Pharma, payers and providers

The most common assumption about this market is that pharma will pay for clinical expert data because pharma pays for everything else. The record does not support it.

Every disclosed pharma-AI transaction from 2024 through 2026 purchased model access, deployment, co-development of discovery workflows, or trial infrastructure. No pharma contract for expert-labelled clinical data or evaluation sets was found.

The deal wave itself is real and large. Novo Nordisk signed with OpenAI and Anthropic added Novartis's CEO to its board, both in mid-April 2026 (Investor's Business Daily). A cluster of five pharma-AI capital events between January and April 2026 reportedly exceeded $5.5B — though that total comes from an aggregator and should not be quoted onward (agentmarketcap) [WEAK]. Anthropic's Claude for Healthcare launch names Medidata as a clinical-trial data partner (Anthropic) — trial operational data, not expert judgement.

In every deal examinable at the primary-source level, the purchased item is software, compute, a connector or a corpus. Not a rubric. Not an adjudicated disagreement. Not a physician's reasoning.

#But the priced market for clinician judgement is right there, and it is enormous

Pharma does buy structured clinical review at scale. It buys it under a different name, through different vendors, for a different regulator.

Pharmacovigilance outsourcing was $7.24B in 2025 rising to $8.51B in 2026, forecast to $16.09B by 2030 (The Business Research Company, corroborated by Research and Markets). Mordor puts 2026 at $9.15B growing at 15.6% CAGR and names the incumbents: Accenture, IQVIA, ICON, Qinecsa and Oracle (Mordor).

Medical affairs outsourcing is smaller but the same shape: $3.05B in 2026 (Market Research Future), or $2.59B in 2025 on a more conservative read (Cervicorn).

Roughly $11–12B of clinician judgement is already priced, adjudicated and audited in 2026 — and none of it is AI data. These are case-level review businesses: a qualified clinician reads a case, applies a coded framework, records a determination, and the determination survives an audit trail because a regulator will one day inspect it.

#That is the right comparable, and the right competitive set

It is the right comparable because it is the only evidence at scale that anyone will pay for structured clinical review as an outsourced service. The willingness-to-pay is established. So is the operating model — case-level review, blinded double-reads, disagreement adjudication, documented reviewer credentials, full audit trail. That is not a coincidental resemblance to what an AI expert-data business would build; it is the same machine pointed at a different regulator. The FDA's January 2025 draft guidance asks AI-device sponsors for exactly these artefacts — "a description of the expertise of those performing the data annotation," the "number of participating clinicians and their qualifications," methods for "adjudicating disagreements," and inter-clinician variability statistics (FDA). Pharmacovigilance has been producing that documentation for thirty years. See The regulator wrote your product spec.

It is also the right competitive set, and that is the uncomfortable half. IQVIA, ICON and Accenture already employ, credential, schedule and audit clinical reviewers at a scale no startup approaches. They already hold the pharma relationships. If pharma ever does start buying expert-labelled AI data, the natural supplier is the vendor already doing the safety case-processing — not a new entrant with a physician mailing list. A vertical-specialist company entering here is not creating a category; it is proposing to take work from CROs on their home ground.

The gap the CROs have not closed is the one described in What actually gets sold: they know how to run a review process, and they do not build evaluations. Nobody at IQVIA has written 48,562 rubric criteria. That is the difference between processing judgement and designing the instrument that elicits it.

There is a second reason to keep this complex in view. Pharma is where the physician wage floor comes from. Market research, advisory boards and PV review have been paying practising physicians for their marginal hour for decades, at rates that a new AI-data entrant cannot undercut and expect supply — the argument set out in The physician panels. Pharma is not currently a customer for clinician judgement as AI data, but it is already the price-setter for the input, and it holds the provider databases through which physicians are recruited. A supplier here competes with pharma for the hour whether or not it ever sells pharma anything.

#Payers: a genuine negative

No payer was found purchasing expert-labelled data or evaluation sets for AI. Not one.

This is a negative finding, not an absent search result. CMS runs a Health Technology Ecosystem initiative and has published AI guidance governing its own internal use (CMS, CMS AI Guidance). An AI RFI for Medicare has been characterised as a procurement signal in commentary, but the characterisation rests on LinkedIn posts and nothing more [WEAK]. Neither CMS nor any commercial payer has been observed paying for clinician judgement as data.

That matters because payers are the second name on the same slide four. They employ armies of clinical reviewers for utilisation management and prior authorisation, they are automating those reviews fast, and they have both the money and the regulatory motive. They are simply not in this market yet. If they enter, the entry point is most likely coverage-determination evaluation — and there is no observed instance of it.

#Providers buy, but they pay in kind

Fifteen health systems signed enterprise AI deals in 2026 (Becker's). None of those deals is a purchase of clinician judgement. The direction of payment runs the other way: the health system supplies the clinicians and receives the software.

Microsoft AI's Mayo Clinic partnership is the clearest instance, and it is disclosed in a job posting rather than a press release. The Member of Technical Staff — AI Evaluations, Health role involves working with "Mayo clinicians as subject-matter experts to design high-quality tooling" and "recruiting physician raters" (microsoft.ai).

That single posting is the sharpest competitive threat in this whole section. An institutional partnership gets a lab specialty-matched clinicians and a credibility halo that no vendor can supply — Mayo's name on an evaluation is worth something a marketplace's name is not. It costs the lab a co-development relationship rather than an hourly rate. Where that structure is available, it beats a vendor on both price and prestige.

The same dynamic runs through provider-side procurement bake-offs, where hundreds of clinicians evaluate competing products on their employer's time. This is a very large volume of expert clinical evaluation labour that is structurally unpurchasable, and it depresses the addressable market by an amount nobody can measure.