UK public sector AI adoption stuck in pilot mode, leaders warn

TL;DR:

• Public sector digital leaders say that while AI pilots are widespread, very few organisations have moved to production use where AI completes tasks and resolves cases autonomously.
• Legacy systems, siloed data, and GDPR contractual negotiations that outlast the pilots themselves are blocking progress.
• Oxford University Hospitals NHS Foundation Trust and the University of London both prioritised governance frameworks before scaling, an approach they say is essential but time-consuming.

The UK government’s AI Opportunity Action Plan has raised political ambitions for AI adoption across public services, but the organisations doing the actual work say they are struggling to move beyond experimentation. That was the message from digital leaders at the ServiceNow AI Summit in London.

Aaron Neil, ServiceNow’s VP for UK public sector, summarised the gap bluntly: the technology is ready and the opportunity is recognised, but legacy systems, siloed data, and limited skills make meaningful integration difficult.

Governance before deployment

Lee Massie, head of IT at Oxford University Hospitals NHS Foundation Trust, described a deliberate approach: develop policy and governance with staff and patients first, then scale deployments. The Trust has tested AI-assisted clinical documentation, where tools transcribe clinician-patient conversations and generate structured notes. Some clinics report more time with patients, but concerns about transcription accuracy with complex medical language remain.

Richard Michel, chief information and digital officer at the University of London, took a similar path. Rather than a standalone AI strategy, the university embedded AI within a five-year digital vision and developed an AI policy framework covering teaching, research, and professional services.

Both leaders highlighted a practical bottleneck: GDPR contractual negotiations for AI pilots frequently take longer than delivering the pilot itself.

Data problems exposed

Generative AI tools are surfacing data quality problems that public sector organisations had not fully recognised. Massie noted that once AI systems can see an organisation’s data environment, they reveal issues that siloed legacy systems had hidden.

Michel drew a line between use cases where public, trusted data makes AI effective, such as helping students find courses, and sensitive areas like wellbeing services where queries must go straight to a human.

The consistent advice from both leaders: start with friction. Identify processes that already frustrate staff and apply AI there. Do not start with ambitious transformation goals. This mirrors the pattern emerging from the PAC’s questioning of government AI claims this week: the gap between what AI promises at pilot stage and what it delivers at scale remains the central challenge for UK public services.