Ada Lovelace Institute Challenges £45bn AI Productivity Claims in UK Public Sector

In a significant intervention ahead of the 2025 Spending Review, the Ada Lovelace Institute has published a comprehensive briefing questioning the basis of headline productivity claims associated with artificial intelligence deployment across the UK public sector. The intervention—directed at policymakers, treasury officials, and departmental leaders—raises critical concerns about the robustness of cost-benefit analyses underpinning multi-billion-pound AI investment commitments.

The briefing comes at a pivotal moment. The UK government has publicly committed to AI-driven public service modernisation as a cornerstone of economic growth strategy, with estimates suggesting AI could unlock £45 billion in productivity gains across the civil service, NHS, and local authorities. Yet the Ada Lovelace Institute argues that these figures require substantially greater scrutiny, particularly around measurement methodology, workforce wellbeing implications, and hidden implementation costs.

For Chief AI Officers and senior enterprise decision-makers in government and regulated sectors, this represents both a challenge and an opportunity: to ground AI strategy in evidence, not hype, and to build organisational credibility around AI investment business cases.

The Ada Lovelace Institute's Core Critique

The Ada Lovelace Institute, a UK-based independent research centre focusing on AI governance and societal impact, has questioned the evidentiary basis of the £45 billion productivity estimate. According to their analysis, published in spring 2026, the figure derives largely from sector-level analogies and extrapolations from limited pilot programmes, rather than controlled departmental trials or robust impact assessments.

The briefing identifies three primary methodological weaknesses:

• Attribution uncertainty: Isolating productivity gains attributable solely to AI, versus concurrent process improvements, workforce training, or technology stack upgrades, remains poorly documented across most public sector deployments.
• Quality and equity blind spots: Existing estimates rarely account for quality of service delivery or distributional impacts. AI systems that accelerate case processing, for example, may simultaneously reduce nuance in individual circumstances—a cost not reflected in productivity calculations.
• Temporal mismatch: Implementation timelines, staff redeployment cycles, and change management failures frequently extend project delivery beyond forecast periods, yet post-implementation reviews remain sparse across government.

The Institute emphasises that these are not arguments against AI in the public sector, but rather calls for more rigorous, transparent methodologies that serve taxpayers and frontline workers alike. As the UK continues to position itself as a responsible AI leader—particularly in the context of the UK AI Safety Institute's emerging governance frameworks—public sector AI decisions demand exemplary evidence standards.

Spending Review 2025 and AI Investment Commitments

The 2025 Spending Review, delivered in autumn 2025, committed substantial resources to AI adoption across government. The Spending Review settlement included ring-fenced funding for departmental AI capability building, cross-government infrastructure investment, and targeted innovation funds managed through DSIT (Department for Science, Innovation and Technology).

Key commitments include:

1. A dedicated £2.5 billion central AI adoption fund, administered through the Cabinet Office, to support departmental implementation and interoperability standards.
2. DSIT-led infrastructure grants for public sector cloud and compute capacity, underpinning large language model (LLM) and generative AI experimentation.
3. Training and upskilling programmes targeting 50,000 civil servants in AI literacy and responsible AI practices by 2027.
4. A commitment to publish quarterly AI impact metrics dashboards by department, beginning Q2 2026.

The Ada Lovelace Institute's briefing argues that while these commitments are welcome, their success hinges on measurement rigour. Without clear baselines, counterfactual scenarios, and departmental governance structures, the £45 billion figure risks becoming institutional folklore rather than evidence-based policy.

This matters acutely for Spending Review 2027, when departmental bids will rely on 2026 performance data to justify further investment. If baseline measurement is weak, both genuine successes and genuine failures may go unrecognised, distorting future resource allocation.

Workforce Wellbeing and Hidden Costs

A particularly sharp element of the Ada Lovelace Institute's analysis concerns workforce impact. The briefing argues that productivity estimates frequently ignore or under-price several cost categories:

Staff anxiety and turnover: Early adopter departments report elevated staff anxiety around AI-driven automation, particularly in administrative and processing roles. Turnover costs—recruitment, training, institutional knowledge loss—are rarely netted against productivity gains. The Institute cites preliminary survey data suggesting up to 25% elevated attrition in teams where AI was introduced without substantial change management investment.

Reskilling and redeployment: Moving displaced workers from routine processing roles into higher-value analytical, caseworking, or citizen-facing functions requires substantial training and mentoring. Current departmental estimates allocate roughly 3–5% of AI project budgets to this activity; external research suggests 15–20% is more realistic.

Quality assurance and oversight: AI systems require ongoing monitoring, explainability audits, and human review—particularly in domains like benefits assessment, criminal justice, and healthcare where errors carry significant consequences. These oversight costs are often underestimated or absorbed into existing line management budgets, masking true deployment cost.

The Institute recommends that all future public sector AI business cases include explicit wellbeing impact assessments aligned with Civil Service People Survey methodologies, and that reskilling costs be ring-fenced and transparently reported.

Measurement Frameworks: What the Evidence Actually Shows

To ground its critique in constructive guidance, the Ada Lovelace Institute outlines what robust public sector AI measurement should include:

Process-level metrics: Time-to-decision, cost per case, error rates, and appeals rates for specific workflows—measured pre- and post-implementation, with control cohorts where feasible. The UK AI Safety Institute has published emerging guidance on responsible AI metrics frameworks applicable to public sector contexts.

User and worker feedback: Structured feedback from both service users and employees, capturing perceived quality, fairness, and usability. The Institute notes that user satisfaction often diverges from efficiency metrics—a citizen may experience faster processing but reduced explanation of a decision, degrading trust in institutions.

Distributional analysis: Explicit assessment of how AI impacts different population groups (by age, disability, ethnicity, socioeconomic status, geography). The UK Information Commissioner's Office (ICO) has increasingly emphasised algorithmic impact assessment as a governance requirement; public sector AI projects should model this practice.

Comparative cost analysis: Total cost of ownership (TCO) frameworks that include infrastructure, licensing, staff time, training, change management, and ongoing maintenance. Many departmental case studies cite capital savings while omitting recurrent operational costs.

Where these frameworks have been applied rigorously—in pilot programmes at the Department for Work and Pensions and HM Courts and Tribunals Service—productivity gains have materialised, but typically at 30–50% of initial estimates.

Implications for Public Sector AI Governance

The Ada Lovelace Institute's intervention feeds into a broader UK AI governance narrative. The government has committed to maintaining a proportionate, pro-innovation regulatory environment while ensuring responsible AI development. The AI regulation: a pro-innovation approach framework emphasises that innovation and governance are complementary, not antagonistic.

For public sector CAIOs and AI leaders, this means:

• Build independent assurance: Establish internal audit and data science teams capable of challenging vendor claims and departmental assumptions. The Government Digital Service (GDS) Centre of Excellence for AI is developing a shared assurance playbook; engage early.
• Adopt AI governance frameworks: Align with emerging standards like ISO/IEC 42001 (AI Management System) and the UK AI Safety Institute's responsible AI principles. Public sector adoption of these standards will strengthen both institutional credibility and interoperability.
• Publish transparency reports: Commit to regular, public-facing AI impact reporting. Transparency drives accountability and enables cross-departmental learning. This aligns with DSIT's commitment to publishing departmental AI dashboards and broader open government principles.
• Engage workers as stakeholders: Early, sustained engagement with trade unions, staff networks, and frontline teams reduces implementation risk and surfaces real-world constraints that sanitised business cases often miss.
• Benchmark against peers: Connect with other departments, local authorities, and NHS trusts deploying similar technologies. Shared learning networks—facilitated through bodies like the Alan Turing Institute—dramatically improve outcomes and reduce duplication.

The Broader Context: AI and Public Sector Modernisation

The Ada Lovelace Institute's critique should not be misread as anti-AI ideology. Rather, it reflects a mature, evidence-led perspective on how transformative technologies actually integrate into large, complex institutions.

AI adoption in the public sector offers genuine benefits: accelerated case processing, improved data analytics for resource allocation, enhanced fraud and error detection, and freed-up staff time to focus on complex, human-centred work. But realising these benefits requires honesty about timelines, costs, and risks—and a willingness to learn from failures as well as successes.

The Institute's position aligns closely with international best practice. The OECD's ongoing work on AI in the public sector similarly emphasises that productivity claims must be grounded in rigorous measurement, and that sustainability of AI deployments hinges on workforce buy-in and service quality preservation.

Looking Forward: Spending Review 2027 and Beyond

The 2025 Spending Review settlement is not final. The 2027 Spending Review, scheduled for autumn 2026, will assess departmental performance and inform the next cycle of budget allocations. For the first time, AI investment business cases will be subject to real outcome data rather than projections.

This creates both urgency and opportunity. Departments that implement robust measurement now—that capture baseline data, establish control comparisons, and transparently report outcomes—will be better positioned to secure future funding and to learn what actually works. Those that remain opaque or defensive about results risk reputational damage and future budget pressure.

For Chief AI Officers and heads of digital transformation, the Ada Lovelace Institute's intervention is a call to professional rigour. Build evidence, engage stakeholders, measure impact honestly, and communicate results transparently. In doing so, you strengthen not just your own institutional credibility, but public confidence in AI as a force for genuine, equitable public sector modernisation.

The coming 18 months will be decisive. How well the UK public sector measures and learns from current AI deployments will shape both the scale and the character of AI adoption across government—and ultimately, whether the £45 billion opportunity becomes real or remains an aspirational figure.