In August last year, we published a blog post setting out the Prime Minister’s ambition for a portfolio of AI Exemplar projects across government, and why a test‑and‑learn approach is needed to accelerate responsible AI adoption in public services. At that point, departments were beginning delivery and the focus was on experimentation: scanning for opportunities, piloting emerging uses of AI, and learning quickly from real projects.  

Ten months on, this blog reflects on what we have learned from the AI Exemplars programme, and how those lessons have directly shaped our move towards a smaller number of large, transformational AI projects delivered in close partnership with departments. 

A portfolio approach to learning 

From the outset, the Prime Minister’s AI Exemplars programme was designed as a test‑and‑learn portfolio.  The Exemplars explore where AI can add genuine value in government transformation and the barriers to scaling and help us collectively learn how best to support AI innovation.   

Central support from GDS and DSIT focused on helping departments accelerate delivery and on surfacing learning from across the portfolio, with the aim of generating practical evidence that could inform future decisions about how government should adopt and use AI at scale. 

What worked well 

The programme demonstrated that departments can identify meaningful AI use cases and pilot them effectively. Across the portfolio, teams progressed projects through delivery stages, tested tools in live environments, and began to demonstrate public value – from supporting frontline decision‑making to improving productivity and service outcomes.  

AI tools in probation services have already summarised over 150,000 meetings, reducing administrative burden for frontline staff, while the award winning NHS AI Diagnostic Fund has supported millions of scans, helping to reduce turnaround times for critical diagnostics. 

The portfolio approach also showed the value of collaboration across government. Bringing together teams working on different use cases made it easier to share insights, compare approaches and develop a clearer picture of common challenges.  

Central support added value where it focused on shared needs, such as evaluation and impact measurement, AI assurance and safety, legal and commercial considerations, and adoption planning. 

What we learned needed to change 

The Exemplars programme has been equally valuable in highlighting the limits of scaling AI through individual pilots alone. 

While many projects showed promise, scaling beyond pilots proved slow and resource intensive. Teams often encountered similar barriers: difficulties evidencing value early on, commercial and procurement constraints and data readiness challenges. 

The programme also highlighted duplication across government – not just in solutions, but in effort. Similar work was repeatedly being done in parallel across departments, making it harder to realise benefits at scale and limiting the impact of central support when spread across a large number of individual projects. 

Key lessons from the Exemplars programme 

Taken together, a year of delivery across the AI Exemplars portfolio has sharpened our understanding of what effective support for public sector AI adoption looks like: 

• Test‑and‑learn portfolios are an effective way to explore new opportunities, but they need to be designed to inform what comes next. 
• Departments can and do pilot AI successfully, but scaling is most often constrained by system‑level barriers. 
• Central action adds greatest value when it tackles those shared barriers once, rather than repeatedly at project level. 
• A smaller number of large, mission‑driven programmes, delivered in close partnership with departments, are more likely to deliver transformational outcomes than many standalone pilots. 

How the learning is shaping the next phase 

What is changing is how central effort is focused. Building on what we’ve learned, we are shifting towards a smaller number of transformational AI projects delivered in partnership with departments, alongside a stronger focus on addressing the barriers that prevent AI from being adopted and scaled across public services. 

The influence of the Exemplars programme is already visible in areas of government. Increasing our ambition from the Exemplars, we recently announced work on AI in education, including the development of a national AI tutoring programme designed to support teaching and learning at scale. Exemplar projects have acted as stepping stones, providing important foundations for this next ambitious phase. 

We are also seeing this next phase in action in healthcare. The government has announced new investment to scale AI diagnostic tools across the NHS, helping clinicians analyse scans more quickly and supporting earlier diagnosis and treatment for patients. This builds on the success of the AI Diagnostic Fund and demonstrates how exemplar learning is being applied at national scale. 

Looking ahead 

The Prime Minister set out a clear expectation that government should move quickly, learn from experience and scale what works. The AI Exemplars programme has played a critical role in building the evidence base needed to do that well. 

As we move into the next phase, the lessons from Exemplars will continue to shape how government approaches AI adoption; with a sharper focus on where central action can unlock the greatest impact, and on supporting departments to deliver transformational outcomes for the public.