UK government's £45B AI savings pitch built on broad-brush guesswork, MPs told (www.theregister.com)

The UK government has pitched a headline figure of £45 billion in productivity savings from AI across the public sector, but MPs were told that number rests on broad-brush assumptions rather than rigorous evidence. Nick Davies of the Institute for Government told the Commons Science, Innovation and Technology Committee that most public spending is wages or infrastructure, so achieving £45B would require either major headcount cuts or large capital savings—or else the figure is simply overstated. Government studies claim AI could free up nearly two working weeks per person and that half of office work can be helped by AI, but Davies warned those gains may be delivered within existing budgets rather than as net spending reductions. For the AI/ML community this is a reality check: demonstrable savings hinge on scalable deployments, clear definitions of “routine” tasks, robust measurement of productivity, and high-quality data and architectures—areas where many public bodies are weak. Early trials, such as a departmental Copilot pilot, showed no clear productivity uplift, and vendors like Microsoft continue to face ROI skepticism. The takeaway for practitioners and policymakers is to focus on targeted pilots, invest in data infrastructure, standardize metrics, and be transparent about whether AI is improving service value or simply redistributing existing spend.