Inside the Agentic Back Office Race for Specialty Practices: How a YC Company Hit 17x Growth and Seven Figures in ARR by Doing the Fax, Referral, Scheduling, and Collections
🎧 Part I Podcast free on Spotify.
🎧 Part II Podcast episode for paid subscribers only. Also available on Spotify.
To listen to paid episodes in Apple or Spotify, link your Substack subscription via the show settings on those platforms (instructions inside the Substack app under Subscriptions → Podcast).
Table of Contents
The pitch that quietly skips the software part
Anatomy of one referral, and why the fax pile never empties
The vise: costs up, rates down, payers automating the denials
The 605 million dollar elephant and the rest of the field
Why ARR here is not the ARR you think it is
Where this whole thing could break
The read for anyone writing checks
Abstract
Plena, a 2025 YC company, says it grew 17x and crossed seven figures in committed ARR in roughly eight to ten months running referrals, fax intake, scheduling, procedure compliance, and collections end to end for specialty practices, across six specialties and eight states, with 80% of growth from word of mouth and 40+ EMR integrations.
The pitch is not “buy our software.” It is “we do the work.” That framing is the whole story.
Macro backdrop is brutal for independents: practice opex up 11.1% from 2024 to 2025, Medicare conversion factor down 2.8% to $32.35, denial rates climbing, payers running AI to downcode and deny faster than humans can appeal.
Tennr sets the comp at the top of the category: $101M Series C, $605M valuation, RaeLM trained on 100M+ docs and 8,000+ payer rules, 10M docs/month.
Real debate is business model (services-as-software priced on labor, not seats), defensibility (integration sprawl vs EHR incumbents bolting on agents), and the exception rate, which is the metric that quietly decides whether any of this is actually cheaper.
Investor read: TAM is labor spend, not software spend, but the moat is operational and boring, and that cuts both ways.
The pitch that quietly skips the software part
Scroll healthcare X long enough and the launches blur together. Then one shows up that is worth stopping on, not because the demo is slick but because of one sentence buried in the founder story. Plena, a company that came out of Y Combinator and launched publicly in the last day or so, describes itself as the AI operating system for specialty medical practices. It says it automates the administrative workflows that keep clinics running, referrals, fax intake, scheduling, procedure compliance, and collections, end to end across the systems a practice already uses, and that in roughly ten months it grew 17x and crossed seven figures in committed ARR across six specialties and eight states, with 80% of growth through word of mouth.  The YC post on the screenshot frames it as eight months and contracted ARR, so somewhere between those two there is a real number, and either way the slope is steep.
The cofounder, Eyad Abdalla, has the resume you would build in a lab if you wanted to sell trust to doctors and engineers at the same time. Doctor’s son, engineer by trade, Waterloo then Datadog then Shopify, and his mom is a physician whose clinic became Plena’s first office and first user.  There is something funny and a little perfect about a career arc that runs through two of the better-known scaling stories in tech and lands, of all places, on the fax machine. The fax is the cockroach of American healthcare. It will outlive everyone in this essay.
Here is the line that matters, paraphrased so nobody can accuse anyone of quoting a tweet: after three years embedded inside specialty clinics, Abdalla concluded they do not need more software, they need someone to do the work, and Plena does it. Read that twice if you spend your days in health tech, because it is a quiet repudiation of the last fifteen years of the category. The dominant model was to hand a clinic a dashboard, a login, and a quarterly invoice, then call the staff time spent learning it “adoption.” Plena is selling the absence of a dashboard. The product is the outcome. The software is plumbing the customer is not really supposed to think about.
That is not a marketing trick, it is a different shape of company, and most of what follows is an attempt to figure out whether the shape holds.
Anatomy of one referral, and why the fax pile never empties
To understand why a clinic would pay someone to just make the work disappear, you have to sit with how absurd a single referral actually is. A specialty practice gets a referral by fax, portal, or email. Somebody reads it. Somebody matches it to the right patient, or creates a new one. Somebody extracts the clinical detail, checks insurance and eligibility, updates the EHR, attaches the documents, routes the task, and then chases whatever is missing, which is usually something. Plena’s own materials describe coordinators spending three to eight minutes per fax on a bucket that never empties, with 15 to 25% of referrals never getting scheduled at all.  That last number is the one that should make an operator wince. A double digit slice of demand simply evaporates inside the intake process, which means the practice paid for marketing, referral relationships, and front desk labor to generate patients it then loses to its own paperwork.
Multiply that by the dozens of distinct workflows a clinic runs every day and you get the real picture. One Plena case describes a four month, 4,800 referral backlog sitting in Epic Care Link that got cleared, with coordinators afterward processing two days of intake per shift and doing same day outreach.  A 4,800 item backlog is not a software problem in the way people usually mean it. The clinic already had software. It had Epic. The backlog existed because there were not enough human hours to feed the software, and the software did nothing on its own while it waited.
This is the gap the whole agentic back office wave is crawling into. Every year more than a third of Americans get referred to a specialist, and the process still lives in a patchwork of faxes, emails, and portals, which produces delayed care, overwhelmed staff, and missed revenue.  The technical claim Plena and its peers are making is that vision and language models can now read the messy unstructured artifacts, the scanned PDF, the handwritten note, the fax of a fax of a fax, and then take action inside the existing systems rather than dumping a structured blob into a new pane of glass for a human to retype. Plena says it reads and writes in the formats the EMR already understands, with no data migration and no parallel system, across 40 plus EMRs including the long tail like Allscripts, Practice Fusion, Office Practicum, and Intergy.  That long tail is the whole ballgame for specialty, and we will come back to why.
The detail that gives the model away as serious, not vaporware, is small. Plena notes that patients tell the clinic they think the agent is one of the staff.  If true at any volume, that is the difference between a tool and a teammate, and it is also the thing payers and compliance teams will eventually want to poke at. For now it is a clean proof that the bar has moved from “summarize this document” to “handle this human on the phone without anyone noticing.”
The vise: costs up, rates down, payers automating the denials
None of this lands without the macro, because the macro is why a busy physician would even take the meeting. Independent practices are getting squeezed from both ends at once, and the squeeze got measurably worse this year. Practice operating expenses climbed 11.1% from 2024 to 2025 while the Medicare conversion factor fell 2.8%, from $33.29 to $32.35, which is the structurally insane situation of paying more to do the same work for less money per unit of that work.  A clinic cannot cut its way out of that with a better scheduling template. It needs either more revenue per patient or radically less labor per task, and the labor market for medical front office staff is not exactly loosening.
Then there is the other side of the table, where the payers have already shown up with their own robots. This is the part of the story that deep readers will find the most interesting, because it reframes the whole thing as an arms race rather than a productivity story. Per Experian Health’s 2025 State of Claims, 41% of providers now report that more than one in ten of their claims gets denied, up from 30% three years ago, while HFMA and Kodiak data put initial denial rates near 12% in 2024 and find that up to 65% of denied claims are never reworked at all.  Sit with that 65%. Most denied revenue is not lost in a fight, it is abandoned without a fight, because the appeal costs more than the claim and a tired biller triages it into the void.
It gets sharper. Kodiak found that insurers denied more claims on clinical grounds in 2025 than in 2024, driving a 25% increase in net revenue leakage at hospitals, with payers using AI and analytics to review entire datasets instead of samples and compress audit timelines.  And the payers are not stopping at denial. Major insurers have been citing aggressive provider coding as a driver of higher 2025 and 2026 medical spend and responding with across the board downcoding strategies.  Translate that out of jargon: the plans are using automation to quietly knock a level 4 visit down to a level 3 and dare the practice to notice and contest it across thousands of claims. A human billing team cannot win that volume war. An agent that appeals every single denial with the supporting documentation attached, at zero marginal labor cost, changes the math, which is exactly why one practice administrator described that capability as finally leveling the field.
So the demand for the back office wave is not really about convenience. It is about a small business getting outgunned on automation by its largest counterparties, and reaching for the only weapon that scales the same way the other side already scaled.
The 605 million dollar elephant and the rest of the field
Plena is not the first mover here, and pretending otherwise would be silly. The company that defined investor expectations for this category is Tennr, and the gap in funding between them is enormous, which is the interesting part rather than a knock. Tennr raised a $101 million Series C led by IVP in June 2025, with Andreessen Horowitz, Lightspeed, GV, ICONIQ, Foundation Capital, and Frank Slootman in the round, founded in 2021 in New York, and by that point processing 10 million documents a month across hundreds of healthcare organizations.  That round put the valuation at roughly $605 million.  The technical pitch is a purpose built model rather than a wrapper. Tennr’s RaeLM was trained on more than 100 million deidentified healthcare documents and over 8,000 payer specific documentation requirements, so it can read scanned, handwritten, and PDF inputs and check them against payer criteria before submission.  Worth remembering the arc, because it is a tidy lesson in how this category compounds. Tennr’s first notable raise was an $18 million round from a16z in 2024 framed entirely around automating things that start with a fax.  Fax to $605 million in about fifteen months tells you how much pent up money is hunting this exact problem.
The distinction Plena is drawing, and it is a real strategic fork, is point solution versus operating layer. Plena’s framing is that unlike point solutions covering a slice of the problem, it becomes the operating layer across the whole practice, connecting the systems, teams, and workflows.  Tennr started laser focused on the referral, the highest value and most painful single workflow, and has been widening from there. Plena is claiming the wide footprint from the jump. Both bets are defensible and the honest answer is nobody knows yet which sequencing wins. Land and expand from one killer workflow is the classic SaaS playbook and it derisks early sales. Show up as the whole back office is harder to sell but stickier if you survive the install, because ripping out the thing that runs your referrals, your phones, your scheduling, and your collections is a near death experience for a clinic, and switching costs are the only moat that has ever reliably worked in healthcare software.
The rest of the field is filling in fast, and a lot of it is coming at the same labor from the revenue side rather than the intake side. There are clinical documentation and coding plays merging note generation with revenue cycle intelligence so the clean claim gets built at the point of documentation rather than reconstructed weeks later. There are ambient scribes drifting downstream into coding. There are RCM incumbents and billing platforms bolting agents onto eligibility and denials. The category does not have clean borders, which is both the opportunity, because the back office is genuinely one connected mess, and the risk, because everyone is converging on the middle from a different edge.
Why ARR here is not the ARR you think it is
This is the section the analysts should linger on, because the headline metric is doing something sneaky. When a traditional SaaS company says seven figures of ARR, it means it sold a number of seats or a platform fee, and the cost to deliver the next dollar of that revenue rounds to nothing. When a company whose entire pitch is “we do the work” says seven figures of ARR, the revenue is sitting on top of an operation that has real variable cost, because somewhere there are agents running, exceptions getting escalated, and almost certainly humans in the loop catching what the models miss. The gross margin question is therefore not a footnote, it is the entire investment thesis, and it is the thing a polished growth chart can hide for a surprisingly long time.
The broader market has a name for this now, services as software, and the bull case is enormous precisely because it is not measured against software budgets. The TAM is the labor line. A clinic does not compare Plena to its EHR subscription, it compares Plena to the salaries of the coordinators it would otherwise hire and cannot find. That is a far bigger pool, and it is why the entire enterprise software world is nervously rerating. IDC expects pure seat based pricing to be obsolete by 2028, with 70% of software vendors refactoring pricing around consumption, outcomes, or capability.  ServiceNow paid $2.85 billion for the AI agent platform Moveworks in early 2025, which is a legacy incumbent buying its way into the new model rather than waiting to be eaten by it.  The capability curve underneath is what makes the timing plausible rather than premature. Bain noted that the cost of OpenAI’s o3 reasoning model dropped about 80% in two months even as accuracy improved, which is the kind of curve that turns a workflow from “human plus app” into “agent plus API” inside a single budget cycle. 
But here is the part the cheerleaders skip. McKinsey’s framing of the near term is that agents mostly act as users of existing software, converting enterprise spend from labor to software while value accrues to whoever controls the underlying data and process.  For Plena that is encouraging, because sitting across 40 plus EMRs and actually doing the work is exactly the control position. For everyone pricing these companies, it means the revenue quality depends on how much of the labor genuinely got eliminated versus relocated. Which brings us to the uncomfortable operational truth nobody puts on a launch tweet.
Where this whole thing could break
The most clear eyed thing written about this category recently did not come from a VC, it came from practice operators, and it is worth taking seriously because it is the bear case from people with no incentive to be bearish about their own automation. One account, drawn from auditing around 1,000 eligibility checks across live physician practice environments, argues that AI often relocates administrative burden instead of reducing it, and that the pivotal metric is the exception rate, because unresolved edge cases create downstream labor that just moves the work around rather than deleting it.  If an agent handles 85% of referrals beautifully and quietly fumbles the other 15% into a new and harder to see pile, the practice has not saved money, it has changed the shape of its problem and possibly made the remaining 15% worse because the easy cases that used to keep staff in practice are gone. The honest version of every demo in this space hides inside that exception rate, and it is the first number a serious diligence process should demand and the last number any founder volunteers.
The second crack is the same one that has swallowed healthcare software companies for thirty years, and Plena is leaning straight into it on purpose. Forty plus EMRs is a flex on the website and a maintenance nightmare on the engineering roadmap, because every one of those integrations is a living thing that breaks when the vendor ships an update, and the long tail systems for specialty are exactly the ones with the worst APIs and the least incentive to keep them stable. The operating layer pitch is only as good as the day to day reliability of writing back into Practice Fusion or Intergy without corrupting a chart, and that reliability is unglamorous, expensive, and never finished.
The third crack is the incumbent question, and it is the one that should keep founders up at night more than the competitors they can see. The EHRs are not going to sit still while a startup becomes the operating layer on top of them. The large platforms have the integration that the startups are paying enormous engineering cost to fake, they have the distribution, and they are all racing to ship their own agents. The downside scenario for the whole independent agentic back office cohort is not that they lose to each other, it is that referral and intake automation becomes a checkbox feature inside the EHR a practice already pays for, the way e prescribing and patient portals did. Specialty’s fragmentation is the best defense against that, because the giants chase the big health systems first and the long tail specialty clinic is genuinely underserved, but “the incumbent is distracted” is a moat with a clock on it.
And the fourth crack is the system one, the one policy people will care about most. The Peterson Health Technology Institute has warned that under current incentives, efficiency gains from these tools accrue inside organizations rather than flowing through to lower system wide costs, and that no stakeholder is positioned to deploy AI in ways that actually reduce friction across the system.  Read in full, that is a forecast of an automation arms race where providers buy agents to fight payer agents, both sides spend more on technology, claims still get adjudicated in a millisecond duel, and the patient and the taxpayer fund both armies. A company can build a wonderful business inside that equilibrium. It is just worth naming that a thriving Plena and a less wasteful health system are not automatically the same thing.
The read for anyone writing checks
Strip the launch energy away and a few things are durably true. The pain is real, large, and getting worse on a schedule set by Medicare math and payer automation, so demand is not the question. The technical capability crossed a threshold in the last two years that makes doing the work, not just summarizing it, actually feasible, so feasibility is mostly not the question either. What is left, and what should drive the valuation, are the boring operational questions that do not fit on a slide: the true exception rate, the real gross margin once the humans in the loop are counted, the cost to keep dozens of integrations alive, and the half life of the moat before an EHR ships the same feature for free.
For investors the category framing is the trap and the prize at once. Pricing these against labor budgets makes every TAM slide look spectacular, and the services as software wave is real enough that the rerating of software multiples is not hype. But labor replacement businesses live and die on delivery quality at scale, which is an operations discipline, not a software one, and the firms that win will look more like exceptionally well run BPOs with a model in the loop than like classic high margin SaaS. That is not a criticism. It is a different animal that deserves a different comp set and a different set of diligence questions, and the funds that already know how to underwrite outsourced operations may have an edge here over the ones pattern matching to seat based SaaS.
Plena is a genuinely interesting entry because it picked the harder go to market, the whole operating layer, in the most fragmented and underserved corner, specialty independents, with the credibility of a founder who watched the problem from inside his mother’s exam room. The 17x growth and 80% word of mouth, if both hold up, suggest the work is landing in a way that makes customers do the selling, which is the only growth signal in healthcare that ever really meant anything.  Whether that becomes a $605 million comp of its own or gets absorbed into a platform it was politely sitting on top of is the open question. The pile of fax paper, at least, is not going anywhere, and somebody is finally getting paid to make it disappear instead of just selling the clinic a nicer way to look at it.
This is a sensitive enough commercial topic that the numbers above are worth double checking against the companies’ own current filings and pages before anything gets published, since launch day figures and funding rounds move fast and a few of the macro stats come from vendor research with a point of view.