Ambient Voice Technology: transitioning from early enthusiasm to business-as-usual benefits

NHS Navigator Karla Richards reflects on a recent roundtable, hosted by DigitalHealth.London and the Health Innovation Network South London. The discussion brought together representatives from London’s Integrated Care Boards (ICBs), hospital trusts, and primary care with representatives from industry to look at the rise in use of Ambient Voice Technology in clinical practice.

Ambient Voice Technology (AVT) refers to AI powered tools that capture and process spoken conversations, converting speech into text. Operating under human supervision, the AI component can analyse the spoken words, provide overviews of conversations and automate tasks like writing patient clinical notes, drafting referral letters, and taking down actions.

The current landscape

As AVT continues its entry into clinical spaces, the integration of this new technology highlights the complexities of driving the shift from analogue to digital in healthcare while safeguarding quality, safety and trust. The current situation is marked by a mix of excitement and uncertainty, with a clear need to clarify regulation and safety.

Healthcare inherently operates with an accepted baseline of risk. Measured risk and accepted levels of human error are facts of daily practice, from prescribing decisions to communication breakdowns.

AVT enters this space with the obvious potential to reduce avoidable transcribing and documentation errors and reduce the cognitive load on clinicians, allowing them to focus their attention on making decisions that matter.

However, AVT also brings its own set of risks – risks where unfamiliarity understandably breeds caution. Whether this caution is fully justified is harder to quantify; one participant at our roundtable queried how often human-written patient notes are audited for accuracy or errors and wondered what audits of AI-written notes are being compared to.

Recent guidance from NHS England on ambient scribe technology (an umbrella term for related technologies including AVT) offered a timely nudge for more cautious and informed adoption. It sets out some basic considerations around implementing AVT, including advice on functionality, safety and information governance protocols.

NHS England does not typically offer guidance on individual classes of innovation. However, given the nascent state of regulation for these technologies and the lack of local organisational guidelines, the reality is that AVT use is being shaped by individual clinician preferences. Across care settings, clinicians are increasingly engaging with AVT tools independently, often without formal governance, IT oversight, or awareness from leadership. This raises important questions: Does the technology work as intended? How is benefit quantified? And critically, how are patients and their data being kept safe? This lack of clarity risks scenarios where tools are deployed beyond their capabilities or without adequate staff training, potentially undermining patient safety and clinician trust.

Primary care and secondary care differ significantly in their digital readiness and awareness of safety standards. The absence of a codified approach to AVT implementation and training leaves gaps that could be filled through standardisation efforts. As one roundtable participant put it: "We treat a large trust and a single GP the same for procurement - this approach is not helpful. The ICB doesn't have a formal role in compliance for products procured at the practice level."

Risk and regulation of AVT

As the landscape of AI-driven technologies continues to evolve, AVT is beginning to find its place within existing regulatory frameworks. The current recommendation is that AVT for use in the NHS should be at least a Class 1 medical device, which is a self-certified designation intended for medical technology products with low risk of harm to patients.

However, as capabilities grow and clinical impact deepens, the risk profile shifts, potentially warranting reclassification to Class 2a, requiring assessment from a notified body before it can be marketed for sale.

If AVT is going to deliver against its promise for patients and clinicians, the Medicines and Healthcare products Regulatory Agency (MHRA) and other bodies will need to keep pace to avoid innovations being stuck in regulatory limbo.

Efforts are underway to modernise the 2002 medical device regulation to formally include software-based technologies, yet many guidelines still need reshaping. Oversight needs to be flexible and willing to learn and adapt, with concepts like the “AI Airlock” offering room for safe experimentation and iterative learning.

Questions like post-market surveillance and oversight are also important. Can we deploy tech into clinical environments swiftly but safely, with ongoing evaluation to ensure patient care isn’t compromised?  The ambition is clear: “Get AVT into use, but in a safe and effective way.” But how do we work together to ensure agility doesn’t come at the cost of accountability and patient safety?

Accountability as AVT becomes business as usual

As Ambient Voice Technology (AVT) becomes more integrated into clinical environments, questions have been raised about who is responsible for the technology, and ultimately liable if something goes wrong.

To ensure safe and ethical deployment, it is important to determine the difference between clinical and implementation risks and the universal approach to liability. Traditionally, clinicians have been accountable for reviewing and signing off notes transcribed by administrators. Should the same standard apply to AVT-generated documentation? And how do we safeguard against overreliance?

Following current guidance, clinicians are ultimately responsible for the information they add to patient notes or referrals. However, as the use of AVT spreads and develops, should liability be shared?

Three areas of responsibility began to emerge from discussions during the roundtable:

As with anything else AVT-related, there are new nuances to contemplate regarding accountability. Who bears responsibility for making sure AVT products work well in different dialects, or if clinicians mumble? What are reasonable expectations around an AVT that can differentiate between clinical jargon or acronyms that can have drastically different meanings depending on context? There remains much uncharted territory to cover.

So far, there have been no landmark liability cases surrounding AVT – but the general feeling during the roundtable was that regardless of the checks and safeguards in place, these were certain to come.

Perhaps an interesting counterpoint raised in a recent MPS Foundation whitepaper is whether a time will come when a clinician is held accountable for not using available AI tools. The implication is clear: as AVT becomes more capable, choosing not to use it could eventually be considered negligent.

Culture change: the key to unlocking AVT's full potential

Culture is often identified as a significant barrier to the adoption of innovation in the NHS. Clinicians, in particular, are often sceptical of technologies which offer the world and deliver little beyond added administrative burden.

Our roundtable admittedly over-indexed towards enthusiastic early adopters of AVT – but their testimonies providing promising examples of clinical staff working with technology still slightly rough round the edges rather than rejecting it outright. From vocalising the physical cues of conversation to talking through every element of the consultation (rather than relying on entering notes later), there is some evidence of clinicians accepting that the benefits are worth the added effort  of adapting to AVT.

Concerns from clinicians discussed during the roundtable centred around creating an over-reliance on AVT for note-taking capabilities (an important skill still needed in situations where AVT may not be feasible). It was also acknowledged that taking typing out of consultations was a novel and potentially uncomfortable change requiring clinicians to consciously adjust their style of patient interaction.

From the patient perspective, the culture change required seems to be more linked to understanding the benefits that AVT can provide to them, while also clarifying the protection of their privacy. Early pilots show promising levels of trust on behalf of patients for the results produced by AVT. If patients are kept informed of the use and benefits of AVT, for example in appointment letters or explanations posted in waiting rooms, these levels of trust may be sustainable.

The benefits of AVT for patients are not necessarily cash-releasing or cost-saving directly, but improvements such as the standardisation of notes, patient information being presented without jargon and in an easy to understand format, and information following consultations being available immediately for patient review could help to improve engagement with treatment.

And for providers, the consensus was that those developing AVT cannot afford to rest on their laurels.  Built-in systems are needed to monitor AVT in real time to ensure data quality, accuracy and safety as well as monitoring clinician use in order to support guidance and the measuring of impact. Work needs to continue on improvements relating to environmental challenges such as noisy clinical settings, accents, multiple speakers and quiet speakers.

AVT providers promise significant benefits, and these need to be quantified for business cases. Will the use of AVT mean additional patients can be seen, or that staff can go home on time at the end of their shift? How can we quantify the improved quality of life and reduced cognitive burden on clinicians? Can the quality of notes be shown to be higher than with previous dictation methods? What will be the outcomes that are the most valuable for business cases and that will drive procurement?

Conclusion

Of the myriad potential uses of AI in healthcare, AVT seems to have established one of the firmest initial footholds.

The technology being used is not yet perfect, but the measured success of early implementations and long-term promise has bought enough goodwill among clinicians that wider spread and adoption appears sensible rather than speculative.

However, securing the long-term benefits of AVT requires funnelling some of that enthusiasm towards less immediately satisfying endeavours – ensuring that the technology is adopted in a safe and sustainable way.

Critical to this long-term future will be collaboration and a collective ability to act in the best interests of AVT as an emerging technology. Providers will sometimes need to be prepared to steer interested clinicians towards proper information governance and procurement pathways at the expense of cashing in on contracts; regulators will need to flex their rules to allow technology to continue to progress at pace.

At the Health Innovation Network South London we are proud to have played a role in facilitating some of the critical collaboration that has helped AVT start to deliver real benefits – and we’re excited to see what comes next.