The measurement my colleagues are celebrating reports in February 2027. The rollout finished long before that

Every Friday I argue against the house. Today the house has decided to be cheerful, which makes the job easier.

The lead says the NHS is doing what nobody else does: measuring what AI changes about a job, with the method published first. That is true and I have four problems with what we are making of it.

The clock is the wrong way round

Phase 2 collects data until January 2027 and reports in February 2027. NHS England announced the national rollout on 4 July 2026. The tools are being deployed now, on the strength of a supplier-partnered study, and the independent answer arrives after the deployment is done.

Measurement that lands after the decision is a record, not a check. We have made this exact criticism of government when a review reported after a policy had shipped. It does not become a triumph because we like the institution.

Independent of the vendor is not independent

The evaluation is funded by a health research body, works with NHS England, and selected its four sites partly for their access to routine data and their compliance with NHS England requirements. Every one of those choices is defensible and every one of them means the study sits inside the system that bought the technology.

It is independent of the seller. It is not independent of the buyer. Those are different things, and this paper is usually the first to say so.

The deliverable tells you whose question it is

One of the promised outputs is a freely available return-on-investment tool that NHS organisations can run on their own numbers. Think about who that is for. It is for a trust deciding whether to buy.

Where released time goes sits inside the productivity strand, as one example in one bullet. I am glad it is there. I do not believe a study whose headline deliverable is a purchasing calculator will make the clinician's working day its answer.

We are grading against ourselves

Legal services look negligent this week because we spent three weeks looking at legal services. The NHS looks exemplary because we spent one evening looking at the NHS. That is not a comparison, it is a difference in how long we looked.

Before we print that health is ahead, somebody should check whether the same machinery exists in education, policing or the courts, where the state is also the buyer. If it does, our story is about public procurement rather than about the NHS being unusually honest.

What would change my mind?

Publish anything before February 2027. An interim note on whether released time is showing up as throughput, as shorter shifts or as nothing.

Measure the end of the shift. Documentation done after hours is in the phase one evidence, so the data exists. If unpaid overtime falls, the clinician got the time. If patients per shift rises and unpaid overtime does not move, the employer did.

And name one site where the answer came back as a shorter working day. Not a model of one. A trust.

Until then, my colleagues have found something better than what exists in law, which is a low bar and worth clearing. They have not found a measurement. They have found a measurement scheduled.