Public-sector AI: 65% of UK government staff experimenting, only 30% integrated

TL;DR:

• Dr Jennifer Barth, director of research at FSP, says UK public-sector AI adoption is stuck in the gap between pilots and integration: around 65 per cent of public-sector respondents to a 2025 study with Microsoft were experimenting with AI, but only about 30 per cent had fully integrated it into how they work.
• Barth’s wider workforce data shows 41 per cent of public-sector staff feel unprepared or unsupported in using AI, and 37 per cent of leaders say they would use AI more with clearer training and safeguards.
• Resultsense view: this is the most honest framing of UK public-sector AI we’ve seen from inside the consulting community — the pilot-to-production gap is a workforce problem, not a model problem, and the implications travel directly into how Whitehall procures from the AI vendors now circling government work.

Barth argues the meaningful productivity uses of AI in government today are quiet ones: summarising large evidence packs, drafting policy briefs, analysing consultation responses, and improving triage in citizen-facing services. She frames much of this as “business-as-usual just a bit quicker” — useful, but a long way from the transformational claims used in political speeches and digital strategies.

Where the real change has to happen

The deeper opportunity, Barth says, is using AI to shorten the policy and service-design cycle — moving from periodic reviews to something closer to continuous learning, and making service delivery proactive rather than reactive. That depends on agentic AI being adopted carefully as a decision-support layer, with human judgement, accountability and empathy explicitly kept in human hands. Her operating principle is “augmentation and collaboration, not replacement”.

To get there, she calls for what she terms “workforce resilience”: properly resourced AI enablement with clear governance, role-based AI literacy curricula, playbooks for working with AI agents, a parallel track of human-skills development, and a measured change-management programme.

Why the gender gap matters for delivery

Barth makes a delivery-focused argument for getting more women into AI and data roles, rather than purely an equity one: systems reflect the perspectives of those who build them, and embedding blind spots into public-sector AI risks reproducing them in services that touch millions of citizens. She links the case for diversity to interdisciplinary design — policy, social science and technical expertise working together with each contribution weighted equally.

Looking forward

For UK departments, the honest measure of progress over the next year is not how many AI tools have been bought, but the gap between Barth’s 65 per cent experimenting and 30 per cent integrated. Closing it will demand cultural change as much as procurement. For UK SMEs selling into government — and for the consultancies now hiring deployment specialists at speed — the practical takeaway is that workforce enablement is the deliverable departments actually need, and the vendors who lead with it will outperform those leading with model demos.