On 6 August, Commure's chief legal officer sent a short email to members of the company's partnership programme. Payment agreements would be terminated within thirty days, the message said, and anything owed would be settled in full. The company, a $7 billion health software firm that sells artificial intelligence tools to medical clinics, was quietly winding down a scheme through which it had been paying customers and other parties for referrals to new business. Days later, a STAT investigation revealed what that scheme had looked like from the inside, and why the timing of its closure was unlikely to be coincidental.

Commure's pitch has always carried a certain moral clarity. Its executives describe the company as a corrective force in American healthcare, built to claw administrative revenue back from insurers and hand it to doctors. The firm set out in 2020 to build a new operating system for healthcare, with its leadership casting it as a mechanism to take money and power from industry giants and return it to physicians. That narrative has proven commercially potent. Commure now serves more than five hundred healthcare organisations, including HCA Healthcare and Tenet Healthcare, and its valuation has climbed to $7 billion on the strength of AI tools that promise to automate billing, referrals and clinical documentation.

The business model built around that promise is where the story becomes instructive rather than merely American. Internal documents examined by STAT showed the company offering thousands of dollars in compensation to clinics and other parties willing to refer its products to prospective customers, even as some existing customers reported significant financial losses tied to the deployment of those same products. Commure maintains that such referral arrangements are standard practice in technology sales and that satisfaction across its customer base remains high. Both things can be true. What the episode exposes is a structural weakness that has nothing to do with the underlying quality of the software and everything to do with how buyers evaluate it.

AI systems of this kind update constantly and behave differently across settings, which makes independent performance assessment genuinely difficult even for sophisticated purchasers. Layer a financial incentive on top of that difficulty, paid to the very people whose recommendation shapes a procurement decision, and the result is a market where enthusiasm can be manufactured faster than evidence can be gathered. That is not a uniquely American problem. It is a procurement problem, and the NHS is about to encounter it at scale.

Britain's health service is moving quickly into comparable territory. Ambient voice technology is being rolled out across trusts to cut clinical administrative burden, AI-assisted referral and pathway management tools are entering commissioning conversations, and integrated care boards are under sustained pressure to deliver efficiency savings that make automation vendors an attractive proposition. The commercial logic pulling NHS buyers toward these tools is the same logic that built Commure's customer base: administrative cost is enormous, the technology is improving, and the promised return is hard to resist.

What is less clear is whether NHS procurement is structurally equipped to test that promise before committing to it. The assurance mechanisms that exist, from MHRA oversight of software as a medical device to the Digital Technology Assessment Criteria that trusts are meant to apply, were largely designed before generative and agentic AI became a live procurement category. None of them was built with an eye to the kind of referral or introducer arrangements that shaped Commure's growth, and there is no obvious requirement that NHS bodies disclose or interrogate the financial relationships behind a vendor recommendation reaching their desk.

This need not become a story about scandal. It should become a question about process. Every health system currently weighing AI procurement is being asked to trust claims about performance that are difficult to verify independently and easy to make persuasively. Commure's termination email is a useful reminder that the companies making those claims sometimes recognise the weakness in their own sales model before their customers do. NHS leaders negotiating contracts this year would do well to ask not only what a system can do, but who was paid to tell them so.