Why Healthcare Public-Data Startups Win on the Adjudicated Entity Graph and Not the Thousand Scrapers: Ownership, Fraud, Rollup, and Distress Intelligence for Hospice, Behavioral Health & Post-Acute
đ§ 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 â Podcast).
Abstract
Cheap code generation has quietly nuked the part of healthcare data work that used to feel hard. Writing source-specific crawlers, reverse-engineering undocumented schemas, parsing ugly PDF tables, and babysitting connectors when a state website redesigns itself is now mostly a tooling problem, not a team problem. So the old pitch, âwe aggregate a thousand public sources,â is no longer a moat. Anyone can aggregate a thousand sources by next Tuesday. What stays expensive, and therefore defensible, is the adjudicated temporal entity graph sitting underneath: durable internal IDs for people, organizations, facilities, enrollments, locations, legal entities, owners, and products, all stitched across NPIs, CCNs, TINs, CLIA numbers, and state licenses, with valid-from and valid-to dates, source-document provenance, confidence scores, and a growing pile of human corrections. This essay walks through the dozen public-data product ideas worth building (ownership graphs, fraud and integrity intel, rollup radar, distress prediction, ghost-network truth, site-of-care migration, trial-site feasibility, device early warning, regulatory catalysts, workforce capacity, post-acute counterparty scoring, and environmental demand), ranks them by buyer urgency versus legal blast radius, and argues for one specific opening move. Pick a single opaque, fragmented sector such as hospice, behavioral health, home health, or skilled nursing. Answer four narrow questions about ownership, affiliation, risk, and momentum. Sell the workflow, not the dashboard. Let one buyerâs revenue fund the graph that every later buyer rents.
Table of Contents
One. The bet hiding inside âwe aggregate a thousand public sourcesâ
Two. What cheap code actually changes, and what it does not
Three. The ownership and control graph, or who really signs the checks
Four. Fraud and integrity, where willingness to pay meets the lawyers
Five. Rollup radar and distress prediction, calling transactions before the banker does
Six. Ghost networks and the rest of the public-data menu
Seven. How a moat actually forms out of deeply boring infrastructure
Eight. A build that does not collapse under its own ambition
Nine. The narrow wedge, and the order in which to ship
One. The bet hiding inside âwe aggregate a thousand public sourcesâ
There is a whole category of healthcare analytics companies whose entire reason for existing is that nobody can answer a simple-sounding question without doing two weeks of forensic accounting. Who actually owns this hospice. Is this nursing home about to fail. Has the guy applying for enrollment ever been attached to an excluded provider under a slightly different name. None of these has a single dataset that just tells you the answer, which is exactly why somebody will pay real money for the answer. The good opportunities all rhyme. The buyer is making a high-dollar decision wrapped in risk, revenue, an acquisition, a contract, or regulatory exposure. No one public file closes the loop. The underlying entities are a nightmare to reconcile because the same real-world thing shows up as a dozen NPIs, three DBAs, two prior legal names, a management company, and a real-estate LLC that exists only to hold a building. And the useful output is a narrow operational answer, not yet another general-purpose healthcare data lake nobody asked for.
The trap, and it is a comfortable trap, is to believe the work is the aggregation. It feels like work. It produces impressive slide counts. CMS alone will keep a team busy for a quarter just normalizing what it already publishes, and the published hospital ownership file runs to roughly one hundred fifty thousand rows before anyone touches a county clerkâs office. But aggregation is the part that just got cheap, and cheap things do not protect margins. The defensible asset is what happens after the raw records land: the resolved identities, the longitudinal graph with dates on every edge, the derived scores, and the accumulated record of every time a human looked at two records and decided whether they were the same thing. That correction history is the part a competitor cannot clone by pointing the same crawler at the same websites.
