Beyond AI Pilotitis: Scaling AVT Responsibly in the NHS

There is an old saying that “one swallow does not make a summer”.

The recent comments reported from NHS ConfedExpo 2026 on AI “pilotitis” and ambient voice technology are timely and important. I agree with the central point: the NHS cannot continue to run repeated pilots indefinitely where there is already meaningful evidence that technology can support clinicians, reduce administrative burden and improve patient-facing time.

However, moving beyond pilots does not simply mean buying more licences.

In my view, the real question is not whether ambient voice technology has potential. Clearly it does. Many clinicians who have used AVT, including outside large-scale NHS deployments, describe it as transformational.

The more important questions are:

Who owns the clinical risk?

Where is the audit trail?

Is the original recording retained?

How is the AI-generated content validated?

What happens when there is an error?

How does the output flow into the wider clinical documentation process?

What is the impact on medical secretaries, administration teams and existing clinical workflow?

And critically, is the Trust adopting a governed operating model, or simply deploying another point solution?

Many AVT pilots appear to be conducted in a relatively standalone way. They may demonstrate benefit at the point of consultation, but that is not the same as proving sustainable value across the full clinical documentation pathway.

If the output is not integrated into downstream workflow, document approval, distribution, coding, validation and patient safety controls, then the benefit may be narrower than it first appears.

There is also a workforce issue that should not be underestimated. Medical secretaries and administrative teams hold significant organisational knowledge. They understand consultant preferences, clinical pathways, document quality, local processes and patient communication. If AVT adoption is presented as a replacement exercise rather than a transformation programme, the NHS risks losing expertise that will be very difficult to rebuild.

AI should support people, not simply displace them.

There are also commercial and supplier-risk considerations. The AVT market has attracted many new entrants because it is one of the more accessible entry points into healthcare AI. But Trusts need to understand whether suppliers own their technology, how dependent they are on third-party LLM providers, how pricing may change, and whether the commercial model is sustainable over the long term.

Many NHS organisations have already experienced the difficulty of changing digital dictation and speech recognition systems once embedded. The same risk applies to AVT, potentially at greater scale.

So yes, the NHS should move beyond endless pilots, but national scaling must not become uncontrolled adoption.

It is also important to recognise that NHS executives and senior Trust leaders cannot simply be criticised for being cautious.

Many have experienced technology programmes that promised significant benefits but ultimately failed to deliver the expected outcomes despite considerable investment of time, money and organisational effort. In some cases, deployments have created operational challenges, increased workload or introduced new risks that required substantial remediation.

Against that backdrop, caution is understandable.

If a significant patient safety incident were ever to occur involving AI-generated clinical content, the resulting scrutiny could have consequences far beyond a single organisation. There is a real risk that confidence in AVT could be damaged, potentially slowing adoption across the wider NHS.

The same principle applies to clinicians. Whilst many clinicians who use AVT become strong advocates for the technology, widespread adoption will only occur when users are satisfied that clinical risks have been appropriately identified, mitigated and governed. Trust in the technology is just as important as the technology itself.

The recent NHS England decision to make Microsoft 365 Copilot available to up to 505,000 NHS staff is an interesting example. “The scale, financial strength and technical resources behind Microsoft may provide additional reassurance to NHS organisations However, even with a global technology provider behind the solution, successful deployment will still depend on training, governance, workforce engagement, operational redesign and effective change management.

Technology can be purchased quickly, adoption, trust and sustainable transformation take considerably longer.

Responsible AVT deployment requires:

• Clear clinical safety ownership

• Robust information governance

• Auditability and traceability

• Human-in-the-loop controls

• Integration with clinical documentation workflows

• Workforce engagement and change management

• Clear evidence of end-to-end operational benefit

• Transparent supplier accountability

• Sustainable commercial models

At Prescribe Digital, we have ambient scribe capability within our product portfolio. However, we are taking a cautious and responsible approach to the market. At this stage, we are not pursuing every AVT RFI or tender opportunity simply because the market is active.

For that reason, our immediate focus remains on the areas where Trusts are seeing clear and urgent operational need: transcription support, digital dictation, speech recognition and the reduction of clinical correspondence backlogs, while continuing to develop AI capability within a responsible governance framework.

There needs to be a more mature and level playing field, with stronger governance, clearer accountability and better understanding of what safe and sustainable AI adoption really means in clinical documentation.

The NHS should not stop innovating.

But it must adopt AI responsibly, with patients, clinicians and the wider workforce properly protected.

Without that, the risk is that today’s enthusiasm becomes tomorrow’s governance, safety and workforce challenge.