AI productivity claims need stronger scrutiny according to Ada Lovelace Institute’s findings

The Ada Lovelace Institute has warned that AI productivity claims in the UK public sector need stronger scrutiny, as headline estimates are already shaping spending, workforce planning and public service reform.

In a policy briefing on AI and public services, the institute says UK government communications, industry reports and third-party analyses frequently present AI as a tool for cutting costs, saving time and boosting growth. It argues that stronger evidence is needed to assess whether those claims translate into public value.

The briefing notes that the UK’s 2025 Spending Review committed to ‘a step change in investment in digital and AI across public services’, informed by estimates of potential savings and productivity benefits that run as high as £45 billion per year.

Many current estimates rely on limited or uncertain evidence, the institute argues. Studies often measure first-order effects, such as time savings or cost reductions, while paying less attention to outcomes that matter for public services, including service quality, equity, citizen experience, institutional capacity and worker well-being.

The briefing also warns that productivity claims often fail to fully account for implementation costs, trade-offs, transition periods and the opportunity cost of prioritising AI investment over other public spending.

Several methodological concerns are identified in AI productivity research, including reliance on task automation models, self-reported surveys and limited triangulation across methods. The institute also highlights the growing use of large language models to assess which tasks they can perform, warning that this creates a circular dynamic in which AI systems are used to judge their own capabilities.

Headline figures can obscure mixed evidence, with productivity estimates varying widely and positive findings often receiving more attention than contradictory or null results. Industry involvement can also shape what gets researched and how results are framed, particularly when AI companies fund studies, provide tools or publish their own reports.

To improve the evidence base, the Ada Lovelace Institute calls for productivity research to reflect uncertainty, report ranges rather than single headline numbers and measure outcomes that matter for public services. It recommends more independent research, transparent methodologies, longer-term studies and measurement built into AI deployments from the start, including tracking service quality, error rates, staff well-being and citizen satisfaction.

Why does it matter?

Public-sector AI is increasingly being justified through promises of efficiency, savings and productivity growth. If those claims are based on weak or narrow evidence, governments risk making major investment and workforce decisions before understanding the real costs, trade-offs and effects on service quality.

The briefing is important because it shifts the question from whether AI can save time in isolated tasks to whether AI improves public services in practice. That includes outcomes such as fairness, reliability, staff well-being, citizen experience and institutional capacity, which are harder to measure than headline savings but central to public value.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!