Seizing a rare opportunity: scaling ambient voice technology in the NHS

'When there’s a transformation, as there is with AVT, there’s a latent energy and enthusiasm which leaders can ride the wave of. Staff can challenge the cultural inertia.'
- National leader

Ambient voice technology (AVT) has been met with initial enthusiasm from staff across the NHS and social care. Leaders piloting AVT have emphasised that staff are excited to try a technology that could improve working conditions and patient care by reducing documentation burden and cognitive load. Following several large-scale pilots, there are now national expectations for widespread rollout in the NHS and local systems are rapidly procuring AVT tools. Currently, most users are doctors, but interest in AVT extends to nurses, pharmacists, physiotherapists and social care staff. The promised benefits span clinician experience, administrative burden, patient experience, productivity and cost. This means the market is potentially very large, as is the potential impact. 

But the NHS has been here before with promising technologies that are delivered unevenly with benefits that are at best minimal and at worse never materialise. To do better this time, AVT will need to be scaled safely and optimised in a deeply resource‑constrained environment. This long read explores how the NHS can maximise the benefits of AVT. It synthesises insights from a roundtable with patient representatives and NHS and industry experts, a review of research reports, and individual conversations with NHS staff and leaders.

Is it different this time?

During the roundtable, participants shared how past digital transformation initiatives have too often spent money on technology and not on the change process. The result – repeated across electronic patient records (EPRs), dictation software and other tools – is technology that exists within the NHS but is unevenly used, with benefits felt in isolated patches rather than nationally. This has happened partly as a result of funding and procurement processes that drive single organisational rather than whole pathway transformation. Basic performance of digital systems is not measured, and improvement methods are not routinely used, so the impact and use of new technology is assumed rather than optimised. The implication is that staff are disillusioned by the promise of technology that never materialises.  

AVT is arriving into this fragmented landscape, and deployment approaches risk repeating mistakes of the past. The market is moving fast, suppliers are multiplying, there are increasing concerns on AI sovereignty, and local procurement is running ahead of any national strategy. There is clearly now an opportunity to channel that energy to improve impact and UK AI capabilities. Roundtable participants and interview participants shared how the enthusiasm from staff makes AVT one of those rare technologies that can shift workforce confidence in what artificial intelligence (AI) can do for health care. The challenge is to learn from the past, align incentives for the present, and build the foundations for the future by leading with the change, not the technology.

What AVT does, and where the NHS actually is with it

One roundtable participant set out a 0–3 phase framing of AVT capability. Phase 0 is transcription, so limited to creating text from speech. Phase 1 is one step further, summarising the transcript text, along with basic integration into EPRs so that staff-approved summaries automatically enter into the electronic records without requiring manual copy-paste actions. Phase 2 is advisory coding and suggested actions. This is where there’s AI analysis of the transcript to provide the clinician with a set of categories (codes) that indicate key information for the consultation and suggest a list of actions (eg, scans, referrals, appointments, etc). These codes and actions are presented to the clinician for review and authorisation. Phase 3 is clinical decision support, with proactive nudges for additional information. This final phase is where more in-depth AI analysis of current and past health care interactions would occur to aid clinicians with making the decision on what to do next. Some interviewees have suggested that AVT could unlock a future of voice-based diagnostic capabilities, creating a new diagnostic tool with new ethical considerations.

Yet the NHS is far from this today. Most current AVT tools sit at Phase 1, but based on interviews with clinicians, there’s strong enthusiasm for them to do more. Many interviewees believe that Phase 2 is where significantly greater time savings start to become possible. The real opportunity is not the transcript or the summary, but the automated workflows and greater sophistication.

Emerging evidence shows that the benefits of AVT are variable depending on care setting, acoustics and technical set-up. Emergency departments and ambulances see direct improvements against priority metrics such as waiting times. An example shared during the roundtable was pre‑operative nursing teams who see substantially more patients.

Outpatients and general practice see reductions in time spent on documentation but smaller reductions in appointment duration. The time saving is often a small fraction of the appointment time, due to short appointment durations or patient complexity. But there are time savings for admin staff. Realising the benefits of AVT will require transformation that goes beyond the direct user  and instead takes a multi-profession workflow approach. While impact varies across care settings one shared benefit across settings and as described in evaluations is workplace improvement, as AVT reduces the discretionary evening and weekend work that drives dissatisfaction.

Environmental factors are equally important to effective AVT use. Work by the Centre of Digital Excellence (CoDE) in the South West, including a simulated GP practice, has shown that microphone quality, microphone placement, distance from the consultation, and background noise (construction, a crying baby, heavy rain) directly affect quality of the audio and hence the accuracy of the AVT. Indications are that accents have a minor effect for most interactions.

Despite the enthusiasm shared by many of the staff we spoke with, there are many examples of AVT not working well – for example, hallucinating medications, diagnosis, and the incorrect orientation of organs or limbs. There are also examples of staff double-running – where clinicians are using AVT alongside their existing process. There are a range of reasons for this, from outputs lacking the personal voice of the staff, to notes having superfluous detail, to the process of note-taking failing to support the clinician’s cognitive process of working through a case. A clinical researcher studying the real-world use of AVT relayed how they observed a paramedic using AVT on a clinical call, but alongside the AVT they opened a Notepad window and typed their own summary in parallel. The paramedic did not feel the AVT output reflected their voice and was overly wordy. Our analysis indicates that AVT cannot simply be dropped into existing workflows and create immediate benefits; successful implementation will need culture change, process change, and technology development all in iterative loops.

AVT is a relatively new technology and capabilities continue to advance rapidly. Despite many pilots, there is little transparency of how performance (eg, reliability, error rates, uptime, correction patterns) varies across suppliers to inform purchasing decisions. This gap in knowledge of good specification metrics could limit informed procurement and, in turn, the benefits of using AVT. When safety and effectiveness data are not standardised or transparent, there is a risk that purchasing decisions are made on the basis of cost alone with no understanding of how a tool compares to alternatives. For example, a less effective AVT could have higher error rates and lower impact, but is that represented in the cost, and what is an acceptable level of error rate? Ideally, the NHS would incentivise competition from AVT suppliers based on a minimum of accuracy and cost criteria.

Any national approach must start from a recognition of the real-world situation – one of variation, fragmentation and unclear metrics, rather than an assumed blank sheet. It also has to start from a clear view of where AVT sits in the wider AI landscape. Sovereign AI capabilities are increasingly important, and the UK has unique assets: a universal health system, a large workforce, and a growing domestic AI capability. The NHS has an opportunity to shape the market by stimulating AVT competition and preferentially building the UK’s competitive capability rather than simply reaching for what is available.

Partnering, not just purchasing

To take advantage of this AVT opportunity for NHS, patient and economic benefits, the NHS needs a different relationship with AVT suppliers. There was strong agreement across suppliers, and NHS and national leaders at the roundtable, that AVT deployment, technology development and use must be a joint endeavour between providers and suppliers.

That starts with shared risk–reward contracting using simple, relevant metrics that allow for the NHS and suppliers to realise joint benefits. The competency to develop risk–reward partnership and contracting is not well established in the NHS. There is some equivalent experience in value‑based contracting for medications, but local trailblazers such as East Lancashire Hospitals NHS Trust are beginning to build it for AVT. They have piloted an AVT tool, evaluated the benefits and captured those in the contract for measuring and sharing benefits. During the roundtable, suppliers indicated a clear willingness to work differently with the NHS; they are confident in their products and are willing to accept the risks involved.

Aligned commercial incentives are only part of the picture; the flow of money into services also needs to change. The underappreciated problem is that in situations where clinicians are seeing more patients, it can cost the organisation more but there is no additional funding. Block contracts do not incentivise more activity; getting better at care delivery means more work without corresponding reward within the existing service model. Wholesale changes to consultant or GP contracts in the short term were seen as unlikely by roundtable participants. Instead, senior leaders of hospitals suggested that the NHS Standard Contract and commissioning arrangements could be reshaped now to direct shared benefits between services and suppliers, to hold providers to account for AVT deployments through risk–reward arrangements, and to set expectations for data flow, integration, and NHS–supplier partnership behaviour. 

Holding existing EPR suppliers to account is part of this. Voluntary compliance with data‑sharing expectations has continually failed. During the roundtable, AVT suppliers reported obstruction and delays when trying to integrate with EPR suppliers, forcing clinicians into manual copy‑paste workflows that increase errors, degrade structured coding, and worsen NHS data quality. The NHS also needs to be better at selecting technology that integrates with existing technologies, creating and maintaining data standards, simplifying governance, and investing in technical staff to enable data flows. It needs to be a good partner through workforce development, leadership support, governance changes and capability development.

Metrics, contracts, and the work of change

There was a rich discussion on benefit-sharing during the roundtable and subsequent interviews with chief financial officers. For risk–reward contracting to work, the centre has to mandate a core set of technology metrics (for example, reliability, error rate, correction rate, uptime) that are automatically measured and reported. These metrics should be agreed between suppliers, NHS providers and the Department of Health and Social Care (DHSC). This transparency enables the NHS to be a more informed purchaser by being aware of where a particular AVT tool performance sits in relation to the alternatives.

The outcome metrics that contracts pay against should not aim for perfection. There will need to be realistic expectations on what is expected from the NHS and suppliers for this to work. Metrics should be care setting‑specific and good enough to get started. According to roundtable participants, the priority should be emergency departments, call centres and ambulance services, where there are direct benefits to waiting times and throughput. In general practice, the early win may be reduction of after‑hours administrative work and patient experience rather than additional appointments. In mental health, it is the ability to redesign pathways (a memory clinic patient who currently needs three appointments could reach a consultant in one). Different settings warrant different metrics, and metrics will need to evolve as the technology does. 

Procurement and finance staff need training to build outcome‑based contracting competency, drawing on local trailblazers and on evidence as it accumulates. The system won’t get the contracting right at every site – and it shouldn’t try to. The priority is to get going and iterate to keep improving. 

Contracts then need an iterative cadence: long enough that short‑term organisational benefits can be realised, short enough that the NHS can re‑contract for more complex transformational aims as the technology and confidence grow. 

Crucially, each re-contract must fund both technology and change. Past business cases directed spending to technology and assumed that change would follow but it mostly didn’t. Deployment is the start, so the funding for change must include quality improvement methodology to do the work of adapting workflows and upskilling staff. Organisational implementation alone will not unlock the full benefits of AVT. Integrated care boards and regional commissioning have a role in co-ordinating cross‑pathway transformation, and developments such as the modern service frameworks create an opportunity to embed AVT in pathway redesign from the outset. 

The enabling conditions

From the roundtable discussions and interviews, five enablers emerged that will determine whether incentives and contracts can translate into impact.

Staff capability and capacity. Digital skills remain low across the workforce, and past efforts to digitally upskill staff have not been sufficient. AVT users need baseline national training, covering both how the technology works (so they understand where it can fail) and how to use it effectively (microphone placement, what to do when accuracy drops, how to interpret confidence signals). Training has to evolve with functionality and link to the workflow adaptation. For example, how can the AVT software provide staff with a warning if the audio quality is poor and could affect the accuracy, and what should staff do in response? Procurement and finance staff need a parallel curriculum in outcome‑based contracting and holding suppliers to account, including EPR and AVT suppliers. Clinical safety, information governance and data protection are essential for safe and privacy-protected implementation of AVT. Leaders need to have the confidence and capability to ensure that a good and mutually beneficial partnership with suppliers is developed. However, there are insufficient capabilities in the NHS at present, and our analysis indicates this has been impacted by recent staff cuts. The central, regional and local structures need to invest in these capabilities.

Data flows and infrastructure. The NHS is plagued by islands of data and digital systems that do not connect, which adds workload on staff and makes patients frustrated when they need to repeat their medical history. Adding AVT without addressing this perpetuates those frustrations and inefficiencies. With every patient interaction potentially mediated through AVT, the opportunity to leap forward is significant. It could mean structured clinical coding that is interoperable across all systems as a baseline, alongside standardised data flows across digital systems built into supplier contracts. Right now, Phase 1 summarisation largely produces free text that is copy‑pasted into the record. Getting the foundations right would accelerate attempts to address the decades‑long interoperability challenge and, given the volume of data AVT captures, could in time provide a new structured data layer for the system.

Evidence and knowledge-sharing. The NHS runs many pilots and audits that generate valuable knowledge but stay siloed. Evidence generation should be guided centrally, with a core set of measures covering safety, quality, experience and effectiveness, which is automated where possible. Equally important is real‑world evidence encompassing post-deployment monitoring, observations and behavioural data on how clinicians actually use the technology. Relying on research evaluation alone risks a limited perspective of a white‑glove view of AVT impact, and misses where there might be double-running. Findings should flow through a national platform with peer networks and forums, so that practitioners learn from each other rather than re‑run evaluations in isolation.

Patient involvement. Early engagement is one of the strongest enablers. During the roundtable, a GP shared how their practice brought patients in at the start of its AVT pilot, and found that information governance concerns dissipated, clinical safety concerns were mitigated, and the resulting product was genuinely co‑produced (including in specialist areas such as paediatrics). Patients want to be informed about how AVT is used, in accessible formats (waiting‑room screens, social media, GP practice posters) rather than buried in consumer‑style terms of agreement. They want to have the option to decline AVT use when discussing sensitive topics. During a transition period, AVT will need to be a viable choice, with options to decline; services should track who opts out and why, to continue to improve.

Safety and trustworthiness. Regulation enables adoption, it doesn’t block it. This sits across technology, user interface and training. Clinical oversight is essential, but there is a clear risk towards clinicians becoming checking AVT output without meaningfully engaging. The technology can help: surfacing low‑confidence sections of a summary, flagging clinicians whose correction patterns sit outside the norm, feeding data back into providers for transparent reporting. Some suppliers confirmed that they can already measure these things. Current estimates suggest AVT clinically significant error rates of around 1%, against research suggesting that errors are present in approximately 20% of manual records, but a lower error rate does not mean no harm. One roundtable attendee recounted an AVT summary that recorded the wrong side of the patient, with obvious surgical implications. At scale, safety looks different – infrequent occurrences can become daily incidents, with one tool aggregating the errors of all its users. A collaborative approach between the NHS, regulators and suppliers – including how AVT safety reporting interacts with the Medicines and Healthcare products Regulatory Agency (MHRA) yellow card system – is needed urgently.

Beyond the consultation room

AVT is sometimes framed as a productivity tool, yet it’s more than that. The voice-to-text to summarisation functionality with integration into EPRs (Phase 1) is a bridgehead, not an endpoint. It’s important to recognise that the technology will not stand still, and neither should the system’s ambition for it.

Three opportunities are already apparent. Real‑time translation could transform care for patients who currently rely on family members or scarce interpreters, with particular impact in emergency settings; the functionality needs benchmarks and safe implementation, but the potential benefits are clear. A leap forward in patient‑facing communication clinic letters rewritten in lay language, and patient actions, delivered through the NHS App, paired with patient activation measures, can shift the relationship between patient and clinician. And the structured data layer AVT could provide has opportunities for population health approaches to the development of the nascent Health Data Research Service. This would require navigating questions about access and consent, which the system needs to think through now.

Done well, this becomes an innovation flywheel rather than a one‑off deployment. An initial focus on staff and patient experience creates goodwill; aligned incentives produce shared gains and trust; and trust unlocks the next cycle. Productivity‑focused AI reduces cost; innovation‑focused AI creates value. Getting the NHS–industry relationship right is what makes the difference.

This is not a question of whether to deploy AVT, but of where and how to realise benefits and accelerate national strategies including improving the UKs AI sovereignty. The technology is already in use, and importantly the enthusiasm from frontline staff is real. The focus should be to align incentives, build the enabling infrastructure, involve patients, ensure safety at scale, and create the conditions for an innovation flywheel. The NHS has to do something different. AVT presents that opportunity, but only if the system learns from the past, aligns incentives for the present, and builds the foundations for the future.

Recommendations

• National funding and support for AVT should encompass purchase of the technology and investing in the capabilities to change services. The funding for change should encompass a full cycle of implementation, transformation and QI optimisation of workflows all based on specific outcome- and experience-based metrics.
  

• Value-based procurement will help to enable responsible risk-sharing. The Department of Health and Social Care, the NHS and AVT suppliers should create outcome metrics for each care setting. The metrics should be available for procurement teams to inform cyclical and iterative approaches that match technology development and real-world outcomes.
  

• Government, with professional bodies, should create a dedicated strand of work to understand the impact on clinician skills and consulting practice, and procurement and finance staff expertise. This should involve should providing education, training and skills for clinicians using structured peer-based forums and networks; developing clinical safety, governance and data protection capacity and capabilities at regional level; and building procurement and finance staff expertise in risk-sharing-based procurement.
  

• The Department of Health and Social Care should convene and work with the full range of service providers to agree data and interoperability standards for responsive data sharing. These should be incorporated into the development of the single patient record and EPRs to operationalise incentives like financial penalties.
  

• The Department of Health and Social Care should engage with the public on the red lines and expectations on data sharing, choice and transparency, and co-produce any solutions with them. Staff should be supported to co-produce and share support to improve patient awareness and options.
  

• Regulators and government need to create a collaborative approach for benchmarking and rapid evidence generation. This should involve developing appropriate technological and training approaches to measure, monitor and improve the safety and accuracy of AVT, ensuring errors and safety issues are captured easily and routinely.
  

• Government, the NHS and suppliers need to work collaboratively to leverage the enthusiasm and momentum of AVT implementation to progress the sovereign AI capabilities of the UK.