Date: 22 June 2026

The UK public sector faces a paradox: widespread AI experimentation masks a critical implementation crisis. According to Dr. Jennifer Barth's landmark Resultsense report released this week, 65% of UK public bodies are actively experimenting with artificial intelligence—yet only 30% have successfully integrated AI into operational workflows. The result is an estimated £2.3 billion annual productivity gap as departments cycle through pilots without moving to scale.

For Chief AI Officers and senior technology leaders in government, NHS trusts, local authorities, and devolved administrations, the findings are both sobering and actionable. The gap between exploration and execution reveals systemic governance failures, workforce unpreparedness, and unclear ROI frameworks—all solvable with strategic intervention.

The Resultsense Findings: By the Numbers

Dr. Jennifer Barth's research surveyed 342 UK public sector organisations across central government, NHS, local government, and education in Q1-Q2 2026. The dataset provides the most comprehensive snapshot of public sector AI adoption since the UK AI Safety Institute's sectoral reviews in 2024.

Key metrics:

  • 65% experimenting: Two-thirds of respondents are running AI pilots or proofs of concept, typically in chatbots, document analysis, scheduling optimisation, and risk flagging.
  • 30% integrated: Only three in ten have moved AI from sandbox environments into production, serving real users and generating measurable benefits.
  • 41% workforce unsupported: Nearly half of employees lack adequate training, change management support, or clear guidance on AI tools in their roles.
  • 23% no governance framework: Almost one-quarter admit they have no formal AI governance, risk assessment, or compliance structure in place.
  • 52% struggle with data quality: Over half cite poor data infrastructure, silos, and legacy systems as the primary blocker to scaling AI projects.

These figures align with parallel findings from the UK Government's DSIT (Department for Science, Innovation and Technology) AI Sector Advisory Board, which warned in May 2026 that public sector productivity gains from AI remain 18 months behind private sector equivalents.

Why Pilots Become Graveyards: The Structural Barriers

The gap between experimentation and integration reveals five recurring failure modes in UK public sector AI deployment:

1. Lack of Executive Sponsorship and Governance

Dr. Barth's report identifies governance clarity as the strongest predictor of successful AI scaling. Organisations with a designated Chief AI Officer, cross-functional AI steering committee, and documented risk frameworks move from pilot to production 3.2x faster than those without.

Yet 58% of responding organisations reported no formal AI governance structure. Even larger bodies like NHS England trusts often operate AI initiatives as departmental side projects rather than strategic programmes. Without executive ownership, pilots remain orphaned: no funding, no prioritisation, no accountability for outcomes.

The UK government's AI assurance standards (DSIT, 2025) provide a framework, but adoption remains voluntary and patchy. CAIOs interviewed for the Resultsense report noted that public sector chief executives often view AI governance as a compliance burden rather than a competitive advantage—a misalignment that stalls progress.

2. Workforce Skills and Change Fatigue

The 41% workforce unsupport figure masks deeper cultural challenges. Many public sector workers have experienced multiple failed digital transformation initiatives. AI is often perceived as the latest buzzword, triggering scepticism rather than enthusiasm.

Training programmes, where they exist, frequently focus on tools rather than skills: teaching staff how to use a chatbot without helping them reimagine their role or workflow. Dr. Barth notes that successful organisations combine technical upskilling with change psychology—helping employees understand why AI matters and how it augments rather than replaces their expertise.

Local authorities, which employ 1.7 million people across the UK, are particularly constrained by budget cuts and hiring freezes. Even willing organisations struggle to backfill staff onto AI teams or fund structured learning programmes. The result: pilots are staffed by enthusiasts who lack the political capital to enforce adoption.

3. Data Infrastructure and Legacy System Decay

The 52% citing data quality issues highlights a less visible but critical bottleneck. UK public sector organisations operate hundreds of siloed legacy systems—sometimes 30+ years old—that were never designed to share data or train modern ML models.

For example, a local council running a benefits fraud detection AI pilot must first unify housing benefit, council tax, social care, and planning data—typically locked in separate databases with different schemas, governance, and access controls. This groundwork often costs as much as the AI project itself, and few organisations budget for it.

The UK AI Safety Institute's sectoral analysis (2025) flagged this as a sector-wide risk: public bodies lack the data infrastructure investment that private sector firms took for granted. Fixing it requires sustained capital expenditure, not pilot funding.

4. Unclear ROI and Benefit Realisation

Dr. Barth found that 67% of pilot projects lack clear metrics for success. Teams measure activities (chatbot conversations, documents processed) rather than outcomes (cost savings, faster decisions, reduced errors, improved citizen satisfaction).

This metric fog allows pilots to persist indefinitely: "We're learning" becomes the perpetual status update. Without agreed ROI targets and benefit realisation frameworks, scaling decisions become political rather than evidence-based. A £200k pilot with no clear benefits cannot easily justify £2m in production investment.

5. Siloed Innovation and Procurement Barriers

The UK public sector's fragmented structure—central government, NHS, 370 councils, 1,600+ schools, plus devolved administrations—means innovation rarely spreads. A successful AI solution in one council or NHS trust rarely diffuses to peers due to procurement rules, risk aversion, and institutional autonomy.

When the Met Police developed an AI-assisted crime prediction tool, or when NHS Grampian deployed an AI discharge planning system, those successes remained largely local rather than becoming models for replication. Procurement frameworks, written for traditional IT, favour established vendors over innovative public sector pilots.

Governance Reforms: A CAIOʼs Agenda for Unblocking Scale

Moving from pilot purgatory to production at scale requires targeted governance changes. Dr. Barth and the Resultsense team propose a four-pillar reform framework:

Pillar 1: Establish Chief AI Officer Accountability

Every public sector organisation with annual budget above £50m should have a named Chief AI Officer, reporting to the chief executive or equivalent. This role should own:

  • AI governance, risk, and compliance (aligned with ICO guidelines and forthcoming AI Act provisions)
  • Pilot-to-scale decision gates using standardised benefit criteria
  • Cross-organisational capability building and change management
  • Data infrastructure and interoperability roadmaps

The DSIT should incentivise this through:

  • Fast-track grant funding for early-adopter CAIOs (e.g., £250k for 2-year role establishment)
  • Mandatory CAIOs for any organisation seeking government AI innovation funding
  • Public recognition through annual "AI Maturity Index" benchmarking

Pillar 2: Standardise Benefit Realisation Frameworks

The UK AI Safety Institute should publish public sector-specific AI benefit frameworks, aligned with HM Treasury Green Book guidance. These should define:

  • Outcome metrics: Cost savings, speed improvements, error reduction, citizen satisfaction—not activity metrics
  • Gate criteria: Minimum evidence threshold for moving pilot to production (e.g., 80% accuracy on validation data, positive business case, staff endorsement)
  • Sunset clauses: Pilots without approved business case terminated after 12 months
  • Post-deployment evaluation: Mandatory benefit realisation audit at 6, 12, and 24 months post-launch

The Alan Turing Institute could provide analytical support, helping smaller bodies conduct rigorous pilot evaluations.

Pillar 3: Fund Data Infrastructure as Foundational

The UK government's AI strategy (2024) allocated £2.5bn to AI investment. Only £120m was directed to public sector data infrastructure. This is backwards.

A dedicated public sector data modernisation fund (suggested: £400m over 3 years) should prioritise:

  • Data lakehouse platforms enabling cross-agency analytics (modelled on Estonia's data exchange layer)
  • Data governance and quality assurance tools
  • Legacy system retirement roadmaps (many public sector systems are unmaintainable at current cost)
  • API standardisation enabling safe data sharing under GDPR and forthcoming AI Act frameworks

The ICO's latest guidance on AI and data protection (May 2026) makes clear that public sector organisations must upgrade their data infrastructure to maintain compliance. Treating this as a separate AI enabler misses the opportunity to integrate funding.

Pillar 4: Build Cross-Sector Communities of Practice

The fragmentation problem requires mechanism, not decree. DSIT should fund:

  • Public Sector AI Network: Monthly peer learning forums for CAIOs and AI leads, organised by sector (central government, NHS, local government, education, justice)
  • Open AI pilot registry: Centralised, searchable database of all public sector AI projects with outcomes, code, and lessons learned—enabling replication
  • Shared procurement frameworks: Generic ITT templates and evaluation criteria for common AI use cases (chatbots, fraud detection, resource scheduling)
  • Shared AI talent pool: A secondment programme enabling public sector organisations to borrow expertise from peers and the private sector

This mirrors the success of the Government Digital Service (GDS) in spreading digital best practice post-2010. A similar model—lightweight, practitioner-led, non-prescriptive—could unblock AI scaling.

The Cost of Inaction: Economic and Social Impact

Dr. Barth estimates that the annual productivity opportunity cost of pilot purgatory is £2.3bn across the UK public sector:

  • £890m in duplicate experimentation costs: Multiple organisations conducting similar pilots without learning from peers
  • £640m in deferred operational gains: Unscaled AI projects that could reduce processing times, error rates, or agency workload
  • £480m in opportunity cost to citizen outcomes: Delayed deployment of AI-enabled services (faster welfare assessments, better hospital scheduling, more targeted social care)
  • £290m in staff productivity loss: Change management and retraining for successive AI initiatives that fail to scale

For taxpayers, this translates to visible service delays: NHS waiting lists extended because scheduling AI remains in pilot, benefit claimants waiting longer because fraud detection AI wasn't scaled, social workers spending time on manual tasks that AI could automate.

Politically, this matters. The narrative that "public sector AI innovation is happening but not delivering" is now embedded in tech media and government accountability forums. Fixing it requires visible, measurable progress on actual deployments—not more pilots.

Forward Look: The Next 18 Months

Several signals suggest 2026 is an inflection point for UK public sector AI:

Regulatory Clarity

The AI Act becomes applicable to UK public sector bodies from August 2026 (as EU trade-dependent organisations). This removes the "wait and see" posture. Organisations must now have formal AI governance, even for low-risk pilots. This pressure, while burdensome, creates an opportunity: CAIOs can leverage regulatory compliance to justify governance investment and pilot consolidation.

Spending Review Pressure

The 2026 Spending Review (announced for September 2026) will likely include AI productivity targets for major departments. Meeting these targets requires scaling, not more pilots. This creates executive urgency.

Sectoral Leadership Emerging

The UK Health Security Agency's deployment of AI for disease surveillance (launched Q2 2026), the Home Office's AI-assisted visa processing rollout, and several council pilots in social care assessment are moving from proof-of-concept to genuine production. These successes, if properly documented and shared, become case studies that reduce scepticism and provide evidence-based confidence for scaling.

Capgemini, Accenture, and Deloitte's Public Sector AI Practices Maturing

Large consultancies are now offering public sector-specific AI playbooks, benefit realisation frameworks, and governance templates. While not a silver bullet, this commercial capability availability means public sector organisations no longer need to build from scratch.

Conclusion: From Experimentation to Excellence

Dr. Jennifer Barth's Resultsense findings confirm what many CAIOs sense intuitively: the UK public sector is remarkably innovative at experimenting with AI, but systematically weak at scaling. The 65% piloting, 30% integrated split is unsustainable—it wastes resources, demoralises teams, and delays citizen benefits.

The fix is not more technology; it is governance, clarity, and cross-sector learning. The reforms outlined above—establishing CAIOs, standardising benefit frameworks, funding data infrastructure, and building communities of practice—are politically feasible and cost-justified by the £2.3bn annual opportunity cost.

For CAIOs themselves, the message is clear: the next career milestone is not running a bigger pilot programme, but demonstrating the discipline to retire unsuccessful pilots, scale proven ones, and drive real productivity gains. The organisations that move fastest from experimentation to execution will set the benchmark for AI leadership in the UK public sector over the next decade.

The question is no longer whether public sector AI works. The question is whether we have the governance discipline to scale it.