UK public sector productivity has been stagnant for three decades.
Costs have risen. Demand has intensified. Performance has yet to return to pre-pandemic levels. Productivity is now the central test of fiscal sustainability and service reform. Artificial Intelligence is increasingly presented as the solution.
The economic case is substantial. The Office for Budget Responsibility estimates that effective AI adoption could unlock up to £41 billion annually. The Tony Blair Institute suggests returns of roughly £9 for every £1 invested. The opportunity is significant. But capturing it will require structural change, not incremental efficiency.
To date, adoption has focused on assistive tools: drafting copilots, document summarisation and task automation. These generate measurable time savings. But public sector productivity is defined by outcomes delivered per pound spent. Faster briefings do not reduce backlogs by default. More efficient administration does not automatically improve case resolution. Without redesigning services, system performance remains largely unchanged.
Evidence increasingly points to a different approach: Agentic AI. These are autonomous, goal-oriented systems capable of coordinating workflows across data, tools and teams. Rather than improving isolated tasks, they operate across services. The shift is from individual efficiency to institutional performance.
This report sets out what that shift requires in practice. Drawing on primary interviews and wider research, we introduce a four-phase Agentic Blueprint for Public Sector Productivity — a structured approach to deploying agentic AI with clarity, accountability and measurable impact. The core question is not whether AI can save time. It is whether it can materially improve how public services deliver value.
Ready to go further? Join our live webinar where Elsewhen's Nadav Mordechai and Gabriel O'Brien walk through the research behind this paper and share a practical roadmap for agentic AI adoption in public services. You will hear directly from the teams behind the TRIAL framework, explore what a four-phase implementation looks like in practice, and get answers to your questions live.
