After the pilot, the rollout

Training is one habit, not a module

Most vendor training is a tour of the interface, which nobody remembers and which was never the problem. The thing that changes output quality more than any product difference is whether the clinician speaks the findings they want recorded.

Research puts scribe error highest on the physical examination, which is precisely the part most clinicians perform in silence. A clinician who narrates the examination gets a better note from every product on the market, and one who does not will blame the software for a gap they created.

So teach that, then teach reading the draft before signing, then stop. Two habits, taught in ten minutes, repeated. Everything else can be learned from the interface. Our review of where the errors concentrate is the evidence to show people who think it is a formality.

The monitoring framework, which is not optional

NHS England's ambient scribing guidance asks deploying organisations to implement a monitoring framework, and describes it as ongoing audits of clinical documentation, reviews of incident reports and reviews of system performance. That framework is part of your DCB0160, and a DCB0160 with a monitoring section that was never run is a document rather than a control.

Make it small enough to actually happen. Ten notes a month, read by someone other than the author, against the same criteria the pilot used. The point is not the sample size, it is that somebody is still looking in month seven.

Watch for drift specifically. Products in this category ship model changes frequently, and a note quality that was acceptable in March is not guaranteed in September. That is also why the supplier's clinical safety case report should be updated for every material release, covered in the clinical safety standards.

An incident route people will actually use

A clinician who catches an invented finding at half past six and corrects it has done the right clinical thing and taught the organisation nothing, because nobody else will ever know it happened. The reporting route has to cost less than the correction did or it will not be used.

One line, one place, no form. What the draft said, what was true, which consultation. Read weekly, and separate the harmless from anything touching medication, dose, allergy or a result value.

If the product is a medical device, a serious incident is reportable to the MHRA through the Yellow Card scheme and the manufacturer has its own legal duty to report. Most transcription and summarisation products are not devices after the MHRA guidance of 29 July 2026, which changes the reporting route and not the value of keeping the log.

Plan for the people who quietly stop

Track usage per clinician weekly and treat a collapse as a finding rather than a data problem. The published evidence is full of wide ranges that average into a comfortable number, and a product some people have abandoned looks identical in a monthly report to one everybody uses moderately.

When someone stops, ask once and take the answer seriously. The reasons are usually mundane. It is slow to start, it gets the examination wrong, it does not fit a particular clinic type, the login times out, the headset is in a drawer. Most are fixable and none get fixed if nobody asks.

Some specialties genuinely fit this technology worse than others, and a clinician who tried properly and stopped is evidence rather than resistance. Let them stop, and note why.

Write down what would make you stop

Agree the stop rule before the rollout, for the same reason you agree acceptance criteria before a pilot. It is much harder to decide something is not working once a contract is signed and a rollout has a champion.

Reasonable stop rules look like this. The audit finds errors reaching the record that the clinician did not catch. Usage falls below a threshold across a cohort after the learning curve should have finished. The supplier ships a change that materially alters output without notice.

Stopping is not failure, and the option to stop is worth negotiating into the contract while walking away is still cheap. Ask about exit terms and the price in year two before you sign, which is covered in what to ask a scribe vendor.

Where wai sits

We would rather a rollout worked than a contract renewed, which sounds like a slogan and is mostly a statement about where the difficulty is. The second month is harder than the first and gets less attention from every vendor in this market, including us when we are not careful.

wai shows its working, so a reviewing clinician checks reasoning rather than trusting output, and every patient-affecting action waits for approval. That is what makes the monitoring above possible rather than theatrical.

Ask us what happens in month seven, not month one. Clinicians still decide what enters the record.

Frequently asked questions