What a clinician hour costs

There is a twenty-fold spread between the floor and the ceiling for what is nominally the same act: a clinician looking at a case and saying what they think. A gamified crowd labeller on DiagnosUs earns an implied $10–25/hour [WEAK]; an expert witness averages $475/hour (Physician Side Gigs). Everything commercially interesting in this dossier is a decision about where on that ladder to buy, and the answer is not "as cheap as possible" — it is "as far below the medical degree as the artefact permits".

Two things make healthcare structurally different from every other domain in this atlas. The wage floor is set by pharma, not by AI, so you are a price-taker on supply and your margin has to come from what you do with the hour. And the state maintains your register, which no other domain can say — security has leaderboards and conference badges; medicine has a national provider database with a public API.

#The ladder

Read the top of that table as a constraint, not an opportunity. AI training work has priced itself into the same band as physician market research — roughly $100–250/hour — which is exactly where it must sit to attract a practising physician's marginal hour, because pharma has been paying that for years and will keep paying it. Mercor's cross-domain average is $81/hour (Time); its healthcare bands sit well above that, which tells you medicine is already priced as a premium vertical inside the largest marketplace.

You cannot pay a practising US physician meaningfully under $100/hour and expect supply. You do not have to. The business is below that line.

#The cheap tiers, which are the business

Sources: BLS Medical Records Specialists, Registered Nurses, Nurse Practitioners, Pharmacists, Counselors, Psychologists, Veterinarians, all May 2025; rates from Mercor and OpenTrain.

Two of these rows carry warnings. Pharmacists at 1.1–1.8x are expensive relative to what you can charge — the artefacts are clean and the margin is not (Pharmacy). And veterinarians have no observed AI-data rate at all: Mercor lists no veterinary role, a confirmed absence rather than a search failure, which means you would be setting the price rather than discovering it (Veterinary medicine).

#Verification: the thing no other domain has

This is healthcare's structural advantage and it should be stated plainly, because it is worth more than the wage arbitrage.

Identity and specialty. The NPPES NPI Registry is a free, public, API-accessible national database of every US healthcare provider (CMS API) — a mirror reports 9,360,727 records [WEAK]. It gives name, NPI, primary taxonomy code and practice address, instantly and at zero cost. Anthropic shipped an NPI Registry connector in January 2026, which means the labs already treat NPI as the canonical clinician identity primitive.

But taxonomy codes are self-attested, and NPI tells you nothing about licence status, board certification or discipline. The real stack is four steps: NPI for identity and claimed specialty → state medical board primary-source verification for an active unencumbered licence → ABMS or the relevant board (ABR, ABPath, ABMGG, ABD, ABO) for certification → subspecialty fellowship evidence where the task needs it. That four-step stack is genuinely defensible operational IP, and it is cheap.

Nurses are easier still: Nursys, the NCSBN national licensure and discipline database, offers QuickConfirm verification and e-Notify, and every state board runs an independent public lookup. The national active-RN-licence count is not retrievable as a figure [UNVERIFIED] — use the BLS employment figure of 3,465,400 instead.

Pharmacists run through NABP Verify plus fifty state boards, with NABP e-Profile IDs as stable national identifiers. Veterinarians run through AAVSB's VAULT plus state boards, with ACVR, ACVP and ACVIM for specialty status.

Coders are the cleanest pool of all, and the only one that is counted. AHIMA publishes active credential-holder totals, per credential, date-stamped to 31 December 2025: CCS 36,925, CCA 7,753, RHIT 26,128, CDIP 2,913 (AHIMA). AAPC reports "over 300,000 worldwide members" as of September 2025 across CPC, COC, CIC, CRC, CPB, CPMA and twenty-plus specialty credentials [WEAK — third-party citation; AAPC's own pages 403 to automated fetch]. Both bodies run credential-verification lookups, so quality gating costs effectively nothing per candidate.

Societies as near-censuses. Three are large enough to function as sampling frames rather than mailing lists: the AAO's 32,000 MDs represent >90% of practising US ophthalmologists [WEAK — Wikipedia, not primary], ASCO has >50,000 members across >150 countries, roughly a third international (ASCO), and AVMA counts 133,475 US veterinarians as of 31 December 2025 (AVMA) against 91,100 employed — a 42,000-person gap of licensed vets not in full-time practice, which is precisely the population that takes flexible remote work. Add ACR's >39,000 members, CAP's 20,000+, and RSNA's ~38,000 annual-meeting attendees.

#Where the numbers are missing

Three unfilled figures matter more than their obscurity suggests, and all three are cheap to obtain.

The genetic counsellor rate. The NSGC 2024 Professional Status Survey is robots-disallowed and neither US certified-genetic-counsellor headcount nor salary could be retrieved [UNVERIFIED]. This single number decides whether Genomics and variant curation is a $70/hr business producing physician-grade ACMG classifications — transformative — or a $200/hr business, which is marginal. Get the NSGC executive summaries and the ABMGG diplomate count directly.

Whether Hippocratic AI pays its clinicians, and how much. Its safety page states it "hired over 7.7K U.S. licensed clinicians to make 775K test calls" (Hippocratic), and its RWE-LLM framework describes 6,234 clinicians — 5,969 nurses and 265 physicians averaging 11.5 years' experience, evaluating 307,038 unique calls (Hippocratic). Neither document says whether they were paid. If they were, that establishes the market rate for nursing evaluation at genuine scale; if they were not, clinicians will do this work for access and credit, and the pricing model in Nursing does not survive.

No lab has ever published a rate. Both HealthBench papers say only that "all members of the physician cohort were compensated". Mercor's published bands are the only first-party price points in the sector, and they are a cost input, not a sale price — which is why the sizing at The health read is built from hours rather than invoices.

#The funnel benchmark

Plan acquisition against one number, and it is not a panel's marketing claim.

An independent 2026 study emailed 12,000 physicians drawn from IQVIA's OneKey provider database. It converted 1.2% — 149 completers across 36 states, who produced 1,156 pairwise ratings over 620 questions at 7.8 assignments each and a median 127 seconds per rating. The physicians "were compensated"; the amount is not disclosed (arXiv).

The 127-second median is the other half of the planning arithmetic. At two minutes a rating and $150/hour, a physician judgement costs about $5; a fully adjudicated three-reader item with disagreement resolution costs roughly four times that. Those two numbers, not a rate card, are what a production model should be built from.

Discard the marketing. Sermo's own top-of-funnel claim of "$5,000 to $15,000 per week" is a ceiling for a handful of members, not a typical figure; realistic totals are $1,000–2,000/month with effort, with oncologists and immunologists at $1,200–2,500 and primary care at $600–1,200 (White Coat Investor, SalaryDr). That is the income a clinician is actually choosing between, and it is the number your offer competes with — see The physician panels for why the panels holding those relationships are the real supply-side threat, and What the doctor on the other end is risking for the employment-contract risk that decides whether a recruit can say yes at all.