What Class IIa AVT means for your organisation

The first two pieces argued that ambient voice belongs in the device regime, and described what it takes to operate there. This one is about the decision an organisation actually faces: how to appraise an AVT system responsibly, and bring it in without taking on risk you can't see. The argument is that classification is the most useful lens you have for that appraisal, for three reasons. It tells you whether the expert evaluation has already been done. It tells you where the clinical and legal risk will sit once the system is live. And it tells you whether what you're adopting can grow with you or is a dead end.

Classification tells you whether the appraisal has already been done.

The structural difficulty in evaluating healthcare AI is an expertise asymmetry. The technical capability to assess these systems sits in frontier labs and a handful of specialist teams; almost no organisation has the in-house AI staff to critically appraise a model's safety claims directly. So the practical question is not "can we evaluate this ourselves" but "who has, and can we trust their assessment." That is precisely what classification answers. A Class IIa certification means a Notified Body has evaluated the device to a standard well above any current NHS process such as DTAC, and has staked its own accreditation on that assessment. Adjacent externally-audited certifications, ISO 27001 and the rest, do the same for the other parts of the product.

Once you understand the volume of work required to attain and maintain that bar, the appraisal shortcut becomes clear: a higher classification means much of the evaluation burden you would otherwise carry has already been met, by someone whose own standing depends on having done it properly. It is the equivalent of relying on an investor's deep diligence, the accounts read, the customers interviewed, rather than repeating all of it yourself. A car safety rating works the same way; so does an insured travel product. The lesson for appraising AVT is that self-declaration tells you almost nothing, and an external certification tells you a great deal, so the first question to ask of any system is which one it actually holds.

It tells you where the risk will sit once it's live.

The second thing classification reveals is who is accountable when something goes wrong, which is the question that should worry an organisation most and is asked least. A self-declared Class I tool can be licensed and left at the door, with nobody having assumed enforceable responsibility for how it performs in your setting. The risk doesn't disappear; it quietly defaults to you. A regulated device works the other way: the manufacturer carries product liability for the device performing as intended, the same way a pacemaker manufacturer is accountable when its device is defective, and is bound to keep monitoring it in the real world for as long as it is live in your organisation. When you appraise a system, you are really appraising where that risk lands, and the classification is what tells you.

This does not remove your own duty of care, and you should be wary of any framing that suggests it might. The clinician still reviews and approves every note, and that responsibility stays with the organisation at the point of use. What a regulated structure does is split the risk honestly and hold each party to its part: the manufacturer is answerable for the device working as intended, you remain answerable for the clinical use of its output, and the Notified Body stakes its accreditation on having assessed the device with due care. Three parties, each carrying their share, is the right shape for deploying something this consequential, and it matters more, not less, as these systems grow more capable. The appraisal question is simply whether the system you're considering gives you that structure, or leaves you holding all of it alone.

It tells you whether what you adopt can grow with you.

The last consideration is the one most relevant to a multi-year decision, because it is about where what you bring in can go. Documentation is one function. A clinician does several things in an encounter: synthesise what is already known about the patient, gather new information through history and examination, formulate a diagnosis and plan, document it, and then action that plan. Scribing addresses one of those. Everything else, summarising the record, decision support, placing downstream orders, eventually even conducting parts of the history, is unambiguously a medical device under any intended-use definition: autonomous functionality producing medical outputs used directly in care.

This matters for appraisal because a non-device foundation cannot extend into those functions. The base layer still has to comprehend the encounter, reach agreement with the clinician, and act on that shared understanding. You cannot wire a non-device that listens to a device that acts, any more than you can connect a layperson's ears to a surgeon's hands and expect a safe operation, and a regulated vendor could not incorporate a non-device layer even if it wanted to, because that layer would never meet the same safety bar as everything built around it.

So a system that is not a medical device is a cul-de-sac: it does the one thing it does, and neither you nor anyone else can extend it. When you appraise AVT, you are not only choosing a scribe, you are choosing whether the foundation you adopt has anywhere to go.

So when an organisation appraises ambient voice, classification is the lens that answers the three questions that actually matter: whether the expert evaluation has already been done and by whom, where the clinical and legal risk will sit once the system is live, and whether the foundation can grow into the more capable systems that are coming. None of this is a reason to move slowly. Understood properly, a higher classification is what lets an organisation move quickly, and bring this technology in with its eyes open.