‘Groundhog Day’: NHS leaders say AI success depends on learning from past digital mistakes
NHS leaders at a Global Government Forum roundtable made clear that AI has the potential to support neighbourhood health and preventative care but warned that success will depend on avoiding the mistakes of previous digital transformation programmes
At a recent Global Government Forum (GGF) roundtable, participants stressed that organisations should use AI to solve real problems in healthcare, rather than chasing new technology; that strong data, governance and processes are needed before scaling adoption; and that benefits for staff and patients – not productivity alone – should be the guiding principle.
The roundtable took place alongside GGF’s Global AI Cities event in Manchester and brought together UK digital and clinical leaders from integrated care boards, trusts, academia and central government, while also drawing on international expertise. The discussion came as the government’s 10 Year Health Plan places digital technology and AI at the heart of efforts to shift care from hospitals into communities and towards a more preventative model.
One attendee warned that the NHS was falling into the same traps that had limited previous digital transformation efforts. “I think we are in Groundhog Day” with AI, they said. “We have taken a blueprint for messing this up and we are following it to the letter.”
Despite this concern, participants pointed to AI already delivering benefits at scale, such as in stroke assessment to support faster treatment and improved patient outcomes. One participant described a long-term vision in which AI supports neighbourhood-level planning and population health management by bringing together demographic, outcomes and service data to inform decision-making.
Looking ahead, participants also identified opportunities for AI to support more informed and engaged citizens. Potential applications include helping people better understand clinical information, navigate services, and access trusted health information through digital platforms.
One said: “I can see AI has so many use cases where it can make such an impact, and it can be really, really good. The challenge we’ve got is we’re like a kid in a sweet shop and everyone’s trying little bits everywhere.”
Register for this webinar to understand the decisions shaping one of the NHS’s most significant digital transformation programmes: The NHS Single Patient Record: Turning Ambition Into Reality (Taking place on 13 August).
Solve the right problems
Participants repeatedly returned to the idea that organisations should stop procuring AI products – where “AI” is often poorly defined – and instead start by identifying the clinical or operational problem they are trying to solve. Several argued that too many organisations begin with a technology and then look for somewhere to apply it, rather than designing solutions around genuine service needs.
“I think we almost need to strip AI out… we should be procuring the capability, not the solution,” one attendee said. “If we have the right people – the people who understand the biases, the people who understand methodology – we will build a solution to the problem rather than buying a solution and then trying to retrofit it,” they added.
The discussion also challenged the idea that AI should be treated as a standalone innovation programme.
“My heart sinks when we talk about AI in healthcare as its own thing. Our governance and management of AI should be part of our core clinical governance,” one attendee said.
Framing AI in this way, participants argued, shifts the focus to the right questions: “What’s the thing we’re trying to make, do or improve, and does AI fit in this or not?”
GGF research: Based on interviews with chief digital and information officers in NHS trusts in England, ‘A fresh mandate for digital leadership in the NHS’ highlights the systemic enablers needed to deliver NHS digital transformation – from prioritising digital as a strategic profession to resetting the role of the digital centre. Find out more and download the report here.
Put people first
Much of the discussion around AI typically involves a focus on productivity and efficiency but several attendees argued that the real prize is larger. Instead of asking whether AI can make processes quicker or cheaper, organisations should instead ask “what’s best for the human?”, one argued, fearing there is a risk of getting this “completely wrong”.
Ambient voice technology (AVT), for example, is one of the government’s big digital bets. While it can transcribe patient-clinician conversations, create structured medical notes and draft patient letters, several attendees painted a more complex picture.
A particular concern was that time saved on tasks becomes converted into more throughput, rather than more time with patients. This may leave clinicians “always on” with potential implications for wellbeing and resilience. Further, participants noted that interactions that may appear inefficient are often valuable, providing opportunities for both connection with patients and recalibration for clinicians. And, they pointed out, when AVT generates summaries or documentation, someone still has to read, verify and take responsibility for them, adding an additional step into the process.
One attendee suggested a more valuable use of AVT would be to give it to patients, helping them better understand what a clinician had told them. “I’m not sure we’re pointing the tools in the right direction,” they said.
Fix the plumbing
A key theme throughout the discussion was the risk of layering new technology on top of poor systems and processes.
“If we’re using better technology to put things into a really bad system, we’re solving the wrong problem,” one said.
“We haven’t done enough around data quality to use AI with confidence,” another commented.
The government’s neighbourhood health and preventative care vision outlined in the 10 Year Health Plan requires an integrated and cross-sector approach, extending to local government, social care and beyond. Participants highlighted issues related to uneven digital approaches, data-sharing challenges and “a lack of coordination” across organisations as major obstacles.
One said: “We’re still thinking of AI in a traditional way… but how do you deploy AI where there isn’t a single bucket with all the data, and how do you train across multiple models?”
These challenges become even greater when incorporating complex data types such as genomics and medical imaging. They also argued that there is “an incongruence between policy and delivery and programme oversight”, noting that the NHS often develops “NHS solutions for NHS problems with NHS data”, which does not align with ambitions for neighbourhood-based approaches to health and care.
Participants also stressed that data quality is not simply a technical issue but also one of equity. They noted that models trained predominantly on US populations, or even UK datasets – which typically over-represent white British males – could risk reinforcing existing health inequalities at neighbourhood level.
Create the right conditions
The discussion concluded by turning to the practical changes needed to unlock AI’s potential in healthcare.
Participants argued that central government has a critical role in creating the conditions for the safe, scalable adoption of AI. One said government should play to its “USP” by “setting standards, setting policy, [and] convening communities of interest”. This role extends beyond technical standards to areas such as funding and procurement.
They questioned, for example, whether current funding and productivity expectations support long-term transformation. Funding from the Frontline Productivity Programme requires recipients to demonstrate productivity benefits in year one. That time horizon is “unhelpful”, one said.
Attendees also called for work to better understand the long-term financial implications of AI adoption. “I’m not sure we understand what the cost is in five years”, one participant said, particularly as commercial providers move towards usage-based pricing models, while in-house models also cost significant “time, money and energy” to keep running.
One participant summed the discussion up by saying: “We’re looking at this as a tech problem, it’s not; it’s a transformational change in any organisation. It’s a human change and it’s a people change. And if you’re treating it as a tech change, that’s why we keep failing because we’re just trying to put tech on top of tech instead of sitting back asking the question: what problem are we trying to solve?”
The roundtable was moderated by Andrew Besford, non-executive director and chair of the digital committee at Gateshead Health NHS Foundation Trust. Besford also authored GGF’s report titled A Fresh Mandate for Digital Leadership in the NHS, which was published in October 2025 and based on interviews with digital leaders in NHS trusts in England.
Reflecting on the discussion, Besford said: “There was real enthusiasm in the room for this discussion, but also a strong reality check. We know how big digital programmes come unstuck: starting with the technology instead of the problem, building on data we don’t yet trust, and expecting the tools to deliver the change on their own.
“None of that is a reason to hold back on AI, but it is every reason to focus on what genuinely helps the patient and the clinician.”
For information on future health roundtables and events, sign up to our newsletter.
About Sarah Wray
Sarah has over 15 years’ experience as a journalist with a specialism in the public sector and topics such as digitalisation and climate action. Sarah was formerly the editor of Cities Today and Smart Cities World, as well as a specialist video-based publication in the aerospace sector. She has also written for publications including Smart Cities Dive, Mobile Europe, Mobile World Live and Computer Weekly.
Related Posts
