debit card supplemental benefit guardrails added, significant deregulatory measures finalized under EO 14192

Financial impact: Star Ratings changes estimated at $18.56B net to Medicare Trust Fund over 10 years (2027-2036)

~42,632 public comments received on the proposed rule

MA enrollment is roughly 33M+ beneficiaries; program represents ~$500B+ in annual federal spending

Biggest investor signals: quality measurement contraction favors scaled incumbents, debit card benefit infrastructure gets formalized, health equity tech demand softens near-term, Part D data infrastructure plays get a longer runway

Background: How We Got Here

Anyone who has been tracking Medicare Advantage closely knows the program has been under sustained pressure since 2023. Medical loss ratios spiked, several large insurers (UnitedHealth, Humana, CVS/Aetna) took significant earnings hits, and CMS started signaling that the decade-long era of generous benchmark rates and flexible quality bonuses needed recalibration. The CY2027 final rule, dropped April 2, lands against that backdrop. This is not a massive structural reform. It reads more like a careful tightening – some policy housekeeping, some politically-motivated deregulation, and a consequential set of decisions about the Star Ratings architecture that will ripple through plan economics for the next decade.

To understand why this rule matters, worth remembering what the Star Ratings system actually does. Plans scoring 4 stars or above receive Quality Bonus Payments from CMS, which in turn generate additional rebate dollars that plans use to fund supplemental benefits, reduce premiums, or pocket margin. The difference between a 3.5-star and 4-star contract can easily translate to hundreds of millions of dollars annually for a large plan. CMS has been tinkering with this system for years, but the changes finalized here are among the more significant structural adjustments since the program matured. Eleven measures are getting cut, the Health Equity Index reward is getting shelved, and a new depression screening measure is being added. None of these are cosmetic.

The Part D side of this rule is in some ways more technically dense but less surprising. The Inflation Reduction Act of 2022 made sweeping changes to the Part D benefit structure – eliminating the coverage gap, capping out-of-pocket at $2,000 for 2025 (now $2,100 for 2026, indexed forward), replacing the Coverage Gap Discount Program with the Manufacturer Discount Program – and CMS implemented most of those changes via program instruction because the IRA gave them specific authority to do so through 2026. That authority expires. This rule codifies all of it into permanent reg text. For investors and builders in the PBM, specialty pharmacy, and drug pricing analytics space, that formalization matters because program instructions can be walked back; regs are harder to unwind.

Star Ratings Overhaul: The $18.56B Question

The headline number here is $18.56 billion. That is CMS’s estimate of the net impact to the Medicare Trust Fund from the Star Ratings changes, spread over 2027 through 2036. For context, that works out to roughly 0.21 percent of Medicare payments to private health plans during that period. On a per-year basis, you are talking somewhere in the range of $1.5 to $2B annually depending on enrollment growth. That is a real number with real consequences for plan financials, and therefore real consequences for the vendor and technology ecosystem that serves those plans.

The 11 measures being removed are targeted at what CMS describes as administrative processes and areas where plan performance has converged to the point where beneficiaries cannot meaningfully distinguish between plans. A few of the notable cuts include the Call Center foreign language interpreter and TTY availability measures (applicable starting with 2028 Star Ratings), and the Statin Therapy for Cardiovascular Disease measure on the Part C side. The Depression Screening and Follow-Up measure is being added for the 2027 measurement year, flowing into 2029 Star Ratings. That addition is worth paying attention to – it signals that behavioral health integration into primary care workflows remains a CMS priority, even as the agency trims the broader measure set.

The Health Equity Index (HEI) reward decision is probably the most politically loaded piece. CMS had developed the HEI as a mechanism to reward plans that showed better performance specifically for historically underserved subpopulations – dual eligibles, low-income subsidy recipients, people with disabilities. The concept was that plans gaming overall averages by performing well for healthier, wealthier enrollees while underserving higher-need populations should not receive the same quality bonuses. The HEI was supposed to correct for that. CMS is now shelving the HEI entirely and sticking with the “historical reward factor,” which rewards consistent high overall performance across all measures over time. The administrative record cites ongoing concerns about the methodology, but the political context here matters: this administration has been explicitly rolling back DEI-related initiatives across federal agencies, and the HEI sits squarely in that crosshairs. Worth watching whether a future administration revives something like it.

For investors in health equity analytics, SDOH-focused platforms, and disparity measurement companies, this is a setback. Not a fatal one – Medicaid managed care and commercial value-based arrangements still create real demand for equity measurement – but the MA-specific revenue thesis just got softer. Companies that positioned specifically around the HEI compliance market are going to need to pivot. Companies with broader equity measurement value propositions (risk stratification, SDOH data aggregation, language services) are more insulated because the underlying clinical and operational demand does not disappear just because CMS removed a specific Star Ratings incentive.

The practical implication of cutting 11 measures while adding 1 is that the overall measure set gets smaller and thus each remaining measure carries more relative weight. From an actuarial standpoint, this concentrates risk. Plans that are strong on the surviving measures benefit disproportionately. Plans with weaknesses in surviving measure domains – CAHPS scores, medication adherence, chronic disease management outcomes – face more concentrated downside. Any tech company operating in those surviving measure domains (medication adherence platforms, patient experience survey infrastructure, chronic condition monitoring tools) should be reframing their value prop accordingly.

Part D Redesign Codification: Locking In the IRA Changes

The codification of the IRA Part D changes is in some ways the least dramatic piece of this rule from a market-impact standpoint because the benefit changes were already in effect. Beneficiaries have been living the new Part D world since 2025. What this rule does is convert temporary program instruction authority into permanent regulation, which has a few important downstream effects.

First, it removes ambiguity about durability. The IRA’s program instruction authority to implement the Part D redesign was time-limited. Converting to reg means these policies – no coverage gap, $2,000 OOP cap (now indexed at $2,100 for 2026 and going forward), zero cost sharing in catastrophic, Manufacturer Discount Program replacing the Coverage Gap Discount Program – are now in the regulatory fabric. Reversing them would require new notice-and-comment rulemaking, which is a much higher bar than simply issuing revised program guidance. For any investor or operator trying to model Part D economics over a multi-year horizon, this codification meaningfully reduces regulatory tail risk.

Second, it formalizes the Manufacturer Discount Program at scale. Under the old Coverage Gap Discount Program, manufacturers provided discounts only in the coverage gap phase. Under the Manufacturer Discount Program, discounts apply in both the initial coverage phase and the catastrophic phase for applicable drugs. Manufacturers who participate provide discounts of 10 percent in initial coverage and 20 percent in catastrophic. This is a fundamentally different economics structure for branded drug manufacturers, and it creates a different set of data and analytics needs. Who is responsible for tracking discount obligations at the NDC level across phases? How do plan sponsors and PBMs reconcile manufacturer discount payments? This is a non-trivial operational problem that has already spawned a cottage industry of Part D reconciliation and discount program management tooling. Codification makes that market more durable.

Third, the Selected Drug Subsidy piece is worth flagging for the data analytics crowd. For drugs that CMS has negotiated a Maximum Fair Price under the Inflation Reduction Act’s Drug Price Negotiation Program, the cost-sharing and subsidy structure in Part D is different. CMS now pays a 10 percent subsidy in initial coverage for selected drugs, and 40 percent reinsurance in catastrophic (compared to different rates for non-selected drugs). Tracking which drugs are “selected drugs” during which “price applicability periods,” and applying the correct liability attribution logic for each drug, is exactly the kind of problem that requires real-time reference data infrastructure. The codification of these rules creates a durable, reg-anchored market for companies providing that data layer.

The out-of-pocket cap dynamics deserve a closer look from an investment thesis standpoint. The $2,000 OOP cap for 2025 (indexed to $2,100 for 2026) fundamentally changes the risk profile of high-cost specialty drug users in Part D. Before the IRA, beneficiaries with specialty drug needs could face thousands of dollars in out-of-pocket costs annually. Now they hit a hard cap. This changes the financial calculus for medication adherence – when a patient’s cost sharing is capped at $2,100 regardless, the financial friction to starting or continuing a specialty drug is dramatically reduced compared to the old benefit structure. For companies in the medication access, patient affordability, and specialty pharmacy support space, this is a tailwind that the codification now makes permanent.

Supplemental Benefits and Debit Card Guardrails

The supplemental benefits section of this rule is where MA’s evolution as a consumer product runs directly into CMS’s concern about program integrity. MA plans have increasingly competed on supplemental benefits – dental, vision, hearing, over-the-counter (OTC) allowances, transportation, meal delivery, fitness memberships, and a host of other services beyond traditional medical coverage. The debit card mechanism became a popular administrative vehicle because it gave plans flexibility to offer dollar allowances across multiple benefit categories, and it gave beneficiaries flexibility in how they used those dollars. The problem is that flexibility created real and documented program integrity gaps: allowances used outside of covered benefit categories, funds used for non-health-related purchases, inconsistent point-of-sale verification, and confusion about what the benefit actually covers.

CMS is finalizing a set of debit card guardrails that address the most obvious integrity gaps. The key requirements are that debit cards must be electronically linked to covered items through a real-time verification mechanism at the point of sale, and debit card benefits must be limited to the specific plan year (no carryover). CMS is also requiring enhanced disclosure for beneficiaries about what the card covers and what it does not. Notably, CMS walked back the proposed prohibition on marketing the dollar value of supplemental benefits, which had generated substantial industry pushback. Plans can still advertise “up to $X in OTC benefits” in their marketing materials.

For the health tech ecosystem, the real-time point-of-sale verification requirement is the operationally meaty piece. This essentially requires that the debit card infrastructure be connected to a benefit eligibility determination engine that can validate at checkout whether a specific product or service is a covered benefit for the member. This is harder than it sounds. Product catalogs for OTC benefits are enormous (tens of thousands of SKUs), benefit definitions vary by plan and geography, and the verification needs to happen fast enough not to disrupt the retail checkout experience. There are already vendors operating in this space – companies that provide eligible item catalogs, pharmacy benefit processors, and specialized OTC card program administrators – but the codification of the real-time verification requirement creates a regulatory mandate for capability that was previously more of a best-practice.

Worth noting that CMS is also finalizing the SSBCI (Special Supplemental Benefits for the Chronically Ill) transparency requirements from the CY2026 proposed rule. Plans must now post their SSBCI eligibility criteria publicly on their websites. SSBCI is the bucket that allows plans to provide non-primarily health-related benefits (things like air conditioning units, pest control, home modifications) to members with chronic conditions. The eligibility criteria for these benefits have been notoriously opaque, and this transparency requirement at least creates a reference point for advocates, regulators, and researchers trying to understand who is actually accessing these benefits and who is not.

The cannabis product clarification in this rule is also worth a sentence. CMS is amending the SSBCI regulations to specify that cannabis products illegal under applicable state or federal law are not allowable SSBCI benefits. This is largely defensive – a clarification designed to prevent plans from inadvertently including cannabis products in benefit designs, particularly as state-level legalization continues to expand. It does not open any door for cannabis in MA benefits; it closes off the ambiguity that some might have tried to exploit.

Regulatory Rollbacks: The Deregulation Layer

This is where the current administration’s footprint on this rule is most visible. Under Executive Order 14192, CMS is rolling back a set of requirements across MA and Part D. Some of these are genuinely sensible deregulation removing requirements that generated compliance costs without meaningful beneficiary benefit. Others are more consequential reductions in consumer protection infrastructure.

On the genuinely sensible side: exempting health reimbursement arrangements and HSA-linked plans from creditable coverage disclosure requirements is a reasonable administrative streamlining. The mid-year notice requirement about unused supplemental benefits is also being rescinded. CMS’s logic here is that plans voluntarily communicate this information anyway, and mandating a specific notice format generates cost without clear evidence of improved utilization. These are modest changes.

On the more consequential side: removing the restrictions on when and how licensed agents and brokers can have conversations with beneficiaries is a meaningful rollback. Previous rules had tightened the conditions under which agents could contact beneficiaries, partly in response to documented cases of high-pressure sales tactics and inappropriate switching behavior that increased insurer costs and destabilized plan enrollment. The removal of these restrictions does not eliminate the underlying prohibition on unsolicited contacts, but it loosens the guardrails in ways that could increase aggressive marketing activity heading into the 2027 Annual Enrollment Period.

The elimination of health equity requirements from MA Utilization Management Committees deserves particular attention. Under the CY2024 rule, CMS had required UM committees to include a health equity expert and to conduct annual health equity analyses of coverage criteria, prior authorization decisions, and formulary designs. Plans were required to post those analyses publicly. These requirements were designed to address documented disparities in prior auth denial rates by race and income. CMS is eliminating all of this: no health equity expert requirement, no annual analysis, no public posting. For companies that had built consulting or analytics products specifically around MA UM health equity compliance, this is a direct revenue headwind.

The LINETS (Limited Income Newly Eligible Transition) call center hour waiver is a smaller but symbolically loaded rollback. LINETS is the program that auto-enrolls low-income beneficiaries who age into Medicare into Part D plans. These beneficiaries are disproportionately lower-income and have less digital literacy, meaning they are more dependent on telephone support to understand their coverage. Waiving the requirement that LINETS plans maintain toll-free call centers accessible from 8 AM to 8 PM across all regions reduces cost for plans but at potential access cost for a vulnerable population.

Health Equity in Retreat

It is worth taking a step back and naming the aggregate pattern here because the individual provisions are easier to rationalize in isolation than as a whole. In this single rule, CMS has shelved the HEI reward, eliminated health equity expertise requirements from UM committees, eliminated the annual health equity analyses requirement, eliminated the public posting requirement for those analyses, and waived extended call center access for low-income auto-enrollees. Each of these can be individually justified on administrative efficiency or methodological grounds. Taken together, they represent a significant reduction in CMS’s formal commitment to measuring and incentivizing equity within MA.

For investors in health equity tech, the near-term investment thesis in MA-specific equity work just got materially weaker. The honest framing is that the regulatory tailwind driving plan spending on equity technology was substantially MA-dependent, and that tailwind has reversed. The more durable equity tech plays are in Medicaid (where federal equity requirements are more entrenched and state-level variation creates persistent demand), in value-based care arrangements with progressive health systems, and in employer-sponsored insurance where DEI commitments from large self-funded employers still create demand for equity analytics infrastructure.

That said, the clinical reality that equity gaps produce worse outcomes and higher costs does not disappear because CMS stops measuring it. Risk-adjustment models that ignore SDOH systematically underestimate costs for high-need populations, and that creates actuarial exposure for plans regardless of whether CMS rewards equity performance. The smarter equity tech companies understand that their core value proposition is cost and risk prediction accuracy, not compliance. Companies that positioned primarily as compliance plays are going to struggle; companies that positioned as actuarial accuracy and population health effectiveness plays have a more durable thesis even in this environment.

Investment Signals and Opportunity Map

Pulling all of this together, the CY2027 rule produces a fairly clear set of investment signals for the health tech ecosystem. The Star Ratings measure contraction is net-negative for quality measure infrastructure vendors that built revenue around the 11 eliminated measures, and net-positive for companies focused on surviving measure domains, particularly CAHPS and medication adherence. The Part D codification is net-positive for anyone in the drug pricing data, specialty pharmacy analytics, or Manufacturer Discount Program reconciliation space. The debit card guardrails create a formal regulatory mandate for real-time eligible item verification infrastructure. The health equity rollbacks create near-term headwinds for MA equity compliance tech but do not eliminate the broader actuarial and population health case for SDOH analytics.

Two categories worth highlighting for angel investors and early-stage funds specifically: Part D reconciliation and Selected Drug tracking infrastructure, and OTC benefit eligibility verification technology. Both of these are unsexy, plumbing-layer problems that generate real, recurring revenue from large plan sponsors and PBMs. The codification of the IRA changes and the debit card guardrails respectively create long-duration regulatory underpinning for both categories. These are not moonshot opportunities. They are the kind of B2B infrastructure plays that generate predictable SaaS-style contracts with MA plans and PBMs at scale.

The request for information section of this rule is also worth monitoring even though CMS does not respond to RFI comments in the final rule text. CMS asked for input on future directions for the MA program, modernizing agent/broker oversight, the dramatic growth in dual-eligible enrollment in chronic condition SNPs (C-SNPs), and nutrition policy in MA. The C-SNP enrollment question is particularly interesting – dual eligibles are the highest-cost, highest-complexity population in Medicare, and the rapid growth of C-SNP enrollment suggests plans are competing aggressively for what they presumably view as an attractive risk cohort. Whether the actual risk profiles and actuarial assumptions underlying that competition are sound is an open question, but the growth trajectory makes C-SNP-focused care models, navigation tools, and complex care management platforms a logical area to watch.

The MA market in 2026 is not the same market it was in 2021. Plan exits, benefit reductions, and premium increases have changed the competitive dynamics. The CY2027 rule does not reverse those dynamics but it does lock in a new set of operational and financial parameters within which the market will evolve. For investors and operators trying to read the regulatory tea leaves, the summary version is this: quality measurement is consolidating around fewer, more clinically meaningful metrics; Part D economics are permanently restructured by the IRA; supplemental benefit integrity requirements are tightening; and CMS under the current administration is explicitly prioritizing deregulation and beneficiary choice over health equity measurement. Position accordingly.​​​​​​​​​​​​​​​​

March 26, 2026. HHS and CMS dropped the member list for a new federal Healthcare Advisory Committee. 18 people. Picked from 400+ nominees. Reporting to RFK Jr. and Dr. Oz. Non-binding recommendations across Medicare, Medicaid, CHIP, and the Marketplace. The five stated priorities: chronic disease, outcomes accountability, real-time data, vulnerable populations, and Medicare Advantage sustainability. The roster mixes major health system operators (Cleveland Clinic, Sanford, Intermountain), value-based care builders (VillageMD), health IT infrastructure (Availity), community health (NACHC), and Tony Robbins. Yes that Tony Robbins. This piece covers:

- Who these people are and why the mix is interesting

- The five priority buckets and where the money follows

- Medicare Advantage as the thing everyone should actually be watching

- Real-time data as an infrastructure thesis hiding in plain sight

- MAHA as a policy direction with real market consequences

- Whether any of this matters given it’s technically advisory and non-binding

Table of Contents

The Roster and What It Signals

Breaking Down the Five Priorities

Medicare Advantage Is the Real Story

The Data Infrastructure Play

MAHA Is a Market Signal Whether You Like It or Not

Non-Binding Doesn’t Mean Irrelevant

The Roster and What It Signals

Skip the quotes in the press release. Every federal advisory committee announcement produces the same sentences about practical solutions and putting patients first. It’s not that anyone is lying, it’s just that the quotes are written to be unobjectionable and therefore contain approximately zero information. What contains information is who they actually put in the room.

The health system operators are the first thing worth looking at. Bill Gassen runs Sanford Health, which is one of the largest rural health systems in the country, operating primarily across the Dakotas and upper Midwest with a footprint that touches communities most coastal health tech companies have never thought about. Dennis Laraway is the CFO of Cleveland Clinic, which runs some of the tightest financial operations of any major academic medical center in the country and has been aggressive about data infrastructure and cost management in ways that a lot of comparable systems haven’t matched. Dan Liljenquist is EVP and Chief Strategy Officer at Intermountain Health, former Utah state senator, former Bain consultant, and the person who quite literally invented Civica Rx, a nonprofit generic drug manufacturer he dreamed up while on a treadmill and then actually built into a real company that now has a domestic manufacturing facility coming online in Virginia. That particular combination of policy brain, strategy chops, and willingness to build new institutions outside existing incentive structures is pretty unusual and his presence on this committee is not decorative.

Then there is Clive Fields, who co-founded VillageMD and has spent years building out the primary care infrastructure thesis that Walgreens eventually paid several billion dollars for before the whole thing got complicated. His background is squarely in value-based primary care at scale and he has thought harder than most people about what physician enablement actually requires operationally. Russ Thomas is the CEO of Availity, which is the largest health information network in the country by transaction volume and sits in the pipes connecting payers and providers at a level most people in digital health don’t fully appreciate. If the committee is serious about real-time data and administrative simplification, having the person who runs that particular piece of infrastructure at the table is not an accident.

Kyu Rhee has one of the more interesting career arcs on the list. Former CMO at HHS, former Chief Health Officer at IBM Watson Health during the period when Watson was supposed to transform everything, former SVP and CMO at CVS Health’s Aetna business, and now President and CEO of the National Association of Community Health Centers, which represents 1,512 federally qualified health centers serving roughly 52 million patients. His presence reads as the committee’s anchor on the vulnerable populations priority and also brings a track record of having worked at the intersection of data, AI, and clinical delivery at a point in time when that intersection was still being figured out. Jenni Gudapati holds a PhD and has been involved in value-based care and population health in ways that complement the clinical operator voices elsewhere on the list.

The two ex officio members, Kimberly Brandt and Stephanie Carlton, are the government insiders keeping the committee connected to actual CMS operations. Brandt in particular has deep background in Medicare oversight and program integrity going back to prior CMS leadership roles, which matters when the agenda includes anything touching fraud reduction or quality measurement integrity.

And then there is Tony Robbins. Look, the honest reaction from most people in health policy circles was somewhere between confusion and amusement. He is a motivational speaker and self-help figure who recently appeared at a CMS quality conference and apparently made enough of an impression on Dr. Oz to land a seat at this table. The charitable read is that he brings a consumer behavior and behavior change lens that is genuinely missing from most policy conversations. The less charitable read is that this is an administration that likes names people recognize. Both things can be true. Either way he is one vote among eighteen and the committee’s output will be shaped by the operators and clinicians, not by the person famous for walking on hot coals.

Breaking Down the Five Priorities

Read more

Where This Investment Philosophy Comes From

The Man Behind the Method: What Bill Frist Actually Knows That Other VCs Don’t

The Portfolio Under the Microscope: Riding Policy Waves Across Specialties

When the Tailwind Stalls: The Bicycle Health Case Study

What This Tells Us About How to Build and Fund in Healthcare

The Road Ahead: Regulatory Winds Are Shifting Again

Abstract

Frist Cressey Ventures (FCV), founded in 2016 by former Senate Majority Leader Bill Frist, MD and PE veteran Bryan Cressey, has built a nearly $1B AUM early-stage healthcare firm around a deceptively simple thesis: find the policy tailwind, back the company surfing it. This piece analyzes how FCV’s origin story, deeply rooted in Frist’s firsthand legislative experience passing landmark laws like the Medicare Modernization Act and PEPFAR, has translated into a repeatable investment philosophy. Using FCV’s disclosed portfolio (including Thyme Care, Monogram Health, Bicycle Health, CodaMetrix, Devoted Health, Axuall, and others from the attached funding data), the essay explores which companies are riding specific regulatory tailwinds, how this strategy has played out in markets over time, and where the approach faces genuine risk.

Key data points referenced:

- FCV Fund IV: $425M, oversubscribed, total AUM near $1B (Feb 2026)

- Thyme Care: $97M Series D (Sept 2025) after four rounds totaling $234M+

- Monogram Health: $375M Series C (Jan 2023) from ~$5M Series A in 2019

- Bicycle Health: $50M Series B (May 2022) plus $16.5M venture round (Jan 2025)

- CodaMetrix: $55M Series A (Feb 2023), $40M Series B (Mar 2024)

- Devoted Health: $1.15B Series D (Oct 2021)

- Core regulatory drivers: Medicare Modernization Act (MMA) of 2003, CMS Enhancing Oncology Model (EOM, 2023), ESRD KKHI Executive Order (2019), SAMHSA buprenorphine final rule (2024), 21st Century Cures Act interoperability rules

Where This Investment Philosophy Comes From

There is a concept worth naming directly: regulatory arbitrage in healthcare venture capital. Not in the pejorative tax-evasion sense, but in the structural sense. The idea is that when a major piece of policy shifts the rules of reimbursement, care delivery, or market access, a narrow window opens where smart capital and product teams can get to scale before incumbents figure out what happened. The window is real. It closes. And the firms that see it earliest, usually because they have someone who helped write the rules sitting on their investment committee, tend to win disproportionately.

FCV is the clearest institutional embodiment of this strategy in healthcare venture. Frist himself has said it about as plainly as a VC ever does: “The only way to get to large scale is good policy.” That quote deserves unpacking because it is doing a lot of work. He is not saying policy is good for society, though presumably he believes that. He is saying that in a $3T market where most dollars flow through Medicare, Medicaid, commercial insurance, and regulated pharmaceutical channels, scale is fundamentally downstream of reimbursement, and reimbursement is fundamentally downstream of policy. If you accept that premise, then pattern-matching legislative cycles becomes a legitimate investment input, not just a nice-to-have for a diligence memo.

This is not a new idea. But it is rarely institutionalized the way FCV has done it. Most funds have one policy advisor or a government affairs guy they call occasionally. FCV has the person who was literally sitting in the Senate Finance and HELP committees when most of the modern healthcare regulatory architecture was being built. Frist served on both committees, was Majority Leader when the MMA passed in 2003, and championed PEPFAR before it had mainstream bipartisan buy-in. When he says he sees regulatory tailwinds, he is not using a metaphor. He is describing a pattern he has observed from both inside the legislative engine room and from decades of watching the market respond.

The Man Behind the Method: What Bill Frist Actually Knows That Other VCs Don’t

Read more

Chamber Cardio closed a $60M Series A in February 2026, led by Frist Cressey Ventures, with participation from General Catalyst, AlleyCorp, Optum Ventures, Healthworx Ventures, American Family Ventures, Company Ventures, Black Opal Ventures, and debt from HSBC Innovation Banking. The company is building value-based care infrastructure for cardiology, operating a dual-sided network model that partners with both payers and cardiologist practices simultaneously. Key data points:

- CVD is the number one driver of US healthcare spend, responsible for roughly 1 in 3 deaths

- 500+ cardiologists across 7 states at time of Series A announcement

- Workflow-native AI embedded directly into EHR systems

- Seed led by General Catalyst at $8M

- Strategic investors include Optum Ventures and Healthworx (CareFirst), signaling payer validation

- Competitive set includes Karoo Health, Novocardia, CardioOne, Heart and Vascular Partners

- Regulatory tailwinds: LEAD Model, ACCESS Model (10-year CMMI program), ARPA-H ADVOCATE program for agentic AI in cardiology

Table of Contents

The Problem Nobody Fixed

Why Cardiology, Why Now

The Model: Both Sides of the Table

The Technology Bet

Investor Signal Reading

Regulatory Wind at Their Back

The Competitive Landscape

Where the Risks Actually Live

The Exit Thesis

The Problem Nobody Fixed

Cardiology has a weird distinction in American healthcare. It is simultaneously the specialty most responsible for patient mortality, the single largest category of healthcare spending in the US, and one of the last major clinical areas to get a serious value-based care infrastructure built around it. That is not an accident. Cardiologists are a particular breed of specialist, historically well-compensated under fee-for-service, protective of their clinical autonomy, and deeply skeptical of administrative overhead. Getting them to change how they operate is not a technology problem. It is a trust and incentive problem. And that is exactly what makes Chamber interesting.

The traditional fee-for-service dynamic in cardiology works roughly like this: a patient sees their PCP, gets referred to a cardiologist, the cardiologist does a procedure or schedules a follow-up, bills for the encounter, and moves on. Whether that patient fills their beta-blocker prescription, shows up to their follow-up three months later, or ends up in the ER with a preventable CHF exacerbation is, structurally speaking, not the cardiologist’s financial problem. This is not a knock on cardiologists as clinicians. Most of them went into medicine to help people. But the payment architecture is not wired to reward longitudinal management. It rewards procedural throughput.

The downstream cost of this is enormous. Cardiovascular disease accounts for roughly 17% of total US healthcare expenditure, which works out to somewhere north of $400 billion annually when you add up direct medical costs and indirect costs from productivity loss. Readmissions for heart failure alone cost Medicare billions per year. A meaningful chunk of that is preventable with better population health management, timely medication titration, and early identification of high-risk patients before they escalate. The data exists to do this. The workflows to act on it, at scale, inside cardiology practices, largely do not. That is the problem Chamber is trying to fix.

What makes this moment different from prior attempts is the combination of EHR maturity, AI capability, and payer willingness to actually write checks for value-based arrangements in cardiology specifically. For years, VBC was primarily a primary care story. ACOs were built around PCPs. Risk-based arrangements largely excluded specialists or tried to subordinate them under primary care-led structures. Cardiologists, watching this play out, mostly stayed on the sidelines or got pushed into clumsy shared savings arrangements that did not reflect their actual role in driving cost and outcomes. The infrastructure was not built with them in mind.

Why Cardiology, Why Now

The timing argument for Chamber is not just about the size of the market, though that helps. It is about a confluence of factors that makes cardiology value-based care specifically tractable right now in a way it was not five years ago.

Start with the epidemiology. The US is aging, and cardiovascular disease prevalence scales aggressively with age. The cohort of Americans over 65 is growing faster than any other demographic segment, and this population disproportionately carries multiple cardiovascular comorbidities simultaneously. Heart failure, atrial fibrillation, coronary artery disease, and hypertension do not travel alone. A significant portion of high-risk cardiology patients also carry diabetes, chronic kidney disease, and obesity, all of which interact with cardiovascular risk in complex ways. Managing these patients well requires coordination across specialties, longitudinal follow-up, and proactive intervention. The fee-for-service model handles none of this gracefully.

The data infrastructure has also matured considerably. EHR penetration in cardiology practices is high, and the major platforms, Epic, eClinicalWorks, Athenahealth, have robust APIs that allow third-party applications to embed into clinical workflows at a level of fidelity that was not really possible at scale until the last few years. FHIR-based interoperability, whatever its limitations in practice, has made it meaningfully easier to aggregate claims data, lab data, and clinical notes across the care continuum and do something useful with it in near-real-time. This matters because the intelligence Chamber is selling to cardiologists is only as good as the underlying data pipeline.

The payer appetite has shifted too. Large commercial payers and Medicare Advantage plans have been burned by broad ACO arrangements that promised savings and delivered mediocre performance. They are now more interested in specialty-specific value-based contracts where the risk is more bounded and the clinical levers are better understood. Cardiology is an obvious target because the spend is concentrated, the patient population is identifiable, and the interventions that reduce cost, medication adherence, timely follow-up, early detection of decompensation, are well-understood clinically. Payers can build a reasonably credible actuarial model around this. That is what makes them willing to sign contracts.

And then there is the workforce math. There are roughly 23,000 practicing cardiologists in the US serving a patient population that by most projections will need substantially more cardiology capacity over the next decade as the Baby Boomer cohort ages into peak cardiovascular risk. You cannot simply train more cardiologists fast enough. The only way to extend cardiologist capacity is to make each cardiologist more effective, which means eliminating low-value work from their day, giving them better signal on which patients need attention right now, and offloading routine monitoring and care coordination to a clinical team they can supervise. This is the operational argument for Chamber’s model, and it is the part that tends to resonate most with cardiologists when it is explained clearly.

The Model: Both Sides of the Table

What Chamber is doing structurally is more nuanced than a lot of the press coverage suggests. The company is not a managed care organization, it is not acquiring practices, and it is not simply a software vendor. It is operating as a two-sided infrastructure layer between payers and cardiologist practices, and understanding why that is strategically important requires unpacking both relationships.

On the payer side, Chamber signs value-based contracts that give health plans visibility into cardiovascular performance, total cost of care, and quality metrics across a defined network of cardiologists. The payer is essentially buying a more manageable cardiology population and cleaner performance data in exchange for sharing some of the savings generated by better care management. For Medicare Advantage plans in particular, this is attractive because cardiology spend is a major driver of medical loss ratio, and RAF scores tied to cardiovascular diagnoses have significant premium implications. Chamber’s ability to surface coding gaps and ensure accurate risk capture is a side benefit that plans appreciate but rarely talk about publicly.

On the provider side, Chamber partners with cardiology practices without acquiring them. This is a deliberate and important choice. Practice acquisition is capital-intensive, creates employment relationships that are complicated to manage across states, and tends to generate the kind of cultural friction that drives away exactly the high-performing cardiologists you need in your network. Chamber’s approach is to come in as a technology and care management partner, wrap operational support around the practice, and share the upside from value-based contract performance. The cardiologist stays independent, keeps their brand, and gets access to a care coordination infrastructure they could not build themselves.

The technology sits in the middle of both relationships. Chamber’s platform does population stratification across the practice’s entire panel, identifying which patients are highest risk for near-term hospitalization or deterioration. It surfaces care gaps, specifically instances where a patient is not on guideline-directed therapy or has not had a recommended follow-up. It integrates directly into the EHR, so the cardiologist is seeing Chamber’s recommendations inside the workflow they already use rather than logging into a separate portal. And it routes lower-acuity work to pharmacists, nurse practitioners, and case managers who can handle medication titration and follow-up calls without consuming a cardiologist’s time. The result, in theory, is a cardiologist who is seeing sicker patients, capturing more accurate diagnosis codes, ordering fewer unnecessary tests because they have better longitudinal data, and generating better outcomes metrics that make the payer contract renew at better economics.

The bet here is on network density. The value of Chamber’s platform to a payer scales with how many cardiologists in a given market are inside the network, because that is what allows the payer to actually manage cardiovascular spend across a population rather than just a subset of it. Building this density state by state requires a lot of practice development work, and it is slow. Seven states and 500 cardiologists at Series A is a meaningful start but represents a tiny fraction of the addressable market. The $60M is largely going toward accelerating that build-out.

The Technology Bet

Chamber’s CMO, Dr. Sameer Sheth, put it well in the Series A announcement: cardiologists do not need more data, they need a clearer signal. That framing is actually a pretty sophisticated product thesis disguised as a one-liner. The failure mode for most clinical decision support tools in cardiology has been information overload. You build a dashboard, populate it with every possible risk metric, and then watch the cardiologist ignore it because it requires too much interpretation effort during a 15-minute appointment. Physicians are not against decision support. They are against decision support that adds cognitive load without reducing it.

What workflow-native AI is supposed to do differently is prioritize and translate rather than just aggregate. The system is not presenting a cardiologist with a panel of 400 patients and asking them to figure out who needs attention. It is surfacing the three patients who are highest risk this week and telling the cardiologist specifically what intervention the evidence suggests. It is flagging the patient on suboptimal beta-blocker dosing and pre-populating the message to the pharmacist. It is identifying the patient who missed their echo follow-up six months ago and automatically scheduling an outreach call. The cardiologist touches this stuff, but they are not the bottleneck for it.

The AI layer has another function that is harder to talk about publicly but drives a lot of the ROI. Accurate hierarchical condition category coding is enormously valuable in Medicare Advantage. Patients with cardiovascular conditions carry significant RAF weight, and a practice that does not systematically document comorbidities and disease severity is leaving premium dollars on the table for the MA plan and underrepresenting the true complexity of their patient population. Chamber’s platform, by surfacing coding gaps and prompting documentation during encounters, helps practices capture this more accurately. This is not upcoding. It is closing documentation gaps that exist because cardiologists are busy and documentation is annoying. But the financial impact is real, and payers are aware of it when they are evaluating whether the value-based contract economics work.

The ARPA-H ADVOCATE program announced in early 2026 is worth flagging specifically. ARPA-H is looking to fund the development of FDA-authorized agentic AI technology that can provide what they are describing as 24/7 specialty care in cardiovascular medicine. The government is essentially signaling that it wants to see AI move from clinical decision support into autonomous care delivery in cardiology. Chamber is not ARPA-H, and ADVOCATE is still early-stage, but the regulatory and funding environment it represents is favorable to any company that has already built the data infrastructure and workflow integrations needed to deploy AI recommendations in cardiology at scale. If agentic AI in cardiology is coming, the companies with the richest longitudinal patient data and the deepest EHR integrations start with a significant structural advantage.

Investor Signal Reading

The cap table on this round tells you a lot if you know how to read it. The lead is Frist Cressey Ventures, co-founded by Senator Bill Frist, who is a cardiac and thoracic transplant surgeon. This is not a generalist firm that landed on cardiology. This is a firm with deep sector-specific conviction and clinical credibility that can open doors with health systems, payers, and cardiology groups in ways that a conventional VC cannot. Frist’s quote in the press release was not marketing. His perspective that cardiovascular care delivery is fragmented and fee-for-service driven reflects genuine clinical knowledge, and his firm’s willingness to lead at this size reflects real conviction about the market timing.

General Catalyst’s continued participation matters for a different reason. GC led the seed at $8M, and the fact that they followed on into the Series A signals that what Chamber showed between seed and Series A was sufficient to maintain their conviction. GC is running their Health Assurance thesis, which is broadly about companies that shift healthcare from reactive to proactive delivery. Chamber fits cleanly into that framework, and GC’s portfolio relationships, particularly in payer and health system land, are potentially valuable for Chamber’s market development.

The strategic investors are the most interesting signal. Optum Ventures is the investment arm of UnitedHealth Group’s Optum segment. Optum is simultaneously one of the largest health plans in the country, one of the largest physician groups, and a major health data and analytics business. Their investment in Chamber is not purely financial. It represents a payer and care delivery organization deciding that Chamber’s model is credible enough to merit a strategic relationship. Whether that eventually becomes a commercial partnership, a distribution arrangement, or an acquisition conversation is worth watching.

Healthworx Ventures is CareFirst BlueCross BlueShield’s venture arm. CareFirst is a large regional Blue operating in the Mid-Atlantic market. Again, a strategic health plan investor is not writing a check to make money on the carry. They are writing a check to get access, learn, and position for potential commercial engagement. Two strategic payer investors at Series A is unusual and meaningful. It suggests Chamber has already had enough substantive conversations with payers to generate this kind of interest, and it de-risks the go-to-market story considerably.

The debt from HSBC Innovation Banking is worth a brief note. Healthcare companies with recurring revenue tied to value-based contracts are increasingly attractive to venture debt providers because the revenue streams are relatively predictable once payer contracts are signed. Using debt alongside equity at Series A is smart capital efficiency: it limits dilution while funding growth, and the availability of the debt facility signals that Chamber’s revenue profile was legible enough to a sophisticated lender to underwrite.

Regulatory Wind at Their Back

The regulatory environment heading into 2026 is materially favorable for specialty-focused value-based care infrastructure, and Chamber specifically called this out in their Series A commentary. A few things worth understanding in detail.

The LEAD Model, which CMS announced in December 2025 as the successor to ACO REACH, is designed to extend accountable care to a broader set of Medicare beneficiaries, including those with complex chronic conditions like cardiovascular disease. ACO REACH had some success in improving care coordination for high-risk patients but struggled to meaningfully include specialists in a way that generated aligned incentives. The LEAD Model is structured to address some of these limitations, and companies that can help physician networks perform under LEAD will find a willing market.

The ACCESS Model from CMMI is potentially even more significant for Chamber’s long-term business. ACCESS is a 10-year payment program offering stable, recurring technology payments for chronic disease management across diabetes, hypertension, chronic kidney disease, obesity, and depression. Cardiovascular disease intersects with nearly all of these categories. A patient with CKD and hypertension is a high cardiovascular risk patient. A diabetic patient is at elevated risk for heart failure. If Chamber’s platform is credentialed as the technology layer enabling ACCESS-aligned care management, the recurring payment structure creates a revenue stream that is more predictable than payer contract performance alone.

The ADVOCATE program from ARPA-H is earlier stage but directionally important. Federal investment in agentic AI for cardiovascular care legitimizes the technology direction Chamber is pursuing and creates a pathway for AI capabilities to receive FDA authorization in clinical cardiovascular settings. This matters for reimbursement. Right now, a lot of the AI-enabled workflow improvements Chamber is selling generate value that accrues to the payer in the form of avoided costs. As AI-enabled care management becomes separately reimbursable, which ADVOCATE seems designed to enable, the revenue model for companies like Chamber could evolve substantially.

There is also the broader political context. The current administration has been friendlier to MA plan flexibility and specialist-inclusive payment models than prior administrations. MA enrollment now covers more than half of Medicare beneficiaries, and the MA market is where most of the commercial innovation in value-based care is happening. A political environment that supports MA plan autonomy to design specialist-inclusive value-based contracts is a tailwind for Chamber’s ability to sign and expand payer partnerships.

The Competitive Landscape

The honest answer is that the competitive landscape in specialty-focused value-based care infrastructure is still relatively early, which is both an opportunity and a reason for caution about market validation. Chamber’s named competitors include Karoo Health and Novocardia, with CardioOne and Heart and Vascular Partners occupying adjacent positions.

Karoo Health is building a similar dual-sided cardiology VBC model and has been relatively quiet publicly, which either means they are heads-down executing or not yet at a scale that generates press. Novocardia is physician-led and focused on heart failure specifically, which is a narrower patient population than Chamber’s broader cardiovascular disease approach. The narrower focus could be a moat or a ceiling depending on whether you believe heart failure VBC is a defensible category or a feature of a broader cardiology platform.

CardioOne and Heart and Vascular Partners are more squarely in the practice acquisition or management services model, which is a different structural bet than Chamber’s partnership approach. Practice acquisition is defensible in the sense that you control the clinician relationship more fully, but it is slower to scale, requires more capital, and creates different regulatory exposure. Chamber’s bet is that you do not need to own the practice to embed deeply enough that switching costs become real.

The EHR vendors and large health system players are a longer-term competitive consideration. Epic has population health tools built into their platform that overlap with what Chamber is doing. The difference, at least for now, is that Epic sells to health systems and large groups, while Chamber is targeting independent and semi-independent cardiology practices that often have less resources to configure and operationalize Epic’s VBC tools. If Epic or another large platform significantly deepens their out-of-the-box cardiology VBC capabilities and makes them accessible to smaller practices, that compresses Chamber’s differentiation. The counter-argument is that Chamber’s combination of technology plus actual clinical operations, the pharmacists, nurse practitioners, and care managers, is not something an EHR vendor is going to replicate.

Where the Risks Actually Live

Being honest about where this could go wrong is more useful than just cheerleading. A few areas that any serious diligence would stress-test.

Network density is the core growth variable and also the core risk. Chamber’s value to payers depends on controlling enough cardiologist relationships in a given market to actually move the needle on cardiovascular spend for a plan’s population. If Chamber has 50 cardiologists in a market of 400, a payer cannot realistically build a value-based contract around that. Getting to meaningful density requires practice development work that is intensive, relationship-driven, and slow. The $60M gives them runway to accelerate, but geography-by-geography market penetration is hard. If a competitor gets to density in key markets first, locking up the better practices, Chamber’s ability to sign payer contracts in those markets gets materially harder.

Payer contract economics are also not as simple as the narrative suggests. Value-based contracts in cardiology are negotiated, and payers have a lot of leverage in those negotiations, particularly for a company that does not yet have the multi-year outcomes data to prove its model with hard numbers. The first generation of these contracts probably have reasonable shared savings percentages but also significant performance thresholds that Chamber needs to clear before any money changes hands. Managing the timing between when Chamber invests in building out a market and when payer contract performance dollars actually flow is a real cash management challenge.

Cardiologist adoption and stickiness is the other big unknown. Chamber’s model requires cardiologists to actually change behavior based on the platform’s recommendations. Physicians are hard to change, and the history of clinical decision support is littered with tools that generated great pilot results and then got ignored once the sales team moved on. The workflow integration into EHRs reduces friction but does not eliminate it. The practices that perform well under Chamber’s model are probably the ones that were already inclined toward proactive population health management. Scaling to practices that need more convincing is operationally harder.

Finally, there is the regulatory execution risk around CMS program participation. Chamber’s growth strategy is explicitly tied to new CMS and CMMI payment models. These programs are designed in Washington, implemented by regional MACs and payer intermediaries, and subject to political and administrative changes. The ACCESS Model’s 10-year structure is appealing precisely because it creates durability, but program rules can change, implementation can be delayed, and the actual payments available to technology companies operating inside these frameworks are not always as large in practice as the program descriptions suggest.

The Exit Thesis

For the investors in this round, the exit landscape is actually pretty clear, which is part of why this was fundable at $60M.

Strategic acquisition is the most obvious path. The list of natural acquirers includes large health plans, diversified health services companies, and major health system operators. A large MA plan with significant cardiovascular spend across its membership has an obvious motivation to own the infrastructure Chamber is building rather than pay for it. Optum’s strategic investment creates an implicit optionality on an acquisition, which the management team is certainly aware of. A deal like this, if Chamber demonstrates strong performance under value-based contracts over the next two to three years, could reasonably be sized in the low-to-mid billions depending on revenue scale and network density.

The IPO path is longer and requires more scale, but it is not implausible. There is a precedent market for VBC infrastructure companies in the public markets, and a company with durable payer contracts, a growing cardiologist network, and demonstrable outcomes data has a reasonably legible story for public market investors. The window depends on market conditions and on whether Chamber can get to meaningful EBITDA or at least a credible path to it within the public company disclosure timeline.

The more speculative but interesting outcome involves the agentic AI trajectory. If the ARPA-H ADVOCATE program produces FDA-authorized agentic AI capabilities for cardiovascular care management, and if Chamber is positioned as one of the few companies with the data and clinical network infrastructure to deploy those capabilities at scale, the strategic value of the asset looks quite different. You are no longer just buying a VBC infrastructure company. You are buying the distribution network and data flywheel for the next generation of AI-enabled specialty care delivery. That is a different valuation conversation entirely.

Chamber is, fundamentally, a bet on the idea that cardiology’s transition to value-based care is inevitable, that the transition requires both technology and operational infrastructure that no existing player has built well, and that the company with the deepest payer relationships and cardiologist network density will control the economics of that transition for a long time. The $60M Series A is the fuel for building that density before the window closes. Whether that thesis plays out depends on execution, market timing, and whether the regulatory tailwinds Chamber is counting on actually materialize the way the current policy trajectory suggests they will. But the underlying logic is sound, the market is enormous, and the investor roster is credible enough to take seriously.​​​​​​​​​​​​​​​​

Specialty biologics now exceed 50 percent of total pharmacy spend and represent the fastest-growing component of medical loss ratio across commercial and government payers. Cell and gene therapies introduce single-event liabilities reaching multiple millions of dollars that existing stop-loss and reinsurance structures inadequately address. Current cost management approaches including pharmacy benefit managers, rebate contracting, site-of-care optimization, and prior authorization address pricing mechanics but fail to mitigate underlying actuarial volatility. The structural opportunity exists to build a healthcare-native risk pooling and financial engineering platform that transforms specialty pharmaceutical exposure into structured financial instruments combining reinsurance, asset management, pharmaceutical contracting, and predictive analytics. This company would function as the volatility dampener for biologic spend across self-insured employers, regional health plans, and Medicare Advantage organizations.

Table of Contents

The Structural Failure of Current Specialty Drug Risk Management

What Actually Needs to Get Built

How the Business Model Works

Technical Architecture and Data Infrastructure

The Predictive Modeling Stack

Risk Pooling Mechanics and Capital Structure

Regulatory Positioning and Licensing Strategy

Competitive Landscape and Why Nobody Has Done This Yet

Moat Development and Defensibility

Go-to-Market Execution Over 36 Months

Capital Requirements and Exit Scenarios

Why This Becomes Inevitable

The Structural Failure of Current Specialty Drug Risk Management

The pharmaceutical cost crisis everyone talks about misses the actual problem. Yeah drug prices are high and yeah PBMs are extracting rents and yeah manufacturers play games with list vs net pricing but none of that explains why actuaries at mid-size health plans are losing sleep over their specialty pharmacy book. The issue is volatility not absolute cost levels.

Consider the math facing a regional Blues plan covering 400k lives. Their total pharmacy spend runs maybe 2.5 billion annually with specialty representing 1.4 billion of that. Within specialty you have predictable high-cost maintenance therapy like Humira or Enbrel where utilization patterns follow established curves. Then you have the tail risk stuff. A member diagnosed with spinal muscular atrophy gets Zolgensma at 2.1 million as a one-time dose. Three members start CAR-T therapy for relapsed lymphoma at 475k each. Employer group with 800 lives has four members initiate Wegovy which cascades into 15 members on GLP-1s within six months fundamentally changing their pharmacy spend trajectory.

The actuarial models these plans use to set premiums and reserve requirements cannot accurately predict this kind of stuff. Disease progression modeling for rare conditions requires longitudinal data these plans do not have. Drug pipeline intelligence and FDA approval timing affects utilization curves but nobody integrates this into their forecasting. Real-world effectiveness data that would let you model adherence and outcomes sits in fragmented claims databases that are not linked to genomic or biomarker signals.

What you end up with is conservative pricing by stop-loss carriers who know they cannot model the risk accurately so they build in massive buffers. Small and mid-size payers get hit disproportionately hard because they lack scale to absorb the statistical noise. One bad case can blow up your medical loss ratio for the year. This creates premium instability, forces benefit design distortions like putting gene therapies in medical instead of pharmacy benefit to push them above stop-loss thresholds, and generally makes the whole system inefficient.

The tools payers currently use do not address actuarial volatility at all. PBMs negotiate rebates which affect net cost but do nothing for the timing or probability distribution of high-cost claims. Prior authorization just delays spend and creates administrative friction without changing the underlying exposure. Site-of-care steering moves a 15k infusion from hospital outpatient to physician office and saves 40 percent on admin fees but the drug cost is the drug cost. Outcomes-based contracting with manufacturers sounds good but the contracts are mostly vaporware because nobody has the data infrastructure to actually measure outcomes at scale or enforce clawbacks.

The gap in the market is someone who can take the volatility itself and turn it into a structured product that institutional investors will buy. You need to aggregate exposure across multiple payers to get statistical smoothing, build predictive models that actually work for tail risk, negotiate at portfolio scale with manufacturers, and then slice the risk into tranches that can be priced and sold. This is fundamentally a capital markets problem dressed up as a healthcare problem.

What Actually Needs to Get Built

Read more

The Datasets Worth Linking and Why

Home Health: A Perfect Fraud Ecosystem

The Entrepreneur’s Playbook for Monetizing Open Source Fraud Detection

Abstract

- The new HHS Medicaid provider spending dataset (NPI x HCPCS x month, 2018-2024, 227M rows) is powerful alone and exponentially more useful when joined against other free public datasets

- Key linkable datasets: NPPES provider registry (8.6M+ providers, entity formation dates, authorized officials), OIG exclusion list (sanctioned providers and individuals), CMS Open Payments (Sunshine Act financial relationships), PECOS Medicare enrollment data, [SAM.gov](http://SAM.gov) federal debarment records, state corporate registry filings, Census TIGER geographic data, HUD housing data, BLS employment statistics

- Home health is the highest-spend taxonomy in the dataset at $288B+ and the highest fraud-density category per dollar due to: zero verifiable clinical artifacts, self-attesting documentation, caregiver identity impossible to audit at scale, and federal matching rate economics that give states weak policing incentives

- The entrepreneur opportunity is a hybrid model: open source data analytics as the top-of-funnel fraud signal generator, wrapped with human investigative services, qui tam legal partnerships, and outcome-based government contracts

- False Claims Act whistleblower recoveries returned $2.80 per $1 spent on enforcement in the HCFAC program; the qui tam relator share is 15-30% of recovered funds

- Business model options range from SaaS sold to MCOs and state Medicaid agencies, to investigative services on contingency, to a qui tam legal referral engine, to a fully integrated fraud-to-recovery platform

The Datasets Worth Linking and Why

The Medicaid provider spending dataset that dropped last week is genuinely novel but it is also genuinely incomplete as a standalone fraud detection tool. What it gives you is a time series of how much money flowed from Medicaid to a specific provider NPI for a specific HCPCS procedure code in each month from January 2018 through December 2024. What it does not give you is almost everything else you need to decide whether that billing pattern represents fraud, waste, or legitimate healthcare delivery. The fraud detection signal lives in the gap between what the spending data shows and what a comprehensive picture of that provider’s real-world existence looks like. Closing that gap requires joining the Medicaid data against a stack of other public datasets that are all free, all downloadable, and all dramatically underutilized by anyone outside the payment integrity industry.

The most important linkable dataset by a wide margin is the NPPES provider registry, maintained by CMS and publicly downloadable in full from [download.cms.gov](http://download.cms.gov). NPPES contains registration records for every provider with a National Provider Identifier in the United States, which as of 2025 is over 8.6 million records. The fields that matter most for fraud detection are not the obvious ones. The NPI itself is useful as a join key. The authorized official name and contact information for organizational providers is more useful, because it lets you identify whether the same individual appears as the authorized official across multiple NPI registrations, a pattern that shows up consistently in bust-out schemes where operators open successive LLCs under their own name or under family members. The entity formation date is arguably the single most valuable field in the whole registry for fraud purposes, because the new-entity-plus-rapid-billing-escalation pattern is so reliably predictive of fraud in the behavioral health and home care taxonomies that it functions almost as a rule rather than a signal. An LLC that did not exist eighteen months ago and is now billing Medicaid at the 95th percentile for its taxonomy in its state has a prior probability of fraud that dwarfs almost any other indicator.

The OIG exclusion list is the second essential join. The Department of Health and Human Services Office of Inspector General maintains a publicly downloadable list of individuals and entities that have been excluded from participation in federal healthcare programs, typically as a result of fraud convictions, license revocations, or other misconduct. The list is searchable and downloadable at [oig.hhs.gov](http://oig.hhs.gov) and is updated monthly. The most common fraud pattern it enables detection of is NPI laundering: an individual who was personally excluded from Medicaid billing opens a new organizational entity under a spouse, parent, or business partner’s name, obtains a new organizational NPI, and resumes billing under the new entity. The exclusion list alone cannot catch this pattern because it tracks individuals by name and Social Security number, not by organizational affiliation, but when joined against NPPES authorized official data it becomes significantly more powerful. An entity whose authorized official shares a surname and address with an excluded individual is not proof of fraud but it is a screening flag worth following up on.

CMS Open Payments, the dataset created by the Physician Payments Sunshine Act, tracks financial relationships between pharmaceutical manufacturers, medical device companies, and healthcare providers. At first glance this seems unrelated to Medicaid home health fraud. In practice it is useful for a specific subset of fraud patterns involving referral schemes, particularly in durable medical equipment, specialty pharmacy, and behavioral health service lines where manufacturers pay providers for referrals or consulting arrangements that serve as kickback vehicles. A home health agency operator who appears as a recipient of significant Open Payments transfers from a DME supplier while simultaneously billing Medicaid for home health services at anomalous rates has a fact pattern worth examining even if neither data point is conclusive alone.

PECOS, the Medicare Provider Enrollment Chain and Ownership System, is CMS’s enrollment database for Medicare participation and is publicly available through [data.cms.gov](http://data.cms.gov). It contains provider enrollment dates, practice location histories, specialty classifications, and reassignment of benefits relationships that show which individual practitioners have routed their billing through which organizational entities. The reassignment data is particularly valuable for detecting the organizational shell game that sophisticated fraud operators play. A physical therapist who reassigns their Medicare billing through six different LLCs over a five-year period, each of which also happens to have been a high Medicaid biller during its existence, is a pattern that PECOS makes visible in a way that the Medicaid spending data alone cannot.

[SAM.gov](http://SAM.gov), the federal System for Award Management, maintains a database of entities that have been debarred or suspended from federal contracting and program participation. It overlaps partially with the OIG exclusion list but covers a broader range of federal programs and includes some individuals and entities that appear on one list but not the other. For Medicaid fraud purposes it is most useful as a supplementary screen rather than a primary signal, but the crosswalk between SAM exclusions and active Medicaid billers has historically surfaced cases that the OIG list missed.

State corporate registry data is technically fifty-one separate datasets rather than one, but the major states all publish searchable corporate registration records that identify when an LLC was formed, who its registered agent and members are, and whether it is currently in good standing. Several states have made this data downloadable in bulk. For the states that haven’t, it is often scrapeable through public records requests or commercial data vendors who aggregate it. The corporate registry data is the piece of the puzzle that closes the entity relationship graph. When you know from NPPES that a Medicaid billing entity was formed in 2022 and when you know from the state corporate registry that its registered agent is the same person who was the registered agent for three other LLCs that billed Medicaid heavily between 2018 and 2021 and then dissolved, you have a fact pattern that is not just anomalous but narratively coherent as a fraud scheme.

Census TIGER geographic data and HUD housing datasets round out the stack in a less obvious but analytically important way. A provider claiming to deliver home health services at a residential address that appears in HUD data as a vacant lot, a commercial property, or a federally subsidized housing unit with no registered healthcare operations is a geographic implausibility check that costs nothing to run and surfaces a meaningful share of phantom billing fraud. The van in rural New Mexico billing 1,006 claims per workday would have failed a basic geographic plausibility screen if the claimed service delivery addresses had been cross-referenced against population density and dwelling unit data for that provider’s service area.

Home Health: A Perfect Fraud Ecosystem

Understanding why home health dominates both the legitimate Medicaid spending taxonomy and the fraud pattern taxonomy requires understanding what home health actually is as a service category and what the structural incentives look like from every angle. The short answer is that home health is a perfect fraud ecosystem not because fraudsters are unusually clever but because the program was designed in a way that makes fraud the path of least resistance and legitimate oversight nearly impossible at scale.

The core problem is that home health services, including personal care, attendant care, and home-based behavioral health support, are fundamentally unverifiable using any of the mechanisms that work for other healthcare claim types. When a physician bills Medicaid for a surgery, there is an operative note, an anesthesia record, a facility record, a post-operative nursing note, and typically imaging or pathology results that collectively make it nearly impossible to bill for a surgery that did not happen. When a pharmacy bills Medicaid for a prescription, there is a dispensing record, a prescriber NPI, a patient signature or delivery confirmation, and a drug supply chain that leaves multiple independent verification points. When a home health aide bills Medicaid for four hours of personal care services delivered to a beneficiary in their home, there is a caregiver attestation, a supervisory visit note that in most states is required only quarterly, and perhaps an Electronic Visit Verification timestamp if the state has implemented EVV well. That’s it. The service itself is a human interaction in a private residence that leaves no independent evidence of having occurred.

Electronic Visit Verification, mandated by the 21st Century Cures Act and required in all states by 2020 for personal care services and by 2023 for home health services, was supposed to address exactly this problem. EVV systems require caregivers to check in and out of visits electronically, typically via a smartphone app or telephonic system, creating a timestamp and often a GPS location record. The implementation reality has been deeply uneven. States had significant flexibility in how they implemented EVV, which vendors they chose, and how rigorously they enforced compliance. In some states EVV data is submitted to managed care organizations who are supposed to match it against claims before payment. In practice the matching is often done retrospectively after payment has already been made, which converts EVV from a pre-payment fraud prevention tool into a post-payment audit trigger. In other states EVV compliance is treated as a documentation requirement that generates a corrective action plan when violated rather than a payment denial, which means a provider who never submits EVV data faces a compliance letter rather than a clawback. The DOGE Medicaid spending dataset does not include EVV data, which means analysts working from the public data cannot directly assess whether billed visits have EVV confirmation. That gap is significant.

The caregiver workforce dynamics of home health create additional fraud vulnerability that is structural rather than incidental. The industry is characterized by high turnover, low wages, part-time employment, and significant use of gig-style labor arrangements. A home health agency operator who wants to commit billing fraud has two basic choices: bill for services that were partially delivered at a higher unit count than actually occurred, or bill for services that were never delivered at all using real caregiver names and real beneficiary names without any actual service taking place. Both schemes are enabled by the labor structure of the industry. In the partial delivery scheme, caregivers who show up for two hours get billed as four, with the caregiver sometimes complicit and sometimes unaware that their attestation is being inflated by the agency billing department. In the phantom billing scheme, the operator may employ a handful of actual caregivers to create a legitimate-looking operation while billing far beyond what that workforce could physically deliver. The physical impossibility threshold that flags the van in New Mexico at 1,006 claims per workday is the extreme version of this pattern. The more common version is an agency with twelve W-2 employees billing for a volume of visits that would require thirty-five workers, which looks unusual in a staffing audit but does not surface in billing data alone without workforce size as a denominator.

The Medicaid managed care structure, which now accounts for roughly two-thirds of total Medicaid spending nationally, creates a diffusion of accountability that home health fraud operators have exploited systematically. In a fee-for-service Medicaid world, the state Medicaid agency pays the provider directly and receives the claim directly, creating at least the theoretical possibility of state-level anomaly detection. In managed care, the state pays a per-member-per-month capitation to the MCO, the MCO pays the provider, and the MCO is nominally responsible for fraud detection within its network. In practice MCO fraud detection programs vary enormously in sophistication and most are focused on the highest-dollar, most obvious patterns because the economics of payment integrity under capitation are different from fee-for-service. Under capitation the MCO absorbs the cost of fraud from its medical loss ratio unless it can recover from the provider, which creates incentive to detect and claw back fraud but also creates incentive to keep enrollment high and networks broad in ways that can compete with rigorous credentialing. The result is a system where the entity theoretically responsible for catching home health fraud is also the entity that signed the contract with the home health agency and has commercial relationships with its operators.

Federal matching rate economics complete the fraud-enabling picture. Because the federal government pays between fifty and ninety cents of every Medicaid dollar depending on the state’s Federal Medical Assistance Percentage, states bear only a fraction of the cost of fraudulent payments made within their programs. A state with a 70-30 federal-state matching ratio that allows $100M in fraudulent home health billing to persist only loses $30M from its own budget. The federal taxpayer absorbs the rest. This creates a documented pattern where states with higher federal matching percentages have historically had weaker fraud detection infrastructure, not because state officials are corrupt but because the incentive structure makes aggressive fraud enforcement a worse financial proposition than it appears from the outside. Fixing this requires either changing the matching rate structure in ways that give states stronger financial stake in their own program integrity, or creating federal detection infrastructure that operates independently of state incentives. The new public dataset is a meaningful step toward the latter.

The Entrepreneur’s Playbook for Monetizing Open Source Fraud Detection

The business opportunity created by the Medicaid spending dataset combined with the other public datasets described above is real, meaningful in scale, and genuinely underserved. But it is not a simple data product play and anyone who approaches it that way will discover quickly that the incumbents are better positioned to sell pure analytics than a startup, and that the government procurement cycle for pure SaaS is measured in years rather than quarters. The entrepreneur opportunity is a hybrid model that uses open source data analytics as the top-of-funnel signal generator and wraps it with investigative services, legal partnerships, and outcome-based revenue structures that align incentives in ways the incumbents structurally cannot.

The starting point is the data infrastructure itself. Building a pipeline that ingests the Medicaid spending dataset, joins it against NPPES, the OIG exclusion list, PECOS, Open Payments, [SAM.gov](http://SAM.gov), and state corporate registry data, and runs it through a set of anomaly detection models is probably a two-to-four person engineering effort over three to six months to get to a production-grade system. The models at this stage do not need to be exotic. The highest-signal fraud indicators in this dataset are detectable with relatively straightforward approaches: new entity formation date joined to billing ramp rate, authorized official network graphs identifying shared principals across multiple NPIs, geographic implausibility screens comparing claimed service delivery locations against population and housing data, procedure code billing concentration analysis that flags providers in the top one percent of their taxonomy and state, and temporal pattern analysis identifying the ramp-and-exit signature of bust-out schemes. None of these require a large language model or a deep learning architecture. They require clean data joins and thoughtful feature engineering. The competitive moat at this stage is not the algorithm but the data assembly and the domain expertise to know which features actually predict fraud versus which features predict legitimate high-volume providers.

The output of that system is a prioritized list of provider NPIs that warrant investigation, ranked by anomaly score and annotated with the specific signals that triggered the flag. That output is not a fraud finding. It is a fraud candidate list. The distinction matters legally, commercially, and ethically. Publishing a ranked list of suspicious providers without human investigation and expert review is how startups generate defamation lawsuits and regulatory backlash. The data layer generates leads. Humans close them.

The investigative services layer is where the business model gets interesting. There are several revenue structures worth considering and a sophisticated entrepreneur should probably pursue multiple simultaneously while the market develops. The first is direct sales to state Medicaid agencies and managed care organizations. State Medicaid directors and MCO medical directors are perpetually under pressure to demonstrate fraud recovery outcomes, and a vendor that can deliver a prioritized investigation queue derived from systematic cross-dataset analysis is solving a real operational problem. The sale is slow, typically twelve to twenty-four months from first contact to signed contract, and pricing is either a flat annual SaaS fee or a percentage of documented recoveries. The percentage-of-recovery model is more palatable to government procurement because it converts a budget line item into a contingency, but it requires the vendor to have the working capital to operate through a long investigation-to-payment cycle before revenue materializes.

The qui tam legal referral model is a second revenue stream that can operate in parallel with the government sales channel and does not require winning procurement contracts. The False Claims Act allows private citizens with original information about fraud against the federal government to file suit on the government’s behalf as a qui tam relator, receiving fifteen to thirty percent of recovered funds if the government intervenes and the case succeeds. The HCFAC enforcement program returns $2.80 for every dollar spent on prosecution. A company that systematically identifies high-confidence fraud cases from open source data, builds an evidentiary package around each case through investigative fieldwork, and refers those packages to healthcare fraud plaintiff firms on a fee-sharing arrangement is essentially a case origination engine for qui tam litigation. The public disclosure bar in the FCA, which limits suits based primarily on publicly available information, requires careful navigation, but a company that combines public data analysis with original fieldwork including caregiver interviews, site visits, and beneficiary contact creates original information that can survive that bar. The legal partnership structure works because plaintiff firms have the litigation infrastructure but are bottlenecked on case origination. A systematic data-driven lead generator that pre-screens cases for strength is genuinely valuable to them.

The investigative fieldwork component is not optional in this model. It is the part that transforms a billing anomaly into an actionable legal referral and it is the part that is hardest to automate. Confirming that a home health agency is billing for phantom visits requires actually attempting to reach the beneficiaries supposedly being served, visiting the claimed service delivery addresses, interviewing former employees, and reviewing whatever documentation the agency has filed with state licensing authorities. That is labor-intensive work that requires people with investigative skills, healthcare program knowledge, and the ability to conduct interviews in community settings. The staffing model for this layer looks more like a private investigative firm than a software company, which is why most pure-tech founders underestimate it and most incumbents in the payment integrity space who have the investigative capacity lack the technical infrastructure to generate leads systematically.

The third revenue layer is market intelligence and compliance services sold to legitimate industry participants. Home health agencies that are operating legally have significant interest in understanding where their billing patterns fall relative to peer providers, both because anomalous-looking billing can trigger audits even for legitimate operators and because understanding the competitive billing landscape helps with rate negotiation and contract management. An analytics platform that gives compliant home health agencies visibility into how their billing looks from a fraud-screening perspective, and helps them maintain documentation practices that will survive scrutiny, is a defensible SaaS product with recurring revenue and a buyer who is motivated by risk management rather than government procurement. The same data infrastructure that generates fraud leads for enforcement purposes generates peer benchmarking and compliance tools for legitimate operators.

The flywheel that makes this business compound over time is labeled outcome data. Every investigation that results in a confirmed fraud finding, a successful qui tam recovery, or a state Medicaid audit action creates a labeled data point that improves the model’s ability to distinguish true fraud from billing anomalies with innocent explanations. A company that has been running this system for three years has a labeled dataset of confirmed fraud patterns that no one else can replicate from the public data alone, because the public data tells you who billed anomalously but not which of those anomalies turned out to be fraud. That labeled dataset is the actual moat, and it accumulates automatically as the investigative operation generates outcomes. The company that runs this flywheel fastest in a specific taxonomy, whether that is home health, behavioral health, or DME, builds a compounding advantage in that space that is genuinely hard for either incumbents or new entrants to replicate quickly.

The realistic near-term revenue model for a company in this space, if executed well, is a combination of three to five state Medicaid agency analytics contracts in the $500K to $2M annual range, a pipeline of qui tam referrals generating contingency fees over a two to four year litigation cycle, and a compliance SaaS product for legitimate home health operators generating $5K to $20K annually per customer with relatively low churn. None of those revenue streams is enormous in year one. All of them are growing, all of them are somewhat defensible, and the combination of a government contract channel with a litigation referral channel with a commercial SaaS channel creates the kind of revenue diversification that makes the business survivable through the slow procurement cycles that characterize this market.

The political tailwind from the current administration’s focus on Medicaid fraud is real but should not be over-indexed. DOGE’s involvement in the data release creates short-term attention that benefits companies in this space by elevating the problem in the minds of state Medicaid directors, MCO executives, and Congressional oversight staff. That attention is useful for accelerating sales conversations that would otherwise take longer to initiate. But the underlying fraud problem predates this administration by decades and will persist through multiple political cycles. The companies that will win in this space are the ones building durable detection infrastructure rather than chasing the current news cycle, because the news cycle moves on and the structural fraud vulnerability in Medicaid home health does not.​​​​​​​​​​​​​​​​

What Actually Dropped Today (and Why It Matters)

The Dataset Itself: What’s in It, What’s Missing, What’s Useful

The Problem Space in Numbers

The Incumbent Landscape and Its Structural Weaknesses

Where the Venture Opportunity Actually Lives

Watch-Outs, Political Risk, and Things That Could Go Wrong

So What Do You Actually Do With This?

Abstract

- On February 13, 2026, DOGE’s HHS team published what it’s calling the largest Medicaid claims dataset ever made publicly available, accessible at [opendata.hhs.gov](http://opendata.hhs.gov)

- Medicaid total program spend: ~$849B (federal + state) in 2023, serving ~90M enrollees

- Medicaid improper payments: estimated $31.1B in FY2024 (5.09% improper payment rate per CMS), with some estimates going 3-4x higher when eligibility errors are included

- GAO estimated $100B+ in combined Medicare/Medicaid improper payments in FY2023

- The underlying T-MSIS (Transformed Medicaid Statistical Information System) data covers 4 claims file types: inpatient, long-term care, other, and prescription

- DOJ’s 2025 healthcare fraud takedown charged 324 defendants for $14.6B in alleged fraud

- Digital health VC hit $12.3B in 2025 (PitchBook Q3 annualized), with payment integrity specifically getting PE consolidation attention

- This essay argues the public release is a genuine inflection point for a cluster of health tech use cases, and tries to map where the real build-vs-buy-vs-partner opportunities exist

-----

What Actually Dropped Today (and Why It Matters)

Medicaid data has historically been one of the most fragmented, hardest-to-access, least standardized bodies of administrative information in all of American healthcare. Fifty-one state programs, each with their own eligibility rules, managed care contract structures, fee schedules, and reporting formats. CMS has been collecting T-MSIS data from states since the mid-2010s, and even the research-accessible version (the TAF, or T-MSIS Analytic Files) required a CMS Privacy Board approval process and a data use agreement that could take the better part of a year for academic researchers, to say nothing of what it meant for commercial operators. The de-identified, aggregated public version that existed before today was useful for high-level trend analysis and not much else.

What DOGE’s HHS team dropped today is a different animal. HHS spokesperson Andrew Nixon described it as the first time the department is “expanding public access to de-identified, aggregated data to increase transparency and accountability beyond what is currently available.” DOGE itself announced this as the largest Medicaid claims dataset in department history. The context for the release is overtly political - the same announcement mentioned that the tool “could have helped detect large-scale autism diagnosis fraud in Minnesota,” a reference to a scandal that’s been a DOGE talking point for months - but the underlying data is real, it’s publicly downloadable, and it covers provider-level spending patterns across Medicaid in a way that hasn’t been available before.

For investors and builders, the politics are a sideshow. The data is the story.

To understand why, you need a quick refresher on how Medicaid data has worked historically. States submit claims data to CMS through the T-MSIS system, which includes four claims file types covering inpatient, long-term care, other services, and prescription drugs, plus a financial transactions file and beneficiary eligibility data. CMS runs over 6,000 data quality checks on state submissions. The resulting dataset is massive, with all 50 states (and DC, Guam, Puerto Rico, and the Virgin Islands) reporting. But the version accessible to researchers and policy shops has always required institutional approval, been restricted to academic use, lagged 18-24 months behind real time, and been structured in a way that made commercial product development extremely cumbersome. The new open data portal cuts a layer of that friction.

The specific dataset published today focuses on provider-level spending patterns - essentially, what individual Medicaid providers are billing, in aggregate, across states. That’s the table stakes version of the data that fraud investigators have always wanted but haven’t had easy public access to. The implications of that shift are not small.

The Dataset Itself: What’s in It, What’s Missing, What’s Useful

Before anyone gets too excited and starts building product roadmaps on a Friday afternoon, it’s worth being specific about what the data actually is and isn’t.

On the positive side: provider-level Medicaid spending aggregated across states in a publicly downloadable format is genuinely novel. The T-MSIS underlying source data is more granular than anything previously available through public channels. For the first time, someone with a laptop and some Python can start asking questions like which NPI numbers are in the top billing percentiles for specific procedure codes in specific states, how provider spending patterns compare across Medicaid programs with different managed care penetration rates, and where outliers cluster by geography, specialty, and service category. That’s a meaningful starting point for anomaly detection, peer benchmarking, network adequacy analysis, and a bunch of other downstream applications.

On the limitations side: a few things to keep in mind. First, this is de-identified and aggregated - you’re not getting beneficiary-level claims files here, which means use cases that require patient-level longitudinal analysis still require the full T-MSIS TAF research files and the DUA process. Second, the data quality across states varies significantly. CMS’s own OBA (Outcomes Based Assessment) framework applies more than 600 high and critical priority data quality checks, and state performance on those checks is publicly tracked. Some states have meaningful data completeness and accuracy gaps, particularly in managed care encounter data, which in many states represents the majority of Medicaid spending. Third, the data still lags real-time operations by design - this is administrative claims data, and the pipeline from state submission to public availability introduces delay. Fourth, and this is subtle but important: about two-thirds of Medicaid spending nationally now flows through managed care organizations (MCOs), and the quality of encounter data submitted by MCOs varies dramatically. A state where MCO encounter data is sparse or incomplete will show misleadingly low apparent provider spending.

None of those limitations make the dataset not useful. They make it a starting point rather than a finished product. The people who will extract the most value from it are the ones who understand those limitations deeply enough to account for them in product design, and who can combine the public data with proprietary signals to fill the gaps.

The most immediately useful applications of the new dataset in roughly decreasing order of data quality dependency: provider outlier identification for fraud investigation (high data quality needed, but lots of signal in the aggregate even with gaps), peer benchmarking for state Medicaid agencies and MCOs (moderate quality needs, relative comparisons are more robust than absolute ones), network adequacy and access analytics for managed care plans (lower quality sensitivity, directionally useful even with gaps), policy analysis and advocacy (the broadest and most forgiving use case, where trends matter more than precision), and market intelligence for health tech vendors trying to understand where Medicaid spending concentrates by geography and specialty.

The Problem Space in Numbers

Let’s ground this in why the problem is interesting enough to build for and invest in.

Medicaid is the second-largest health program by expenditure in the US. Total federal and state Medicaid spending hit approximately $849B in 2023 for roughly 90 million enrollees. That number has been growing faster than CMS projected throughout the 2020s, driven by a combination of the ACA expansion, pandemic-era continuous enrollment policies, increased per-enrollee spending, higher drug costs, greater use of long-term services and supports, and state-directed payment mechanisms. Even as COVID-era enrollment unwinding reduced beneficiary counts modestly in 2024 and 2025, per-enrollee spending kept climbing.

Improper payments are where the product opportunity crystallizes. CMS’s official estimate for Medicaid improper payments in FY2024 was $31.1B, representing a 5.09% improper payment rate. That’s the number HHS puts in its annual financial report. It’s also probably conservative. The Payment Error Rate Measurement (PERM) program that CMS uses to calculate that figure has historically excluded eligibility errors from its calculation, and when researchers at the Paragon Institute applied a broader methodology that included eligibility verification, the estimated cumulative improper Medicaid payments from 2015-2024 came out around $1.1 trillion, roughly double GAO’s $543B estimate for the same period. The debate about the right methodology is real and ongoing, but even at the conservative official number, $31.1B a year in improper payments on an $849B program is a massive problem with an obvious technology component.

Beyond pure fraud, there are structural inefficiencies that are arguably bigger in aggregate. The $51B in Medicare improper payments and $50B in Medicaid improper payments that GAO flagged for FY2023 are just the identifiable surface. The DOJ’s 2025 National Health Care Fraud Takedown, the largest in US history, resulted in criminal charges against 324 defendants for schemes involving over $14.6B in intended loss. Operation Gold Rush alone, targeting a transnational criminal organization running multi-billion-dollar Medicare schemes, illustrated how sophisticated and organized the fraud ecosystem has become.

State-to-state variation is another piece of the picture that the new dataset should illuminate. KFF’s analysis of T-MSIS data shows that Medicaid spending per enrollee varies by a factor of more than 3x across states, even controlling for eligibility group. An aged, blind, and disabled enrollee in New York costs about $30,000/year in Medicaid, while a similar enrollee in Texas costs closer to $12,000. Some of that reflects legitimate differences in benefit design, provider rates, and local cost-of-care. Some of it reflects differences in managed care penetration, administrative rigor, and fraud control effectiveness. A dataset that lets analysts start decomposing those differences at the provider level is genuinely useful for policymakers, MCO executives, and any vendor trying to understand where they can add value in specific state markets.

The specific Minnesota autism fraud case that DOGE keeps referencing is worth understanding on its own terms as a case study. The allegation, which the US Attorney for Minnesota flagged as potentially involving upwards of $9B in losses over recent years, centered on providers paying parents to have children diagnosed with autism specifically to maximize Medicaid billing. The scheme was allegedly detectable through billing pattern anomalies that would have been visible in provider-level claims data had that data been more readily available and analyzed. Whether or not DOGE’s characterization of that case is accurate in its specifics, the structural vulnerability it illustrates is real: when a provider-level billing outlier can persist for years across a multi-billion-dollar program because no one has easy visibility into cross-state spending patterns, that’s a data availability problem as much as it’s a human oversight problem.

The Incumbent Landscape and Its Structural Weaknesses

The healthcare fraud and waste detection market has been around for decades, which is part of the problem. The current ecosystem is dominated by legacy players whose products were built in a different era of data availability, computing capability, and healthcare program structure.

Cotiviti, Optum (through its Clinical Intelligence and payment integrity units), Change Healthcare (now part of UnitedHealth’s Optum stack), and a collection of smaller regional firms like Conduent and NCI Information Systems have historically held most of the large Medicaid and Medicare payment integrity contracts. These are not small businesses - Cotiviti processes billions of dollars in claims reviews annually, and Optum’s analytics business generates billions in revenue. But their core products were designed around rule-based retrospective auditing: you build a library of billing code combinations that are known fraud patterns, you run claims through those rules after payment, you identify recoveries, and you take a percentage of what you recover. It’s a useful capability and it will always have a role, but its limitations are well understood.

Rule-based systems are inherently reactive. They can only catch fraud patterns that someone has already documented and codified. A sophisticated operator who stays just outside the known rule violations can extract enormous value for years without triggering anything. The autism case in Minnesota is a good example of exactly this failure mode. The billing patterns were anomalous in a way that would have been obvious to any unsupervised learning model looking for outliers - providers with rapid enrollment growth, high per-beneficiary billing, claims concentrated in procedure codes with weak documentation requirements, unusual geographic clustering - but those aren’t the kinds of patterns rule-based systems are designed to catch proactively.

The other structural weakness of the incumbent landscape is its dependence on proprietary data silos. A firm like Cotiviti gets its advantage partly from having accumulated benchmarks across many client programs - essentially, a private claims database that lets them flag statistical outliers. But those benchmarks are siloed within their client relationships and are not refreshed at anything close to real-time. The T-MSIS data that was available to them was subject to the same access constraints and data quality issues that affected everyone else.

In 2025, New Mountain Capital made a significant PE bet on the consolidation thesis, acquiring Machinify and combining its AI capabilities with legacy payment integrity providers including Apixio, Varis, and The Rawlings Group into a combined entity. The bet is essentially that combining established distribution (relationships with MCOs and state Medicaid agencies) with modern ML infrastructure creates a more defensible business than either could build alone. That’s probably right as far as it goes, and it signals that smart money sees the space as ripe for AI-native disruption, even if the PE approach is more roll-up than greenfield innovation.

The gap being created here, and where the venture-scale opportunity lives, is in the combination of three things that the incumbents structurally can’t provide: real-time or near-real-time anomaly detection using modern ML on the new public data combined with proprietary signals, workflow tools that actually integrate into state Medicaid agency and MCO operating environments in a way that creates stickiness, and the ability to move quickly as new fraud patterns emerge rather than waiting for a rules library to be updated.

Where the Venture Opportunity Actually Lives

The new dataset is useful, but it’s a raw ingredient. The venture-scale businesses get built on top of infrastructure that transforms that ingredient into defensible, recurring products. Here’s an honest map of where the opportunity clusters.

The most direct application is Medicaid-specific payment integrity SaaS. The market has historically been dominated by the incumbents described above, operating under contingency-fee audit contracts (typically 10-30% of recovered dollars) or flat-fee managed care subcontracts. The contingency-fee model creates weird incentives: auditors go after the lowest-hanging, easiest-to-document patterns because the economics of chasing sophisticated fraud are worse. A SaaS model that charges state Medicaid agencies or MCOs a platform fee for continuous monitoring rather than a percentage of recoveries aligns incentives better and creates more predictable revenue. The challenge is that state Medicaid agency procurement is slow, relationships matter a lot, and the switching costs from incumbents are real. But the market is large enough that even a narrow wedge is interesting - a state like California with $140B+ in annual Medicaid spending can justify paying real money for a meaningfully better tool.

The second cluster is what you might call Medicaid market intelligence for health tech vendors and health plans. This one is underappreciated. Any company selling technology or services into the Medicaid ecosystem (and that’s a lot of companies - EHR vendors, RCM companies, managed care tech platforms, pharmacy benefit managers, home health operators) needs to understand how Medicaid spending is distributed across states, specialties, and service categories to allocate their sales resources intelligently. The public dataset creates a foundation for commercial intelligence products that are analogous to what companies like Symphony Health or IQVIA have built for the pharmaceutical industry - essentially, turning claims data into market maps that help operators make better go-to-market decisions. This is a faster-moving, lower-regulatory-barrier opportunity than the payment integrity space.

Network adequacy analytics is a third use case that doesn’t get enough attention. Federal managed care regulations require MCOs with Medicaid contracts to demonstrate that they have sufficient providers in their networks to serve enrollees within time and distance standards. Compliance with those requirements is monitored by state Medicaid agencies, but the monitoring tools have historically been inadequate. Provider-level Medicaid spending data, combined with geographic and specialty information, creates a much richer basis for network adequacy modeling than what’s existed before. This is interesting not just as a compliance tool but as a strategic asset for MCOs figuring out where to invest in network development.

The fourth area is genuinely underserved and potentially the largest: beneficiary navigation and enrollment integrity. About 77 million people were enrolled in Medicaid and CHIP as of September 2025. The “unwinding” of pandemic-era continuous enrollment policies in 2023 and 2024 resulted in millions of eligible people losing coverage due to administrative failures rather than actual ineligibility changes. Meanwhile, the new OBBBA budget reconciliation package signed on July 4, 2025 introduced Medicaid work requirements and other eligibility changes that will create new administrative complexity at scale. The combination of more complex eligibility rules, more frequent redeterminations, and a data asset that can help identify where eligibility mismatches are occurring is a real product opportunity. Fortuna, a YC-backed startup, is going after part of this with an end-to-end Medicaid navigation platform - the concept is right even if the execution is early.

The fifth cluster is public health and policy analytics, which is more consulting-adjacent than pure SaaS but shouldn’t be dismissed. State health departments, advocacy organizations, academic medical centers, and a growing number of large health systems want analytical capabilities that let them work with Medicaid data at scale. The new public dataset, combined with other open data assets (Census, HRSA, AHRQ, OpenPayments), creates a rich substrate for these kinds of analyses. A platform that makes it easy for less technical users to query and visualize this data - think Palantir-lite for Medicaid analysts - has a real market, even if the buyer profile (state agencies, health systems, foundations) makes it a slower sale than a purely commercial customer.

Watch-Outs, Political Risk, and Things That Could Go Wrong

None of this is risk-free, and some of the risks are non-obvious, so they’re worth naming directly.

The biggest single risk is political instability of the data asset itself. This dataset was released by DOGE’s HHS team as part of a politically motivated narrative about fraud in Medicaid. Administrations change, political priorities shift, and data that was made public can be made less accessible, less maintained, or simply deprecated in a future administration. The T-MSIS program itself depends on continued CMS funding and state cooperation. DOGE has simultaneously been cutting CMS staff and budget in other areas, which creates a tension: the agency being asked to publish and maintain better data is also the agency being stripped of the capacity to do it well. Founders building product on this dataset need a data diversification strategy that doesn’t make them entirely dependent on continued government openness.

The data quality issue is both a risk and an opportunity, but it tilts toward risk at the product design level. The T-MSIS OBA framework tracks state performance on over 600 high and critical data quality checks, and a meaningful number of states have persistent gaps, particularly in managed care encounter data. A product that uses the public dataset without explicitly accounting for state-level data quality variation will produce systematically biased outputs. In managed care states with poor encounter data submission - and there are several large ones - the spending patterns visible in the dataset may reflect the quality of reporting infrastructure as much as they reflect actual care delivery patterns. Any ML model trained on the raw data without state-level quality weights will find the quality artifacts as reliably as it finds the fraud signals.

Privacy risk is real even with de-identified, aggregated data. There’s a well-documented academic literature on re-identification attacks on aggregated healthcare data, particularly when geographic and demographic cut-points are fine enough. HIPAA’s de-identification standards were written in a different era of data availability, and the combination of the new Medicaid dataset with other publicly available data sources (OpenPayments, NPPES, provider directories, social media) creates re-identification vectors that weren’t contemplated when the data was designed. Vendors building on this data need to be thoughtful about what they aggregate and at what granularity, not just because of legal risk but because a high-profile re-identification incident would generate regulatory blowback that could constrain the entire space.

The incumbent response is also worth modeling. Cotiviti and Optum are not going to ignore a new public data asset that could theoretically commoditize part of their business. They have the relationships with state Medicaid agencies, the existing contracts, and the resources to build or acquire the ML capabilities to compete on the new data. The venture window here is probably 18-36 months before the large incumbents have incorporated the public dataset into their products in a way that narrows the differentiation available to startups. That’s enough time to build and scale something interesting, but it’s not infinite, and the sales cycle for selling into state Medicaid agencies is long enough that the clock matters.

Finally, there’s the OBBBA policy risk. The reconciliation package Trump signed on July 4, 2025 cuts nearly $1 trillion from Medicaid over the next decade, introduces work requirements for expansion enrollees, restricts state-directed payment mechanisms, and limits provider tax financing arrangements that many states use to increase their federal matching dollars. MedPAC and CBO both projected the bill will reduce Medicaid enrollment by more than 10 million people. A significantly smaller Medicaid program means a smaller total addressable market for Medicaid-focused products, and the states most affected by the cuts are also the states that often have the most interesting data and the most sophisticated MCO infrastructure. This is a macro headwind for the space that any investor or founder needs to think through carefully - the opportunity is real, but it’s operating inside a program that is under genuine political pressure to shrink.

So What Do You Actually Do With This?

For investors with existing portfolio exposure to payment integrity, RCM, or healthcare data infrastructure, the dataset release is a forcing function to re-underwrite competitive positioning. Any company in those categories whose defensibility depends on proprietary access to data that just became significantly more available has a potential moat problem. The conversation to have with those portfolio companies this week is not “how do you use the new data” but “how does this change the competitive landscape for your core product.”

For investors evaluating new opportunities, the dataset creates a few categories worth accelerating diligence on. Medicaid-specific AI/ML payment integrity platforms with real state agency relationships, strong data quality handling, and defensible IP beyond the public data are worth a serious look, particularly if they’ve been quietly building for a few years and have production deployments. Be skeptical of pitches that lean heavily on “we have access to the new public Medicaid data” as a core differentiator - everyone has that access now, and the moat has to be in what you do with it. The real moat candidates are companies with proprietary labeled datasets of confirmed fraud outcomes (because that’s what lets you train a discriminative model that’s actually useful), and companies with deep workflow integration into state or MCO operating environments.

For founders thinking about the space, the product insight that matters is that the public dataset is most useful as a seed layer for building something proprietary. The pattern is roughly: start with the public T-MSIS spending data to build a baseline anomaly detection model, use that model to generate leads that get investigated and confirmed (or not), turn those confirmed outcomes into labeled training data, train a better model on the labeled data, and use that better model to surface better leads. The value accumulates in the labeled dataset, not in the public data itself. That’s the flywheel, and the company that runs it fastest and most rigorously in a specific segment (say, behavioral health billing fraud in managed care, or durable medical equipment patterns in specific geographies) will build a compounding advantage that’s genuinely hard to replicate.

The broader point is that publicly available Medicaid data at provider-level resolution is a fundamentally new capability for the health tech ecosystem, even with all its caveats. The closest analog is probably the Open Payments (Sunshine Act) database that CMS has published since 2014, which tracks payments from pharmaceutical and medical device manufacturers to physicians and teaching hospitals. That data release generated an entire ecosystem of compliance tools, investigative journalism, and policy analysis that didn’t exist before it. The Medicaid provider spending dataset is potentially more impactful because the underlying program is larger, the fraud problem is more acute, and the computational tools available for making sense of it are dramatically better than they were in 2014.

The question for anyone in this space is not whether the data matters. It does. The question is whether you have the specific combination of domain expertise, technical capability, state-level relationships, and data quality savvy to build something on top of it that compounds. That bar is higher than the announcement suggests, and that’s probably why the opportunity is real.​​​​​​​​​​​​​​​​

Virtual cards have penetrated healthcare through claim-based payments, but this represents the most conservative and politically constrained insertion point available. The real opportunity lies in payment workflows where clinical regulation is absent, multi-party reconciliation is manual, and treasury sophistication is low. This analysis examines three higher-leverage domains: bundled episode payments with their inherent disaggregation complexity, post-acute and value-based care enablement where cash flow trumps unit economics, and employer-direct arrangements where payers actively want intermediation. The thesis is that virtual cards become genuine infrastructure rather than interchange arbitrage when inserted at points where payment logic itself creates switching costs. Bundled payment conveners, particularly consulting firms that design episode models but don’t control disbursement, represent an underexploited wedge into sticky, high-margin payment orchestration.

Table of Contents

Why Claims Are Actually the Wrong Reference Point

Bundled Payments as Structural Card Territory

The Hayes Opportunity and Convener Economics

Post-Acute: Better Margins, Less Political Heat

Value-Based Care Platforms That Don’t Want to Touch Money

Employer-Direct and the Interchange Blindspot

Patient-Mediated Payments With Embedded Controls

What Actually Creates Moats in Healthcare Payment Rails

Why Claims Are Actually the Wrong Reference Point

Optum runs virtual cards through claim payments because volume justifies the build and providers already expect remittance bundled with payment. The pitch is administrative simplification, which blunts pushback on interchange fees that would otherwise trigger provider revolts. But claims also come with baggage that caps both economics and design flexibility. They’re price-regulated through fee schedules, audited by states and CMS, politically visible when margins look extractive, and deeply embedded in existing clearinghouse and remittance infrastructure that took decades to stabilize. The result is that virtual card penetration in claims looks more like a tax on existing rails than actual innovation in payment architecture.

The better hunting ground is off-claims entirely, where payments are operational instead of clinical, margin opacity is higher, and nobody has entrenched opinions about how money should move. These are environments where cards can encode business logic instead of just replacing ACH. The pattern to look for is fragmented payables with weak treasury operations and tolerance for fees that stay invisible to end buyers. Healthcare has more of this surface area than almost any other sector, but most of it sits outside the claim flow that gets all the attention.

Bundled Payments as Structural Card Territory

Bundled payments get talked about as a Medicare innovation play or risk-sharing mechanism, but they’re actually a structural opportunity for payment infrastructure because disaggregation is inherently painful. A typical episode bundle involves a single lump sum payment to a convener, which then has to split funds across the hospital, multiple physician groups, anesthesia, post-acute facilities, durable medical equipment suppliers, sometimes imaging or lab vendors, and occasionally home health or rehab services. Each downstream participant has different contract terms, different performance metrics, sometimes different risk arrangements. The disaggregation process today is almost entirely manual, driven by spreadsheets and contract PDFs, reconciled weeks or months after the episode closes, and paid out through a mix of ACH transfers and actual paper checks in some markets.

Virtual cards let you collapse that entire operational mess into programmable payment logic. You can time-gate disbursements based on outcomes windows, like holding back surgeon bonuses until ninety-day readmission data clears. You can encode episode-specific rules directly into card controls, so post-acute payments only release after acute discharge documentation hits the system. Remittance metadata flows with the payment instead of requiring separate reconciliation, which matters enormously when you’re splitting a single bundle check across eight different entities with different EINs and different GL codes. And you can apply merchant fees per downstream participant without the convener needing to negotiate or track those costs separately. ACH cannot do any of this without custom middleware that somebody has to build and maintain, which is why most bundled payment programs still run on spreadsheets and monthly reconciliation calls.

The Hayes Opportunity and Convener Economics

Hayes Management Consulting sits in a weird but lucrative position in the bundle ecosystem. They design the clinical pathways, define the gainsharing formulas, build the actuarial models that determine how much risk each party takes, and often manage the ongoing performance reporting. What they typically don’t do is own the actual payment rails. Money flows from the payer to some designated convener, then that convener manually distributes funds based on whatever contract Hayes helped design. There’s a gap between program design and payment execution, and that gap represents uncaptured economics.

A Hayes-anchored payment platform would look like this: Hayes or a joint venture becomes the actual payment orchestrator, not just the consultant. Bundle funds flow into a controlled treasury account that Hayes operates. Downstream distributions happen via virtual cards issued directly to participating providers. Hayes earns their existing program fees for clinical design, but now also captures merchant processing fees on every disbursement, plus potential float value from timing differences between when the payer funds the bundle and when downstream participants get paid. This reframes Hayes from risk consultant to payments infrastructure with embedded clinical logic, which is a completely different business model with completely different margin potential.

Could they own the whole market? No, because bundled payments remain a small fraction of total healthcare spend and the programs themselves are bespoke by design. But they could absolutely dominate their existing client base because bundles have massive switching costs once clinical workflows and contract terms get operationalized. Payment logic becomes tightly coupled to episode design, which means changing payment vendors requires renegotiating the entire bundle structure. This looks more like vertical SaaS economics than fintech scale plays, but gross margins can hit seventy or eighty percent if the platform is the only thing that knows how to execute the payment waterfall correctly.

Post-Acute: Better Margins, Less Political Heat

Post-acute care is where bundled payment theory crashes into operational reality, and it’s also where virtual cards have better product-market fit than anywhere else in the acute episode. Skilled nursing facilities, home health agencies, and rehab providers are fragmented, undercapitalized, running on thin margins, and almost universally bad at revenue cycle management. They get paid slowly, have limited access to working capital, and often lack the treasury infrastructure to optimize payment terms. They’ll accept interchange fees if it means getting paid faster with clear remittance, because cash flow is worth more than rate optimization when you’re operating on fifteen percent margins.

Use cases are straightforward but high-volume: per diem payments to SNFs that participate in bundled episodes, visit-based payments to home health agencies, episode-based disbursements to outpatient rehab facilities. The pain point isn’t complexity of payment logic, it’s speed and clarity. Most post-acute providers are still waiting thirty to sixty days for bundle reconciliation, then another two weeks for ACH settlement, then manually matching remittance advice to their billing system. Virtual cards can cut that entire cycle to same-day settlement with embedded remittance metadata, which is genuinely valuable to providers even after interchange.

This is arguably a better insertion point than acute bundles because acute hospitals have leverage and sophistication. They can push back on interchange, they have treasury teams that optimize payment terms, and they’re politically connected enough to make noise if fees look extractive. Post-acute providers have none of those advantages, which makes them both more willing to accept cards and less able to extract concessions on pricing. The total addressable market is smaller than acute care, but margin realization is much higher and regulatory scrutiny is much lower.

Value-Based Care Platforms That Don’t Want to Touch Money

The value-based care enablement layer is full of companies that are good at clinical coordination or quality reporting but terrible at financial operations. These are care management vendors that help primary care groups hit quality benchmarks, remote patient monitoring platforms that generate savings through hospital avoidance, quality reporting firms that manage HEDIS and Stars submissions, and various flavors of enablement software that coordinate MSSP or direct contracting arrangements. They all have the same structural problem: they receive performance-based revenue that’s contingent on shared savings or quality bonuses, then have to distribute portions of that revenue to downstream provider groups based on contribution models nobody really agrees on.

Most of these vendors have no interest in becoming payment processors, but they end up doing it anyway through manual ACH transfers and reconciliation spreadsheets because nobody else wants to own the problem. Virtual cards let them become the financial hub without building actual treasury infrastructure. Funds flow into the platform, disbursements happen via virtual cards with embedded attribution logic, and the platform earns interchange while staying completely out of regulated claim flows. This is especially powerful in MSSP-adjacent programs where clinical entities want to participate in upside without becoming an ACO themselves, because the enablement vendor can own the money movement without touching clinical decision-making.

The wedge here is that these platforms already own the reporting that determines who gets paid what. Adding payment execution is a natural extension that generates economics without requiring the platform to take on balance sheet risk or regulatory burden. It’s pure margin expansion on an existing customer relationship, which is why the opportunity keeps getting overlooked by fintech companies that want to build horizontal payment rails instead of embedding into vertical workflows.

Employer-Direct and the Interchange Blindspot

Employer-direct contracting is the one corner of healthcare where buyers genuinely do not care about interchange fees, which makes it nearly perfect territory for virtual card insertion. When a self-insured employer contracts directly with a center of excellence for bariatric surgery or orthopedic bundles, the payment workflow involves the employer or their TPA sending a lump sum for the case rate, then some intermediary managing execution and paying downstream providers. Speed and reporting matter, unit cost optimization does not, because the employer is already getting a discount versus network rates and cares way more about predictability than basis points on payment processing.

Virtual cards work in this context because employers want control and visibility, TPAs want to avoid building payments infrastructure, and providers want fast settlement. The employer doesn’t see interchange as a cost because it’s embedded in the program fee, the TPA earns margin by offering a differentiated service, and the provider gets paid faster than they would through traditional claim cycles. Nobody in the transaction has an incentive to optimize away the card, which is completely different from payer-driven models where providers have enough volume to demand better terms.

This is greenfield compared to traditional payer claims because the number of transactions is lower but the margin per transaction is higher, and there’s no political sensitivity around whether payment rails are extractive. Employer-direct is also growing faster than people realize, particularly in surgical bundles and specialty drug management, which means the TAM is expanding while legacy infrastructure is still mostly absent. The challenge is distribution, because you need relationships with benefits consultants and TPAs rather than provider networks, but the economics are cleaner once you’re in.

Patient-Mediated Payments With Embedded Controls

This one is more controversial but potentially higher margin: virtual cards issued to patients for specific healthcare spending with encoded restrictions on what they can buy. Fertility benefits are the clearest example, where employers or insurers provide a fixed subsidy for IVF or egg freezing, and patients need to coordinate payments across multiple providers for procedures, medications, monitoring, and lab work. The traditional model is reimbursement, where patients pay out of pocket and submit receipts, which is administratively painful and creates cash flow problems for patients who don’t have the liquidity to float thousands of dollars.

Virtual cards solve this by giving the patient a spending instrument with embedded controls: it only works at approved fertility clinics and pharmacies, it can have procedure-specific spending limits, it generates automatic documentation for the plan sponsor, and it completely eliminates reimbursement workflows. The same model applies to specialty pharmacy copay assistance, travel and lodging benefits for centers of excellence programs, and various consumer-directed health benefits that employers are trying to offer without building claims infrastructure.

The economics here are interesting because patients don’t push back on interchange, plan sponsors value fraud reduction over cost optimization, and providers are willing to accept cards if it means guaranteed payment instead of chasing patient collections. This is already happening in benefits administration tech, but healthcare-specific players are underutilizing it because they’re stuck thinking about cards as a replacement for claim payments rather than as a patient spending control mechanism. The regulatory question is whether these arrangements trigger any Medicaid or Medicare kickback concerns, which they generally don’t as long as the benefit is structured as a plan feature rather than a provider inducement, but that requires actual legal architecture.

What Actually Creates Moats in Healthcare Payment Rails

The strategic pattern across all these insertion points is that virtual cards work best when payments are multi-party, logic is contractual rather than clinical, cash flow matters more than unit economics, and nobody wants to build the infrastructure themselves. Claims fail most of these tests, which is why Optum’s implementation looks more like interchange arbitrage than genuine platform defensibility. The places where cards actually create moats are where payment logic itself becomes the product, not just the rails.

Bundled payments have this property because the disaggregation rules are specific to each episode design, which means switching payment vendors requires re-implementing business logic that’s tightly coupled to clinical workflows. Post-acute has it because speed and clarity are worth more than rate optimization, which means providers will tolerate interchange to solve cash flow problems. Value-based care enablement has it because the platform already owns the performance reporting that determines payments, so adding disbursement is margin expansion on an existing sticky relationship. Employer-direct has it because buyers don’t care about interchange and want control more than cost reduction. Patient-mediated benefits have it because spending controls are the actual product, not just a payment method.

The common thread is that all of these are environments where somebody already owns the relationship or the data or the contract structure that determines how money should move, but they don’t own the actual money movement. That gap is where card economics hide best, because you’re not competing on price, you’re competing on whether the buyer wants to build the operational complexity themselves. In most cases they don’t, which means the willingness to pay is high even if the absolute dollar volume is lower than claims.

The next question is which actors in these spaces are positioned to own the episode or the program but currently don’t own the treasury function. Bundled payment conveners like Hayes are the obvious example, but there are equivalent players in every vertical: care management platforms in MSSP, benefits administrators in employer-direct, patient navigation companies in centers of excellence programs. The ones that figure out how to embed payment logic into their existing product will capture margin that currently leaks to generic ACH infrastructure or gets lost in manual reconciliation. The ones that don’t will keep running on spreadsheets while somebody else builds the rails underneath them.​​​​​​​​​​​​​​​​

Kidney Contracting Entities (KCEs) in the CMS Kidney Care Choices Model face a simple but brutal economic problem: they bear financial risk for patients progressing to dialysis, but most discover these progressions only after they happen, when intervention is too late and costs have already spiraled. The model’s structure creates asymmetric risk where a single patient starting dialysis can cost a KCE $90,000-120,000 annually in incremental spending against their benchmark, yet the typical nephrology practice identifies high-risk progressors using clinical judgment and static lab snapshots rather than predictive analytics. This gap represents a clear product opportunity for a dialysis progression prediction platform that ingests longitudinal kidney function data, calculates personalized decline trajectories, and flags patients likely to reach end-stage renal disease within specific time windows. The value proposition is straightforward: if a platform costing $8 per patient per month helps a KCE delay or prevent dialysis initiation in just 2% of their at-risk population through earlier intervention, the ROI exceeds 10:1 for most entity sizes. The technical challenge involves building models that outperform simple eGFR thresholds by incorporating rate of decline, proteinuria patterns, comorbidity interactions, and adherence signals while remaining explainable enough for nephrologists to trust and act on. The go-to-market strategy requires direct sales to the 74 KCEs through nephrology conferences and digital channels, with implementation timelines under 60 days and immediate value demonstration through retrospective analysis of their own patient populations. This analysis examines why existing clinical tools fall short, what features drive adoption, how to build sustainable competitive advantages, and why the total addressable market extends far beyond KCC into dialysis organizations and nephrology practices managing fee-for-service populations.

Key Points:

- Average dialysis costs in Medicare approach $90,000-120,000 annually per patient, creating massive financial exposure for KCEs under two-sided risk

- Typical nephrology practices identify progressors reactively when eGFR crosses thresholds rather than predicting trajectory months in advance

- 30-40% of incident dialysis patients could potentially delay initiation through earlier intervention on modifiable risk factors

- Platform ROI calculation: $8 PMPM across 500 at-risk patients ($48,000 annually) versus preventing/delaying dialysis in 10 patients ($900,000-1,200,000 in cost avoidance)

- Market extends to 7,500+ nephrology practices and dialysis organizations beyond the 74 current KCEs

-----

Why Dialysis Progression Prediction Matters More Than Any Other Risk Model

The economics of kidney disease create a specific inflection point that dominates financial outcomes for any organization bearing population risk: the transition from pre-dialysis chronic kidney disease to end-stage renal disease requiring dialysis. Before this transition, a patient with Stage 3 or 4 CKD costs Medicare perhaps $15,000-25,000 annually across nephrology visits, labs, medications, and associated primary care. After dialysis initiation, costs jump to $90,000-120,000 annually just for dialysis services, plus increased hospitalization, medication, and specialist costs that often push total spending past $140,000.

For Kidney Contracting Entities operating under the KCC model, this transition represents existential financial risk. A KCE managing 1,000 pre-dialysis beneficiaries might have 50-80 patients at high risk of progression within the next 12-24 months. If all these patients start dialysis on schedule according to natural disease progression, the KCE faces incremental annual costs of $4.5-7.2 million versus their benchmark. Under the Global option with 50% downside risk, a bad progression year could cost the entity $2-3 million in losses at reconciliation.

The cruel math is that preventing progression entirely is usually impossible given that chronic kidney disease reflects permanent nephron loss, but delaying initiation by 6-12 months through better management of modifiable factors is achievable in a meaningful percentage of cases. Research suggests 30-40% of dialysis initiations happen earlier than clinically necessary, driven by factors like uncontrolled blood pressure, poor medication adherence, acute kidney injury episodes, or lack of conservative management options. A KCE that identifies high-risk progressors early and intervenes aggressively on these modifiable factors can materially change their financial trajectory.

The problem is that most nephrology practices identify progressors reactively rather than predictively. The typical workflow involves seeing patients quarterly, checking eGFR and creatinine, and initiating dialysis planning conversations when eGFR drops below 20 or the patient becomes symptomatic. This reactive approach misses the opportunity for earlier intensive intervention when patients are at eGFR 25-35 and still have time to modify their trajectory. By the time eGFR hits 15 and dialysis seems imminent, most modifiable factors have already caused irreversible damage.

The clinical tools available to nephrologists reinforce this reactive approach. They look at the most recent eGFR value and mentally compare it to previous values, but they rarely calculate actual decline rates or project time to dialysis using statistical models. Laboratory information systems show the latest result but don’t visualize trajectories or flag accelerating decline. Epic and other EHRs might have flowsheets showing trends, but they require manual navigation and don’t proactively alert clinicians to concerning patterns.

This gap between reactive identification and predictive opportunity creates the product space. A platform that continuously monitors every patient’s kidney function trajectory, calculates personalized time-to-dialysis predictions, and flags those at highest risk for progression in the next 6-12 months would fundamentally change how KCEs allocate intensive care management resources.

The Technology Architecture for Dialysis Progression Prediction

Read more

The Transparency in Coverage (TiC) rule promised to revolutionize healthcare price analytics by forcing payers to publish machine-readable files of negotiated rates. Instead, it created a data engineering nightmare. The TiC v2 specification was supposed to reduce file sizes and improve usability, but early implementations show file sizes increasing by an average of 10% rather than decreasing. For analytics companies attempting to build commercial products on this data, the fundamental challenge is not just obtaining the files but making them computationally tractable without burning through cloud budgets.

This essay examines the technical architecture required to operationalize TiC data at scale, focusing on storage optimization, query performance, and cost containment strategies. The analysis covers columnar storage formats, streaming ingestion patterns, serverless compute architectures, and the tradeoffs between different database technologies. For CTOs building analytics platforms on TiC data, the engineering decisions made in the first six months will determine whether the product has defensible unit economics or becomes a margin-destroying money pit.

Table of Contents

Why TiC v2 Failed to Solve the File Size Problem

The Real Cost of Storing and Processing TiC Data

Columnar Storage and Compression Strategies

Streaming Ingestion and Incremental Updates

Query Optimization for Analytical Workloads

Serverless vs Dedicated Compute Tradeoffs

Data Normalization and Entity Resolution

Building Programmatic Access Layers

Cost Containment and Unit Economics

Competitive Moats in TiC Analytics

Why TiC v2 Failed to Solve the File Size Problem

The TiC v2 specification introduced new fields intended to reduce redundancy and improve data quality. In theory, payers would deduplicate billing codes, normalize provider identifiers, and structure negotiated rates more efficiently. In practice, most payers approached v2 as an additive compliance exercise rather than a data architecture optimization. They bolted new fields onto existing file generation pipelines without rethinking the underlying data models.

The result is that v2 files contain more metadata, more granular rate structures, and more detailed provider information, all of which increase file sizes. Some payers are now publishing individual JSON files exceeding 50GB for single insurance products. The largest payers produce hundreds of these files monthly, creating a data corpus that exceeds multiple petabytes across the industry.

For startups attempting to build analytics products on this data, the first realization is that you cannot just download these files to an S3 bucket and query them with Athena. The economics do not work. A single query scanning unoptimized TiC data can cost hundreds of dollars in compute charges. Multiply that by thousands of queries per day from customers running analyses, and you quickly burn through seed funding without generating useful insights.

The second realization is that most existing healthcare data infrastructure was not designed for this scale. The industry runs on SQL Server instances managing claims databases measured in hundreds of gigabytes, not distributed systems managing petabytes of semi-structured JSON. The technical patterns that work for claims adjudication do not translate to TiC analytics. This is fundamentally a big data problem, but most healthcare data engineers have never built big data systems.

The Real Cost of Storing and Processing TiC Data

Consider a mid-sized analytics startup attempting to ingest TiC data from the top 50 national and regional payers. Assume each payer publishes an average of 200 files per month, with average file sizes of 5GB. That is 50 payers times 200 files times 5GB, which equals 50TB of raw data per month. Over a year, you are looking at 600TB of storage before accounting for any processing, indexing, or derived datasets.

At AWS S3 standard storage rates of roughly $23 per TB per month, storing 600TB costs $13,800 monthly just for the raw files. But you cannot serve queries directly from S3 without transformation. You need to parse the JSON, normalize the schema, deduplicate records, and load the data into a queryable format. Depending on your architecture, this might mean writing to Snowflake, BigQuery, Databricks, or a self-managed data lake using Parquet files.

Snowflake pricing for storage is around $23-40 per TB per month depending on the region, similar to S3, but compute costs are where things get expensive. Snowflake charges roughly $2-3 per credit, and a medium warehouse consuming 4 credits per hour costs around $8-12 per hour. If your ingestion pipeline runs continuously to keep up with new file releases, you could easily burn $5,000-10,000 per month just on warehouse compute for data loading, before any customer queries run.

The alternative is building a data lake architecture using Spark or similar distributed processing frameworks. This gives you more control over compute costs but requires maintaining infrastructure and hiring engineers who understand distributed systems. For a seed-stage company, the tradeoff is between paying for managed services that eat into margins versus building custom infrastructure that delays time to market.

The key insight is that raw file storage is a minor cost compared to transformation and query compute. The engineering challenge is designing a data pipeline that minimizes redundant processing. If you naively re-parse and re-load entire files every time a payer publishes an updated version, you waste compute on records that have not changed. Incremental updates require tracking file versions, computing diffs, and merging changes, which adds complexity but dramatically reduces processing costs.

Columnar Storage and Compression Strategies

The most impactful architectural decision for TiC analytics is choosing a storage format optimized for analytical queries. Row-oriented formats like JSON are terrible for analytics because queries typically scan specific columns across millions of records. Reading a single column from a JSON file requires parsing every record in the file, even though you are discarding 95% of the data.

Columnar formats like Parquet, ORC, or Arrow store data by column rather than by row, which means scanning a single column only reads that column’s data from disk. This reduces I/O by orders of magnitude for typical analytical queries. Parquet also supports efficient compression because values in a single column tend to be similar, allowing compression algorithms to achieve higher ratios than they would on heterogeneous row data.

Converting TiC JSON files to Parquet can reduce storage by 80-90% depending on the data distribution and compression codec. A 5GB JSON file might compress to 500MB-1GB as Parquet with Snappy or Zstd compression. This directly translates to lower storage costs and faster query performance because less data needs to be read from disk.

The tradeoff is that Parquet is immutable and does not support updates in place. If a payer publishes an updated file with changed rates, you cannot directly update existing Parquet records. Instead, you need to write new files with the updated data and handle versioning at the query layer. This is manageable with modern data lake formats like Delta Lake, Iceberg, or Hudi, which provide ACID transactions and time travel on top of Parquet files.

Delta Lake, for example, allows you to upsert records into a Parquet-based table, and it manages versioning by writing new files and tracking metadata about which files belong to which version. Queries automatically read from the latest version unless you specify a historical snapshot. This gives you the performance benefits of columnar storage with the operational convenience of update semantics.

Another compression technique is dictionary encoding, where repeated string values are replaced with integer codes referencing a dictionary. TiC data is highly repetitive with the same billing codes, provider identifiers, and rate structures appearing millions of times across files. Dictionary encoding can compress these columns by 95%+ because instead of storing the string “99213” millions of times, you store it once and reference it with a 4-byte integer.

Parquet supports dictionary encoding natively, and most modern query engines can push filters down to dictionary-encoded columns without decompressing the data. This means a query filtering on a specific billing code can skip entire row groups without reading them, further reducing I/O. The combination of columnar layout, compression, and predicate pushdown makes Parquet the de facto standard for big data analytics, and it is the right choice for TiC data.

Streaming Ingestion and Incremental Updates

Most payers publish TiC files on a monthly cadence, but the release dates are inconsistent and unpredictable. Some payers update files mid-month with corrections, others publish late, and a few payers have failed to publish files altogether despite regulatory requirements. This creates an ingestion challenge where your pipeline needs to continuously monitor for new files, detect updates, and process them incrementally without re-ingesting unchanged data.

A naive approach is running a cron job daily that downloads all files, checks for changes, and re-processes everything. This works but wastes compute on files that have not changed. A more efficient approach is tracking file metadata like ETags, last-modified timestamps, or content hashes, and only downloading files that have changed since the last ingestion run.

AWS S3 supports ETags, which are MD5 hashes of file contents for files uploaded in a single part. By storing the ETag of each file you have previously ingested, you can query the S3 API for current ETags and only download files where the ETag has changed. This requires maintaining a metadata store tracking file URLs, ETags, and ingestion timestamps, but it dramatically reduces download bandwidth and processing time.

For files that have changed, you need to determine what changed. Did the payer add new records, update existing records, or delete records? TiC files do not include change logs or CDC events, so you have to infer changes by comparing the new file against the previous version. This requires loading both versions, computing a diff, and generating a changelog that can be applied to your warehouse.

One approach is using content-addressable storage where each record is hashed and stored with its hash as the key. When a new file arrives, you hash each record and check if it exists in your store. New hashes indicate new or updated records, and missing hashes indicate deletions. This allows you to compute diffs without loading entire files into memory, because you can stream records, hash them, and check existence in a key-value store like RocksDB or DynamoDB.

The engineering complexity here is non-trivial, but the cost savings are substantial. Incremental ingestion reduces compute by 10-50x compared to full re-processing, depending on how frequently files change. For a production system ingesting hundreds of files daily, this is the difference between sustainable unit economics and burning money on redundant compute.

Query Optimization for Analytical Workloads

The typical query pattern for TiC analytics is aggregating rates across providers, billing codes, and payers to answer questions like “what is the median negotiated rate for CPT code 99213 in Boston across all payers” or “which providers have the highest rates for knee replacements.” These queries scan millions of records, apply filters, group by dimensions, and compute aggregates.

The naive approach is loading all data into a SQL database and running these queries directly. This works for small datasets but becomes prohibitively slow as data scales to billions of records. A single query scanning billions of rows can take minutes to hours on a traditional database, which is unacceptable for an interactive analytics product.

The solution is partitioning data along query access patterns and using indexes or metadata to prune partitions before scanning. For TiC data, common partition keys are payer, billing code, geography, or provider. If queries typically filter by payer, partition data by payer so each payer’s data lives in separate files. When a query filters to a specific payer, only that payer’s files are scanned.

Partitioning reduces query latency by 10-100x depending on the partition cardinality and query selectivity. A query filtering to a single payer might scan 1/50th of the data if there are 50 payers, which translates to 50x less I/O and proportionally faster execution. The tradeoff is that over-partitioning creates too many small files, which increases metadata overhead and reduces query efficiency. Finding the right partition granularity requires profiling actual query patterns.

Another optimization is pre-aggregating common queries into materialized views or summary tables. If 80% of queries ask for median rates by billing code and geography, you can pre-compute these aggregates and store them in a much smaller table. Queries against the summary table return instantly, and you only fall back to scanning raw data for edge cases not covered by pre-aggregations.

Maintaining materialized views requires updating them when underlying data changes, which adds complexity. The incremental update strategy described earlier helps here because you can identify which partitions changed and only refresh affected materialized views. For data that updates monthly, the refresh cost is manageable and the query performance improvement is dramatic.

Modern data warehouses like Snowflake, BigQuery, and Databricks have built-in support for materialized views, automatic query caching, and adaptive query optimization, which reduces the need for manual tuning. However, understanding the underlying principles is critical for making architectural tradeoffs and debugging performance issues when they arise.

Serverless vs Dedicated Compute Tradeoffs

A fundamental architectural decision is whether to use serverless compute services like AWS Lambda, BigQuery, or Snowflake serverless, versus dedicated compute like EC2 instances, Kubernetes clusters, or Databricks classic clusters. Serverless offers elasticity and pay-per-use pricing, which is attractive for unpredictable workloads. Dedicated compute offers lower per-hour costs but requires capacity planning and pays for idle time.

For TiC ingestion, serverless makes sense because file arrivals are bursty and unpredictable. Using Lambda functions triggered by S3 events allows processing files as they arrive without maintaining always-on infrastructure. Lambda pricing is roughly $0.20 per million requests plus $0.0000166667 per GB-second of compute, which is cheap for intermittent workloads.

The limitation is that Lambda functions have a 15-minute execution timeout and 10GB memory limit, which may not be sufficient for processing large TiC files. Files exceeding 10GB need to be split or processed in chunks, which adds complexity. An alternative is using AWS Fargate or ECS tasks, which support longer execution times and higher memory limits, while still offering serverless-like operational simplicity.

For query workloads, the tradeoff depends on query patterns. If customers run ad-hoc queries sporadically, serverless warehouses like BigQuery or Snowflake serverless make sense because you only pay for queries executed. If customers run continuous dashboards or scheduled reports, dedicated compute clusters amortize fixed costs across many queries, reducing per-query costs.

The break-even point is roughly when utilization exceeds 30-50%. If a cluster is idle more than 50% of the time, serverless is cheaper. If a cluster is busy more than 50% of the time, dedicated is cheaper. For most early-stage analytics products, traffic is low and unpredictable, so serverless is the right default. As usage scales, migrating to dedicated clusters for high-utilization workloads optimizes costs.

Data Normalization and Entity Resolution

TiC data is notoriously messy with inconsistent provider identifiers, duplicate billing codes, and malformed rate structures. Payers use different identifier systems for providers, some using NPIs, others using proprietary IDs, and many using a mix of both. Billing codes are sometimes formatted with punctuation like dashes or periods, sometimes without, and occasionally with typos or invalid codes.

Building a useful analytics product requires normalizing this mess into a consistent schema where providers, billing codes, and rates are uniquely identified and can be joined across payers. This requires entity resolution, which is the process of determining when two records refer to the same real-world entity despite having different identifiers or attributes.

For providers, the gold standard identifier is the NPI, but not all TiC records include NPIs. Some records only include a name and address, which requires fuzzy matching against a reference database like NPPES to resolve the provider’s NPI. Fuzzy matching is computationally expensive because it requires comparing each record against millions of potential matches using edit distance or phonetic algorithms.

A scalable approach is using locality-sensitive hashing (LSH) to reduce the search space. LSH hashes similar strings to the same bucket, so you only compare records that hash to the same bucket, reducing comparisons by 1000x or more. After LSH reduces the candidate set, you apply more expensive algorithms like Levenshtein distance or Jaro-Winkler to score matches and select the best candidate.

For billing codes, normalization is simpler because codes follow standardized formats like CPT, HCPCS, or ICD-10. The challenge is handling variations like leading zeros, dashes, or case sensitivity. A simple normalization pipeline that strips non-alphanumeric characters and uppercases codes catches most variations. Invalid codes that do not exist in reference datasets like the CMS HCPCS file can be flagged for manual review or dropped.

Entity resolution is not a one-time process because new providers and codes appear over time. A production system needs continuous monitoring to detect new entities, resolve them, and update reference datasets. This requires data pipelines that version reference data and propagate updates through the system without breaking downstream queries.

Debugging entity resolution issues is tedious because errors propagate through the system and cause incorrect aggregates. A provider misidentified as two separate entities will have their rates split across two records, resulting in incorrect median calculations. Catching these errors requires comprehensive testing and validation, including sample queries that check for known ground truth results.

Building Programmatic Access Layers

The end goal of operationalizing TiC data is exposing it through APIs that customers can integrate into their applications. The API design determines how easy it is for developers to adopt your product and how much support burden you carry. A well-designed API abstracts the complexity of TiC data and provides intuitive, fast, and reliable access to rates.

The core API endpoints typically include searching for rates by billing code, provider, payer, or geography, returning aggregated statistics like median, percentile, or range, and filtering by plan type, network, or other attributes. These endpoints need to return results in milliseconds to support interactive applications, which requires careful indexing and caching.

One approach is using a caching layer like Redis or Memcached to store frequently accessed queries. When a query is executed, check if the result is cached, return it immediately if so, or execute the query, cache the result, and return it. Cache invalidation is the hard part because cached results become stale when underlying data updates. A simple strategy is time-based expiration where cached results expire after a fixed duration like one hour or one day.

For queries not covered by cache, the API needs to execute them against the data warehouse efficiently. This requires translating API parameters into SQL queries with appropriate filters, joins, and aggregations, while preventing SQL injection or malformed queries that could crash the database. Using a query builder library or ORM helps, but custom validation logic is still necessary to enforce business rules like maximum query size or allowed parameter combinations.

Rate limiting is critical to prevent customers from overloading the API and driving up compute costs. A token bucket or leaky bucket algorithm limits requests per customer per time window, returning 429 errors when limits are exceeded. Setting appropriate rate limits requires understanding your infrastructure capacity and cost structure. Too strict and customers cannot use the product effectively, too loose and a single customer can blow your entire compute budget.

Monitoring API performance and errors is essential for maintaining reliability. Instrument APIs with metrics tracking request counts, latency percentiles, error rates, and cache hit ratios. Set up alerts for anomalies like sudden latency spikes or elevated error rates, which indicate infrastructure issues or data problems. A comprehensive monitoring setup detects issues before customers report them, allowing proactive fixes.

Cost Containment and Unit Economics

The ultimate constraint for any TiC analytics company is whether unit economics work. If customer acquisition cost exceeds lifetime value, the business is not viable regardless of technical sophistication. For infrastructure-heavy products, the dominant cost component is often compute and storage, so optimizing these costs is existential.

The starting point is understanding cost per query or cost per customer per month. Instrument your pipeline to track compute and storage costs attributable to each customer, which requires tagging resources with customer identifiers and aggregating costs in billing reports. This reveals which customers are profitable and which are losing money, allowing you to adjust pricing or usage limits.

A common pattern is that a small fraction of customers drive the majority of costs. The Pareto principle applies where 20% of customers might consume 80% of compute resources. For these high-usage customers, enforcing rate limits or upselling to higher-priced tiers is necessary to maintain profitability. Alternatively, optimizing their specific query patterns can reduce costs without impacting their experience.

Another cost containment strategy is intelligent query routing. If a query can be satisfied by cached data or pre-aggregated summaries, route it there instead of hitting the data warehouse. Only execute expensive queries against raw data when absolutely necessary. This requires a decision engine that analyzes queries and selects the optimal execution path based on cost and latency tradeoffs.

Spot instances and reserved capacity offer significant cost savings for predictable workloads. AWS EC2 spot instances cost 70-90% less than on-demand, though they can be interrupted with two-minute notice. For batch processing jobs like nightly data refreshes, spot instances are ideal because interruptions can be tolerated and retried. Reserved instances offer 30-50% discounts for one or three-year commitments, which makes sense for baseline capacity that runs continuously.

Storage tiering is another cost optimization. Not all data needs to live in hot storage optimized for low-latency access. Historical data older than six months might be queried rarely and can be moved to cheaper cold storage like S3 Glacier, reducing storage costs by 80%+. Implementing lifecycle policies that automatically tier data based on access patterns keeps hot storage lean and costs low.

Competitive Moats in TiC Analytics

The TiC data market is nascent but rapidly attracting startups and established players. For a new entrant, the question is what defensible advantage can be built before competitors catch up. Pure data aggregation is not defensible because anyone can download TiC files. Differentiation comes from data quality, analytical capabilities, and integration ease.

Data quality is the most important differentiator because raw TiC files are nearly unusable without extensive cleaning, normalization, and enrichment. A company that invests in entity resolution, error correction, and completeness checks builds a dataset that is qualitatively better than raw files. This quality advantage compounds over time as the company builds institutional knowledge about payer quirks and data anomalies.

Analytical capabilities that provide unique insights are another moat. If your product can answer questions that competitors cannot, customers have a reason to choose you. This might include linking TiC rates to claims data to estimate actual transaction prices, combining rates with quality metrics to identify high-value providers, or using machine learning to predict future rate changes based on historical trends.

Integration ease determines how quickly customers can deploy your product. If integration takes weeks of custom development, customers will hesitate or choose simpler alternatives. Providing SDKs, pre-built connectors for common platforms, and comprehensive documentation reduces integration friction. Some companies go further by offering white-label solutions that customers can rebrand and resell, creating distribution leverage.

Network effects are possible if your product enables data sharing or benchmarking across customers. For example, if customers contribute usage data back to the platform, you can provide richer benchmarks showing how a customer’s rates compare to peer organizations. This data flywheel makes the product more valuable as more customers adopt it, creating a classic two-sided network effect.

The strongest moats combine multiple advantages. A company with the cleanest data, the best analytical tools, and the easiest integration is difficult to displace even if competitors match any single dimension. Building this requires sustained investment in engineering, data operations, and customer success, which takes time but creates durable competitive advantage.

The opportunity in TiC analytics is real because the regulatory mandate ensures data availability and the market need for price transparency is enormous. However, the technical challenges are substantial and underestimated by most teams. Success requires deep expertise in distributed systems, data engineering, and healthcare domain knowledge. For CTOs willing to invest in the right architecture upfront, the payoff is a scalable, profitable product that solves a genuine market problem. For those who underinvest in infrastructure, the result is a product that cannot scale, burns money on compute, and collapses under its own weight before achieving product-market fit. The engineering decisions made in the first six months determine which path a company follows.​​​​​​​​​​​​​​​​

This essay examines current opportunities in healthcare technology through the lens of recently published compliance guidance from the Office of Inspector General regarding Medicare Advantage programs. The document reveals significant operational gaps, enforcement priorities, and structural weaknesses that represent addressable market opportunities for entrepreneurs and investors. Key themes include:

- Compliance infrastructure represents a $15B+ market opportunity across MAOs, providers, and third parties

- Risk adjustment accuracy remains systematically problematic despite being central to $400B+ in annual payments

- Prior authorization and utilization management create massive friction costs while generating quality concerns

- Provider network accuracy and directory maintenance remain unsolved at scale

- Third party oversight and FDR management lacks adequate tooling

- Marketing compliance presents both risk and opportunity as regulatory scrutiny intensifies

The analysis connects regulatory pressure points to product opportunities, with particular focus on software infrastructure, data analytics, and process automation that can address compliance requirements while improving operational efficiency.

Table of Contents

Introduction and Context

The Compliance Infrastructure Opportunity

Risk Adjustment as a Product Category

Prior Authorization and Utilization Management

Network Adequacy and Provider Directory Solutions

Third Party Oversight and Vendor Management

Marketing and Enrollment Compliance Tools

Quality Measurement and Star Ratings Infrastructure

Integrated Platforms vs Point Solutions

Market Sizing and Investment Considerations

Conclusion

Introduction and Context

The OIG just dropped a 42 page document that basically reads like a product roadmap if you know how to interpret it. The Medicare Advantage Industry Compliance Program Guidance updates their 1999 version, which tells you something about how much the landscape has changed. MA enrollment has exploded to cover more than half of Medicare beneficiaries, creating a roughly $450B market that grows 8 to 10 percent annually. The guidance document is nominally about compliance but really it is a detailed catalog of operational problems that MAOs cannot solve with their current tools and processes.

This matters because regulatory guidance documents reveal where enforcement will focus, which means they reveal where companies will spend money to avoid penalties. More importantly, they reveal systematic operational failures across an industry segment. When OIG publishes 40+ pages detailing specific processes that MAOs should implement, they are essentially admitting that current approaches do not work. Each recommendation represents a potential product.

The key insight is that compliance requirements and operational efficiency are not opposed. The areas where OIG identifies the highest risk tend to be areas where current processes are manual, error prone, and expensive. Building tools that make compliance easier usually means building tools that make operations better. Companies that can deliver both will capture disproportionate value.

The Compliance Infrastructure Opportunity

Start with the basics. CMS requires MAOs to maintain compliance programs with specific elements including dedicated officers, committees, training programs, monitoring systems, and corrective action processes. The guidance makes clear that current implementations are inadequate. MAOs operate with compliance teams that are too small, lack specialized MA expertise, rely on manual processes, and struggle to maintain oversight across complex organizational structures.

The market is roughly 500 MAOs ranging from massive national carriers to regional plans with a few thousand members. Below that are several thousand provider organizations acting as FDRs, thousands of TPMOs handling marketing and enrollment, and various vendors providing everything from utilization management to risk adjustment services. Altogether maybe 10,000 organizations need meaningful compliance infrastructure specifically for MA programs.

Most of these organizations build compliance programs using general purpose tools, spreadsheets, and manual processes. A dedicated compliance platform purpose built for MA requirements could command $50K to $500K annually depending on organization size. Even at conservative estimates, this is a $500M+ market just for core compliance infrastructure.

But the real opportunity is not selling compliance software to compliance officers. It is embedding compliance into operational workflows so that doing the right thing becomes the path of least resistance. Think about how Stripe embedded PCI compliance into payment processing or how Modern Treasury embedded banking compliance into money movement. The companies that win will make compliance invisible rather than making it a separate workstream.

The guidance document essentially provides a specification. MAOs need systems that maintain policies and procedures, track training completion, manage hotline reports, document investigations, oversee third parties, conduct audits, and generate reports for boards and regulators. Each of these represents a distinct module within a larger platform.

Risk Adjustment as a Product Category

Read more

UnitedHealth Group reported full year 2025 revenues of $449 billion with adjusted earnings of $28.15 per share, representing 8% revenue growth but significant earnings pressure from medical cost trends. The company’s results reveal fundamental shifts in healthcare economics that directly impact health tech companies building in payer-adjacent spaces. Key themes include elevated medical utilization persisting beyond COVID normalization, Medicare Advantage margin compression forcing product redesign, Medicaid redeterminations creating permanent coverage gaps, and continued resistance to specialty drug cost trends. UnitedHealthcare’s medical loss ratio increased 330 basis points year over year while Optum Health struggled with risk adjustment headwinds and care delivery margin pressure. The results suggest a structural shift away from the benign cost environment of 2021-2023 and toward sustained pressure on margins, utilization management intensity, and care delivery economics. For health tech companies, this translates to increased scrutiny on ROI demonstrations, longer sales cycles for cost-reduction solutions, and potential opportunities in utilization management infrastructure and alternative site-of-care enablement.

Table of Contents

The Numbers That Matter

What Actually Happened in 2025

Medicare Advantage Is Breaking (In Slow Motion)

The Medicaid Hangover Nobody Talks About

Optum’s Reality Check

What This Means for Digital Health Companies

The Contrarian Opportunities

How To Think About Building Now

The Numbers That Matter

UnitedHealth just closed the books on 2025 and the top line looks fine until you start digging into what’s underneath. The company did $449 billion in revenue, up 8% from 2024, which sounds perfectly acceptable for a company that size. But the adjusted earnings per share came in at $28.15, which if you’re tracking along with Street expectations tells you pretty much everything about how the year actually went. Revenue growth without corresponding earnings leverage means margin compression, and margin compression at a company that touches roughly one in eight Americans tells you something structural changed in the healthcare economy.

The medical loss ratio for UnitedHealthcare hit 85.1% for the full year, up 330 basis points from 81.8% in 2024. For context, that’s the difference between a well-oiled insurance operation and one that’s suddenly spending an extra $12 billion on medical costs relative to premiums. The company blamed it on higher medical cost trends, which is corporate speak for people are using more healthcare and it costs more than we priced for. The fourth quarter was particularly rough with an MLR of 86.3%, suggesting the trend accelerated as the year progressed rather than moderating like everyone hoped.

What makes this especially interesting is that UHC grew revenues to $316 billion across the insurance business while operating earnings actually declined to $16.4 billion from $16.8 billion the prior year. That’s not a rounding error, that’s a fundamental shift in unit economics. When you’re adding billions in top line revenue and losing operating income, something broke in the pricing model or the utilization assumptions or both.

Optum, which everyone has spent the last five years calling UHG’s growth engine and margin expansion story, posted $214 billion in revenues but saw operating earnings decline 17% to $11.2 billion. The company tried to blame this on a cyberattack affecting Change Healthcare, which definitely hurt, but the guidance for 2026 suggests the problems run deeper than a one-time security incident. When your health services division that’s supposed to be delivering margin expansion is instead contracting margins, that tells you something about care delivery economics that should make every value-based care investor pause.

What Actually Happened in 2025

The simple version is that medical utilization stayed elevated all year and cost trends ran hotter than pricing. But that explanation obscures what’s actually interesting, which is why utilization stayed high and why costs surprised to the upside despite three years of health plans supposedly getting better at managing both.

Coming out of COVID, there was this prevailing theory that we’d see normalization of utilization patterns as deferred care got caught up and the system returned to pre-pandemic equilibrium. That happened partially in 2022 and 2023, which created this window where health plans had favorable medical cost trends because people were still being conservative about seeking care but acuity was normalizing. That window closed in 2024 and stayed closed in 2025.

What UHC is seeing, and what other plans are reporting in their earnings, is that utilization is running persistently above pre-COVID baselines across multiple categories. Outpatient activity is elevated, specialty care is elevated, diagnostic testing is elevated, and importantly these increases are showing up in commercially insured populations not just Medicare and Medicaid. The company specifically called out higher outpatient activity levels and increased intensity of services, which is healthcare consultant speak for people are going to the doctor more and getting more stuff done when they go.

The cost side is equally interesting because it’s not just volume, it’s price mix shifting toward more expensive sites of care and more expensive services within sites of care. Hospital outpatient departments are capturing more volume that used to happen in physician offices, specialty drugs are becoming standard of care for conditions that used to get managed with generics, and diagnostic capabilities that used to be limited to academic medical centers are now available at community hospitals. All of this shifts the cost curve independent of utilization volume.

UHC also took a $600 million hit from the Change Healthcare cyberattack, which disrupted claims processing and revenue cycle operations across thousands of providers. The interesting part isn’t the immediate cost, it’s what it revealed about system fragility. When one clearinghouse going down for a few weeks causes that much economic damage, it tells you how concentrated and brittle the administrative infrastructure has become. For health tech companies building in that infrastructure layer, that’s simultaneously a warning about single points of failure and an indication of how valuable resilience and redundancy will become.

Medicare Advantage Is Breaking (In Slow Motion)

The Medicare Advantage story is probably the most important one buried in these results for anyone building companies that touch seniors or value-based care. UHC’s Medicare Advantage membership grew to 8.5 million people, up 5% year over year, which looks like healthy growth until you realize the operating margin on that membership is compressing rapidly.

The company doesn’t break out Medicare Advantage MLR separately, but they made it pretty clear in the remarks that MA was a significant driver of the overall MLR deterioration. The combination of benchmark rate pressure from CMS, higher medical costs from an aging and increasingly complex membership, and continued challenges with risk adjustment coding created a perfect storm for margins.

CMS has been systematically reducing MA benchmarks through the rate notice process while simultaneously tightening risk adjustment documentation requirements and launching audits that claw back prior year payments. The 2025 rate notice was particularly painful, with effective rate increases well below medical cost trend, and the 2026 notice that came out a few months ago continues that pattern. Health plans are stuck between pricing plans that are attractive enough to drive enrollment while somehow making money on benchmark rates that don’t cover medical costs.

What makes this especially problematic is the membership mix is shifting unfavorably. The easy MA members, the ones who are healthy and low-cost and generate positive margins even at compressed benchmarks, have mostly already enrolled. The marginal member coming into MA now is older, sicker, more likely to have multiple chronic conditions, and more expensive to care for. As plans have expanded into rural areas and lower income populations chasing growth, the risk mix has deteriorated.

The risk adjustment model changes CMS implemented over the last few years have also created significant headwinds. The agency is trying to reduce risk score growth because they believe plans have been gaming the system through aggressive coding and diagnosis hunting. They’re probably right about the gaming, but the pendulum has swung far enough that plans are seeing risk scores decline even as their member populations get objectively sicker. UHC specifically called out risk adjustment challenges affecting Optum Health, where declining risk scores directly impact capitation revenues.

For health tech companies, this has huge implications. The entire value-based care infrastructure built over the last decade was predicated on expanding margins in MA creating room to invest in care transformation and technology. If MA margins are compressing structurally, the economics of value-based care change fundamentally. Groups that were planning to scale into breakeven or profitability on volume growth are going to discover their unit economics don’t work anymore. Technology vendors selling into MA-focused groups are going to face much harder ROI scrutiny and longer sales cycles.

The Medicaid Hangover Nobody Talks About

Medicaid should be a footnote in UHC’s results because it’s a relatively small part of their book, but it’s worth understanding because the dynamics are playing out across the industry and affecting companies building in that market.

The Medicaid redeterminations that happened through 2023 and 2024, where states went through their rolls and disenrolled people who were no longer eligible after the continuous coverage requirement ended, created massive disruption in coverage patterns. UHC, like every other health plan, saw significant membership declines as states worked through their backlogs. The company ended 2025 with materially fewer Medicaid members than it started with.

What’s interesting is that the members who remained tend to be higher acuity and more expensive to serve. The redetermination process disproportionately removed healthier, younger members who had other coverage options or who didn’t complete the renewal paperwork. What’s left is a sicker, more complex population with higher medical costs relative to the capitation rates states are paying.

States are also increasingly broke and unwilling or unable to increase Medicaid rates to keep pace with medical cost trends. The combination of state budget pressure, federal matching rate mechanics, and political resistance to Medicaid expansion means rates are lagging costs in most states. Health plans are stuck serving populations where the rate environment doesn’t cover medical costs and the only way to make money is aggressive utilization management, narrow networks, or just exiting markets entirely.

UHC has been relatively disciplined about exiting unprofitable Medicaid markets, which is why their Medicaid book is smaller than it used to be. But for health tech companies building Medicaid solutions, this should be a massive red flag. If the largest and most sophisticated health plan in the country can’t make money in Medicaid, how is a venture-backed startup going to make money solving Medicaid problems? The market opportunity might be large in terms of covered lives, but the revenue opportunity is constrained by rate structures that don’t support innovation investment.

Optum’s Reality Check

Optum is supposed to be the growth story within UHG, the diversified health services platform that delivers margins, creates competitive moats through vertical integration, and generates recurring revenue streams independent of insurance underwriting risk. The 2025 results suggest that narrative needs updating.

Optum Health, the care delivery and value-based care arm, grew revenues to $99 billion but saw operating margins compress significantly. The business serves 31 million people in value-based care arrangements, which sounds impressive until you realize the economics of those arrangements deteriorated materially in 2025. The combination of risk adjustment headwinds, higher medical costs, and challenges scaling their care delivery infrastructure created margin pressure across the portfolio.

The company has been investing heavily in building owned and operated care delivery assets, including primary care clinics, surgical centers, and specialty care facilities. The theory was that owning the care delivery allows better cost management, improved care coordination, and margin capture across the value chain. The reality in 2025 was that care delivery is hard, labor markets are tight, and the fixed cost base of owned facilities creates operating leverage that cuts both ways.

Optum Rx, the pharmacy benefit management business, did $174 billion in revenues, which is an absurd number that reflects both the underlying growth in specialty drug spending and their success capturing share in the PBM market. But the operating margin in the PBM business is compressing because the fundamental economics of PBM are changing. Spread pricing is under regulatory pressure, rebate retention is under scrutiny, and plans are increasingly demanding transparency that reduces the information asymmetry PBMs have historically exploited.

The Change Healthcare cyberattack hit Optum Insight particularly hard, disrupting the revenue cycle management and payment processing operations that serve thousands of providers. The direct costs were substantial but the indirect costs in terms of customer relationships, regulatory exposure, and needed security infrastructure investment are probably larger. For health tech infrastructure companies, this is a reminder that when you become critical infrastructure the consequences of failure become existential.

What This Means for Digital Health Companies

If you’re building a health tech company that touches payers or providers or relies on favorable health economics, these results should recalibrate your assumptions pretty dramatically.

First, sales cycles are going to get longer and ROI thresholds are going to get higher. When health plans are dealing with margin compression and elevated medical costs, every dollar of spend gets scrutinized more carefully. Solutions that might have gotten approved based on directional evidence of impact are now going to need to demonstrate clear ROI with conservative assumptions. Pilots that used to convert to enterprise contracts in six months are going to take twelve to eighteen months as organizations demand more proof before committing budget.

Second, the risk environment for value-based care companies just got materially harder. If Medicare Advantage margins are compressing structurally, the pathway to profitability for groups taking full risk gets much narrower. Companies that were planning to scale into profitability through membership growth are going to discover their unit economics don’t work when benchmark rates barely cover medical costs. The groups that survive are going to be the ones with genuine care transformation capabilities, not just risk aggregation plays.

Third, utilization management is going to become more aggressive across the board. When medical costs are running hot and margins are compressed, health plans have to manage utilization more tightly to hit their financial targets. That means prior authorization requirements expanding, step therapy protocols becoming more restrictive, and network designs getting narrower. Digital health companies that facilitate utilization are going to face more resistance than companies that demonstrably reduce utilization or shift it to lower cost settings.

Fourth, the pricing environment for health tech solutions is going to be under pressure. When customers are margin-constrained, they push back harder on vendor pricing and demand more favorable commercial terms. The days of 20% annual price increases are over for most categories. Companies need to plan for flat or declining unit pricing and make up for it through volume growth or operational efficiency.

Fifth, the regulatory environment is going to stay intense. When health plans are financially stressed, they tend to take more aggressive positions on things like risk adjustment coding, prior authorization, claims denials, and network adequacy. That creates more regulatory attention and enforcement action, which creates more compliance costs and operational friction. Health tech companies operating in regulated spaces need to plan for higher compliance costs and more regulatory scrutiny.

The Contrarian Opportunities

The flip side of all this margin pressure and cost growth is that it creates opportunities for companies that can genuinely solve the underlying problems rather than just selling software.

Utilization management infrastructure is going to be a massive opportunity. Health plans need better tools for managing prior authorization, for identifying inappropriate utilization, for steering members to cost-effective providers, and for preventing avoidable acute care. The legacy systems most plans use for this are terrible, which is why prior auth is so frustrating for providers and members. Companies that can build intelligent, clinically appropriate, operationally efficient utilization management tools are going to find a lot of demand.

Alternative site of care enablement is another big opportunity. One reason costs are elevated is because too much care happens in expensive settings like hospital outpatient departments and emergency departments when it could happen in ambulatory surgery centers, urgent care clinics, or even at home. Technology that makes it easier and safer to shift care to lower cost settings creates real value for payers. This includes things like hospital at home platforms, ambulatory surgical center enablement, diagnostic testing in retail settings, and specialty care telemedicine.

Risk adjustment and clinical documentation improvement is going to be permanently important. The tension between CMS trying to reduce risk score growth and health plans trying to capture appropriate risk adjustment isn’t going away. Companies that can help accurately document member acuity in clinically appropriate ways while maintaining compliance with documentation requirements are going to have durable businesses. The key is genuinely improving documentation accuracy rather than just maximizing scores, because the regulatory scrutiny on risk adjustment is only going to increase.

Specialty drug cost management is a massive unsolved problem. Specialty pharmaceuticals are the fastest growing category of medical spend and the hardest to manage. Prior authorization and step therapy help at the margins but don’t address the fundamental issue that these drugs are incredibly expensive and increasingly becoming standard of care. Companies working on specialty pharmacy logistics, biosimilar adoption, dose optimization, adherence improvement, or alternative business models for specialty drug access are working on problems that payers desperately need solved.

Claims processing and revenue cycle infrastructure is going to get rebuilt. The Change Healthcare incident proved how fragile the existing systems are and how concentrated the risk is in a few large clearinghouses. There’s going to be demand for more resilient, more distributed, more secure infrastructure for claims processing, eligibility verification, and payment processing. This is hard infrastructure work without obvious venture returns, but it’s genuinely valuable and the need is acute.

How To Think About Building Now

If you’re building a health tech company in this environment, the strategy needs to shift pretty materially from what worked in 2021-2023.

Customer selection matters more than ever. Not all health plans or provider organizations are equally attractive customers. The ones under the most margin pressure are going to be the hardest to sell to and the slowest to pay. Focus on customers with relatively stable economics, sophisticated procurement processes, and track records of implementing technology successfully. Avoid customers in financial distress even if they seem desperate for solutions.

ROI demonstration needs to be rigorous and conservative. Hand-waving about directional impact or long-term value creation isn’t going to work anymore. Customers need to see clear ROI with conservative assumptions, validated through pilots or case studies, with attribution methodologies they trust. Build your ROI modeling early and stress test it against skeptical assumptions. If your model doesn’t hold up under conservative assumptions, rework the business model.

Product pricing needs to align with customer economics. If your customers are margin-constrained, they can’t afford expensive software. Think about pricing models that align your success with their success, like performance-based pricing or risk-sharing arrangements. Avoid pricing models that create large upfront costs or ongoing fees that don’t scale with value delivered.

Sales cycles are going to be longer, so burn rate management becomes critical. If it’s taking eighteen months instead of six months to close enterprise deals, you need more runway to get to the same revenue milestones. Either raise more money, reduce burn, or both. The companies that die in this environment are the ones that run out of cash waiting for deals to close.

Regulatory compliance is table stakes, not a nice-to-have. If you’re working in spaces that touch PHI, claims data, or regulated transactions, compliance needs to be built in from day one. The cost of fixing compliance problems after the fact is prohibitive and the regulatory risk is real. Budget for compliance infrastructure, hire compliance expertise early, and design products with compliance requirements in mind.

Understand that health plans are going to be more conservative about vendor relationships. They’re going to ask harder questions about financial stability, business continuity, data security, and regulatory compliance. They’re going to demand more onerous contract terms, longer payment cycles, and more extensive proof of concept before committing. This is just the reality of selling into stressed organizations.

The companies that succeed in this environment are going to be the ones that solve genuine problems, demonstrate clear value, price appropriately for customer economics, and execute operationally at a high level. The market for incremental improvements and nice-to-have features is contracting. The market for must-have solutions to painful problems is still there, but the bar for what counts as must-have just went up significantly.

UnitedHealth’s 2025 results are a signal about where health economics are headed and what that means for companies building in this market. The benign cost environment of recent years is over and we’re entering a period of sustained pressure on margins, utilization management, and care delivery economics. That’s challenging if you’re building business models that depend on expanding margins and loose utilization management. It’s an opportunity if you’re building solutions that genuinely reduce costs, improve efficiency, or enable better care at lower expense. The companies that read these signals correctly and adjust their strategies accordingly are going to be the ones that succeed over the next several years.​​​​​​​​​​​​​​​​

CMS announced a voluntary pledge from major Medicaid technology vendors to deliver $600M in savings to support community engagement implementation across state programs. The commitment involves major players like Conduent, Maximus, and Gainwell offering discounted services for eligibility verification, benefit coordination, and administrative functions. While framed as public-private partnership success, the pledge reveals structural dysfunction in Medicaid technology procurement, creates minimal incentive for innovation, and fundamentally misunderstands where value creation opportunities exist in the program. The arrangement perpetuates incumbent advantages, does little to address core technology deficiencies in state systems, and represents political theater rather than meaningful reform. For health tech investors and entrepreneurs, the announcement clarifies which problems remain unsolvable through traditional procurement and where disruptive approaches might gain traction outside legacy vendor relationships.

Table of Contents

- Introduction: Theater Over Substance

- The Pledge Mechanics: What 600 Million Actually Represents

- Why Legacy Vendors Made This Deal

- Community Engagement as Policy Graveyard

- Where the Real Medicaid Tech Problems Live

- Procurement Dynamics and Innovation Barriers

- What This Means for Startups and Investors

- Conclusion: Following the Actual Money

Introduction: Theater Over Substance

CMS rolled out a press release that probably seemed brilliant in some policy meeting about how Medicaid technology vendors would voluntarily pledge $600M in savings to states implementing community engagement requirements. The announcement hit all the right notes for people who care about announcements instead of outcomes, positioning the commitment as evidence that public-private collaboration works and that technology companies want to be good partners to states struggling with administrative modernization. The reality looks considerably less inspiring when you understand what these vendors actually committed to, why they made these commitments, and what problems in Medicaid technology infrastructure remain completely unaddressed by this arrangement.

The timing matters because community engagement requirements, the ostensible beneficiary of these pledged savings, exist in a policy environment where courts keep striking down implementations and political support swings wildly based on which administration currently runs CMS. Arkansas tried community engagement and lost something like 18,000 people from coverage before a federal court blocked the program. Kentucky got approval, then lost it, then the whole framework got rescinded under one administration and potentially revived under another. Building technology infrastructure to support a policy that might not exist in eighteen months represents exactly the kind of project that states hate funding and vendors love exploiting, which makes this pledge particularly interesting as a case study in how Medicaid procurement dysfunction works.

For investors trying to understand where opportunities exist in Medicaid technology, this announcement functions more as a map of where not to look rather than highlighting viable market entry points. The vendors making these pledges occupy positions so entrenched that they can afford to offer discounts on services that states probably don’t need for programs that might not survive legal challenge. Understanding why this makes business sense for legacy players while offering nothing to innovative startups tells you most of what you need to know about building companies in this market.

The Pledge Mechanics: What 600 Million Actually Represents

The $600M figure sounds substantial until you break down what vendors actually committed to providing. The pledges involve discounted rates on specific services related to community engagement implementation, not $600M in cash or even $600M in measured savings from reduced state spending. The math works more like vendors saying they would normally charge states X dollars for eligibility verification systems, benefit coordination platforms, and administrative support functions, but for community engagement programs they will charge something less than X, and if you add up all those potential discounts across all potential state implementations you get to $600M over some unspecified timeframe.

The participant list reads like a who’s who of companies already completely embedded in state Medicaid operations. Conduent runs eligibility systems in multiple states and processes billions in healthcare payments. Maximus operates customer contact centers and handles eligibility determination for numerous state programs. Gainwell (the former DXC Medicaid business that got sold to Veritas Capital) provides claims processing and program integrity services across a massive state footprint. These companies don’t need to acquire new customers to fulfill their pledges because they already work with the states that might implement community engagement requirements. The discount structure effectively just adjusts pricing on contract modifications or renewals that would happen anyway.

The specific services being discounted fall into categories that legacy vendors already deliver at scale. Eligibility verification systems need modifications to track work requirements, volunteer hours, or educational participation. Customer service operations need training on new program rules and potentially expanded call center capacity. Data integration work needs to happen to verify that beneficiaries meet community engagement criteria through employment records, school enrollment, or approved exemptions. All of this represents incremental additions to existing infrastructure rather than net new technology development, which means the marginal cost to vendors of providing these services runs considerably lower than the stated discount value.

The accounting gets even more creative when you consider how vendor pricing works in Medicaid contracts. States often pay based on per-member-per-month fees, transaction volumes, or percentage-of-spend arrangements that include built-in flexibility for scope changes. When a vendor commits to providing services “at reduced rates,” the baseline rate used for comparison might already include substantial margins or might represent a theoretical price that states would negotiate down anyway through normal procurement processes. The $600M savings number assumes states would pay full freight for these services absent the pledge, which represents a pretty generous assumption about how Medicaid technology procurement actually functions.

Why Legacy Vendors Made This Deal

The strategic logic for existing vendors making these commitments becomes clear once you map out their competitive positioning and relationship to state procurement processes. These companies already hold multiyear contracts worth hundreds of millions or billions of dollars across their Medicaid footprints. Offering discounted services for community engagement implementation strengthens relationships with state officials, demonstrates responsiveness to policy priorities, and creates opportunities to expand contract scopes in ways that might not otherwise get approved through standard procurement channels.

The political optics matter enormously when you operate in a market where government buyers control 100 percent of the demand. State Medicaid directors face constant pressure to demonstrate fiscal responsibility, find administrative efficiencies, and implement federal policy priorities without blowing budgets. A vendor that proactively offers savings on politically important initiatives becomes easier to defend when procurement competitors complain about contract renewals or when state legislators question sole-source agreements. The pledge functions as relationship insurance that costs vendors relatively little in actual margin but provides substantial protection against competitive threats.

The timing also coincides with broader contract renewal cycles across multiple large states. Medicaid technology contracts typically run five to seven years with extension options, which means states periodically need to either rebid major systems or exercise renewal options with incumbent vendors. An incumbent offering proactive savings on emerging policy priorities creates strong incentives for states to avoid expensive rebid processes in favor of contract modifications that incorporate the new services. This dynamic explains why vendors structured pledges around specific policy implementations rather than general cost reductions, it ties the savings to net new work that justifies contract expansions.

The competitive barrier creation matters more than the immediate financial impact. A startup trying to enter the Medicaid eligibility or claims processing market faces extraordinary difficulty displacing entrenched vendors who already integrate with fifty different state systems, employ thousands of workers with security clearances, and maintain decades of institutional knowledge about program rules. When those incumbents offer to absorb community engagement implementation costs at discounted rates, states have even less incentive to consider alternatives that would require system migrations, staff retraining, and integration work with uncertain timelines. The pledge effectively fortifies moats that already keep new entrants out of core Medicaid technology infrastructure.

Community Engagement as Policy Graveyard

The fundamental problem with building technology infrastructure around community engagement requirements comes down to the policy instability and legal vulnerability that has characterized every major implementation attempt. Arkansas launched a work requirement program that resulted in over 18,000 people losing coverage, many of whom actually met the requirements but failed to properly document compliance or navigate the reporting systems. A federal court eventually struck down the Arkansas waiver on grounds that CMS had failed to properly consider how the requirements would affect coverage levels and program objectives. Kentucky went through multiple rounds of approval and rescission before eventually abandoning implementation after years of legal challenges.

The administrative burden created by community engagement reporting falls disproportionately on populations least equipped to handle complex documentation requirements. People working irregular hourly shifts, managing chronic health conditions, or dealing with housing instability face systematic challenges in tracking volunteer hours, maintaining employment verification, or proving exemption status through the kinds of digital systems that Medicaid technology vendors typically deploy. The Arkansas experience showed that technical barriers to compliance, not actual failure to meet requirements, drove most coverage losses. Building more technology to create more documentation requirements probably makes this problem worse rather than better.

The legal framework remains fundamentally unstable because courts have consistently ruled that work requirements undermine Medicaid’s core statutory purpose of providing healthcare coverage to low-income populations. The Trump administration approved multiple state waivers for community engagement programs, the Biden administration rescinded those approvals, and future administrations might reverse course again. Technology vendors building systems to support these requirements face the prospect that their work product becomes obsolete based on election outcomes or appellate court decisions that have nothing to do with system quality or vendor performance.

States investing significant resources in community engagement infrastructure also face opportunity costs relative to other Medicaid technology priorities that might deliver more durable value. Eligibility systems need modernization to reduce churn from procedural denials. Provider payment infrastructure requires updates to support value-based contracting and alternative payment models. Data analytics capabilities lag behind what commercial payers deploy for utilization management and care coordination. All of these represent technology needs that will remain relevant regardless of which party controls federal health policy, which makes them considerably more attractive long-term investments than systems built specifically for community engagement tracking.

Where the Real Medicaid Tech Problems Live

The actual technology deficiencies in state Medicaid programs look nothing like the community engagement implementation challenges that vendors pledged to discount. States run eligibility systems built on COBOL codebases from the 1980s that require mainframe programmers to modify basic business logic. Claims processing platforms struggle to handle encounter data from managed care organizations in standardized formats. Provider enrollment processes that should take days stretch into months because verification workflows depend on manual data entry across disconnected systems. These problems cost states billions of dollars annually in administrative waste and create beneficiary friction that drives coverage churn and access barriers.

The eligibility modernization challenge alone represents a multibillion-dollar problem that has resisted decades of attempted fixes. The federal government funded substantial enhanced matching for states to replace legacy systems under the Affordable Care Act, and most implementations ran years behind schedule while delivering systems that barely met minimum functional requirements. The core issue stems from the reality that Medicaid eligibility rules vary dramatically across states, change frequently based on policy decisions, and interact with federal tax data, Social Security records, and state wage systems in ways that create enormous integration complexity. Building truly modern eligibility platforms requires both technical sophistication and deep domain expertise that few vendors possess.

The claims and payment infrastructure presents equally severe problems for states trying to manage costs and improve care quality. Managed care organizations submit encounter data in formats that don’t match fee-for-service claims structures, making it nearly impossible to build comprehensive views of utilization patterns or spending trends. States need analytics capabilities to identify high-cost members who might benefit from care management interventions, but data quality issues and system integration gaps make this analysis difficult even for states with substantial resources. The downstream effects ripple through everything from rate setting to fraud detection to policy evaluation.

Provider-facing technology creates enormous friction in Medicaid participation decisions. Doctors evaluate whether to accept Medicaid patients based partly on administrative burden, and states running enrollment systems that require manual paperwork, in-person verification, or weeks-long approval processes systematically discourage participation. Prior authorization platforms that force providers to submit clinical documentation through fax machines or web portals that log out after ten minutes of inactivity drive practices toward commercial insurance and away from Medicaid lives. Fixing these problems requires rebuilding large portions of the technology stack that connects providers to payment systems, credentialing databases, and utilization management platforms.

Procurement Dynamics and Innovation Barriers

The vendor pledge illuminates why Medicaid technology procurement systematically favors incumbent players and resists innovation from new entrants. States face enormous switching costs when considering transitions away from existing vendors because Medicaid operations cannot tolerate downtime and integration requirements span dozens of systems. A state might theoretically want to replace a claims processing platform that runs on decades-old technology, but the risk of payment disruptions, provider relations problems, or federal compliance issues makes continuing with a known-bad solution preferable to attempting migration to something potentially better.

The procurement process itself creates barriers that startups cannot easily overcome. Requests for proposals in Medicaid technology typically include requirements for vendor experience with specific transaction volumes, system certifications that take years to obtain, or incumbent knowledge that only existing contractors possess. A requirement that bidders demonstrate experience processing 50 million claims annually effectively excludes every company except the three or four vendors currently operating at that scale. States justify these requirements as risk mitigation, but the practical effect keeps new technology approaches from ever getting evaluated on technical merit.

The financial structure of Medicaid technology contracts makes venture-scale returns difficult even for companies that successfully win business. States pay based on cost-plus models, fixed-fee arrangements, or percentage-of-savings frameworks that limit upside potential and create incentive misalignment. A vendor that dramatically improves eligibility processing efficiency might get paid less under a per-transaction model, not more. Companies that reduce improper payments might negotiate small shares of documented savings, but proving attribution and sustaining those savings over multiyear contract periods creates measurement challenges that usually benefit states rather than vendors. This economic reality explains why private equity firms dominate Medicaid technology vendors while venture capital largely avoids the sector.

The regulatory compliance burden adds another layer of difficulty for companies trying to innovate in Medicaid infrastructure. Technology systems processing Medicaid data need to meet HIPAA security requirements, state-specific privacy rules, federal certification standards, and audit frameworks that vary across programs and jurisdictions. Building compliant infrastructure requires legal expertise, security investments, and operational overhead that consumes capital better spent on product development. Large vendors spread these costs across massive contract values, but startups attacking narrow problems with better technology struggle to absorb compliance expenses that might exceed engineering budgets.

What This Means for Startups and Investors

The strategic takeaway for health tech investors evaluating Medicaid opportunities comes down to avoiding head-on competition with entrenched vendors in core infrastructure and instead finding problems where better technology creates value that states or health plans will actually pay for. The vendor pledge demonstrates that incumbents own the relationship layer with state procurement officials and can offer discounts that startups cannot match. Competing for eligibility system contracts or claims processing platforms represents a losing strategy unless you somehow solve the switching cost and risk perception problems that keep states locked into existing vendors.

The more promising opportunities exist in areas where new technology enables capabilities that states currently cannot access through legacy infrastructure. Real-time eligibility verification that reduces provider abrasion and improves access probably has buyer interest if delivered through lightweight integration rather than requiring system replacement. Analytics platforms that help managed care organizations identify members needing behavioral health interventions might create enough value to justify standalone purchases outside state procurement processes. Tools that reduce member churn by improving renewal completion rates address a problem that states care about but have struggled to solve through existing vendor relationships.

The managed care channel offers better near-term prospects than direct state sales for most health tech startups. Medicaid managed care organizations operate more like commercial health plans than government agencies, with faster procurement cycles, greater willingness to pilot new technology, and clearer ROI frameworks for evaluating vendor performance. A company solving prior authorization friction or improving care coordination might sell to Medicaid MCOs using essentially the same approach that works with commercial payers, then potentially expand to state direct relationships once the technology proves out at scale. This path avoids the incumbent vendor problem and lets startups build traction before attempting to navigate state procurement complexity.

The adjacency strategy also makes sense for companies with core technology that applies across payer types but offers particular value in Medicaid contexts. Social determinants of health screening tools, housing navigation platforms, or transportation coordination services might sell to commercial plans and Medicare Advantage programs while creating disproportionate impact for Medicaid populations. Building a business model that depends on Medicaid revenue creates excessive concentration risk given procurement timelines and policy instability, but incorporating Medicaid as one customer segment within a broader payer strategy lets companies capture value from solving relevant problems without betting everything on state buying patterns.

The impact investment angle sometimes unlocks capital for Medicaid-focused health tech companies that would struggle to attract pure venture dollars. Investors willing to accept venture-like risk with below-market return expectations can back companies addressing social determinants, improving maternal health outcomes, or reducing racial disparities in care access even when the underlying business models look unattractive from a pure IRR perspective. This capital source requires founders comfortable with impact measurement frameworks and investors genuinely aligned with social mission rather than just using impact language to justify marginal deals, but it represents a viable path for companies solving real problems that traditional markets underprice.

Conclusion: Following the Actual Money

The Medicaid technology vendor pledge tells you everything you need to know about where innovation happens and where it doesn’t in public program infrastructure. Legacy vendors pledging discounted services for politically volatile policy implementations represent rational behavior within a broken procurement system that rewards incumbency over technical merit. States accepting these pledges demonstrate the constraint set they operate under when trying to modernize technology while managing enormous operational complexity and political risk. Startups largely excluded from this dynamic need to find different paths to market that avoid direct competition with vendors who own the relationship layer and can absorb margin compression that would kill venture-backed companies.

The actual money in Medicaid technology flows through multibillion-dollar contracts with states and managed care organizations that value reliability and incumbent relationships over innovation and cost efficiency. Trying to displace Conduent or Maximus or Gainwell in core eligibility and claims processing probably requires more capital and longer timelines than venture economics support, which means most health tech investors should avoid these markets entirely. The exceptions live in specific problem areas where new technology creates measurable value that someone with budget authority cares enough about to fund outside existing vendor relationships.

The broader pattern across healthcare infrastructure markets suggests that government programs systematically underinvest in technology innovation relative to private sector comparables. Commercial health plans spend billions on analytics, member engagement, and care management platforms because better technology directly improves margins through medical cost savings and administrative efficiency. State Medicaid programs operate under constraints that make long-term technology investment difficult to justify even when payback periods look attractive, because budget cycles rarely extend beyond current fiscal years and procurement processes favor known vendors over technical risk-taking. This dynamic creates persistent quality gaps between Medicaid and commercial infrastructure that technology alone cannot solve without corresponding changes to how states fund and procure systems.

For founders building in this space, the vendor pledge represents a useful reminder that good technology rarely wins on merit alone in markets where buying decisions depend more on relationships, risk tolerance, and political considerations than product quality or cost efficiency. The companies succeeding in Medicaid technology tend to either become large enough to compete directly with legacy vendors through acquisition and consolidation, or find narrow problems where better solutions create enough value that buyers can justify purchases outside traditional procurement channels. The middle ground where venture-backed startups usually operate looks particularly difficult in Medicaid infrastructure, which explains why smart investors mostly hunt elsewhere for healthcare technology opportunities.​​​​​​​​​​​​​​​​