The UK public sector is running hundreds of AI experiments but cannot point to a single one that has transformed its cost base. That is the uncomfortable conclusion from Deloitte and Re:State’s annual State of the State report, published in February 2026 and based on a survey of 5,847 UK adults and interviews with 118 senior public sector leaders.
The report’s central finding is straightforward: government needs to focus on delivery that citizens can actually see and feel. But buried within its 90 pages is a more specific story about AI adoption that should concern any organisation working with or selling to the UK public sector.
The numbers behind the confidence gap
The survey data paints a picture of a public that is anxious, dissatisfied, and deeply sceptical about technology’s role in their public services.
| Public priority | % citing as top issue | Change since 2023 |
|---|---|---|
| Cost of living | 74% | -4pp |
| The NHS | 69% | N/A |
| Immigration | 53% | +10pp |
| Jobs and growth | 46% | +2pp |
| Crime and policing | 45% | -1pp |
Critical Context: 67% of the public expect the cost of living to get worse, up twelve percentage points in three years. This is the backdrop against which every government AI investment will be judged.
Satisfaction with public services has dropped across the board since 2020. Healthcare satisfaction fell from 65% to 31%. Social care satisfaction halved from 34% to 20%. These are not small shifts; they represent a fundamental erosion of public confidence in the state’s ability to deliver basic services, let alone transform them with new technology.
What the public actually thinks about AI in government
When asked directly about AI in public services, the public is more worried than hopeful. 37% see AI as primarily a risk, compared with 23% who see it as primarily an opportunity. The remaining 25% see it as both equally, while just 2% think it will have no impact at all.
The age split is telling. Among 16-34 year olds, 31% see AI as mainly an opportunity. Among over-55s, that drops to just 14%, with 40% seeing it as primarily a risk. Women are more likely than men to see AI as a risk.
| Top public concerns about AI | % |
|---|---|
| Reduced human contact | 51% |
| Job losses from automation | 50% |
| Over-reliance reducing human oversight | 47% |
| Privacy and data security | 46% |
| Inability to handle judgements like a human | 36% |
| Top public opportunities from AI | % |
|---|---|
| Better data analysis and trend handling | 33% |
| Less admin work for staff | 33% |
| Reduced costs and efficiencies | 26% |
| Faster service and shorter waiting times | 26% |
| Reduction of human error | 24% |
Strategic Insight: The public’s top concern is not bias or ethics (34%) but the loss of human contact (51%). This tells us something about what people actually value in public services: they want to deal with other humans who understand their situation. Any AI strategy that ignores this will face a trust deficit from day one.
A thousand flowers, thin soil
The phrase that captures the AI picture among public sector leaders comes from a senior civil servant: “On Artificial Intelligence, there’s about a thousand flowers blooming. For policymakers, we need to break down the process and figure out where AI can help.”
This is not a sector that has been sitting on its hands. Ministers have allowed public bodies to explore AI themselves, devolving innovation across government. The result is hundreds of pilots and experiments, from correspondence automation to small language models used as internal encyclopaedias.
But there are three problems with this approach that the report surfaces clearly.
First, nothing has transformed the cost base. One senior civil servant put it bluntly: “None of the AI case studies turn out to be as good as the press release, and not a single one will transform government’s cost base. There’s a naive techno-utopianism in Whitehall.” This is not cynicism. It is the observation of someone who has seen the gap between what AI vendors promise and what government teams actually deliver.
Second, councils are using perhaps 5-10% of what AI can do. A council chief executive noted: “As a council, we exploit about five to ten percent of what AI can do. The question is how we shift the dial to 15, 20, 30 per cent. And it’s not the tech that’s the issue, it’s the people and the culture.” The technology is available. The organisational capacity to use it is not.
Implementation Note: The gap between AI capability and organisational readiness is the single biggest barrier to public sector AI adoption. Technology procurement without culture change is money wasted.
Third, civil servants are buying their own AI tools. A substantial number of interviewees admitted to using personal AI accounts at work. One director at a combined authority said: “I pay £20 for my own ChatGPT account and I use it all the time. My boss doesn’t like it, but are we supposed to trade in laptops for typewriters?” When your workforce is paying out of their own pocket for tools their employer will not provide, something has gone wrong with your technology strategy.
The agentic leap and the legacy wall
The most forward-looking leaders in the report identified agentic AI as the next phase. A director of a government software supplier observed: “Government is rising to the Artificial Intelligence challenge really well. Most effort is on dealing with textual artefacts which speeds up intellectual pursuits and the next leap is to get agentic to tackle working processes. But you’re going to need to deal with legacy to take advantage of that.”
This is where the report’s broader findings about government delivery connect directly to the AI story. Legacy systems and data quality issues are not new problems, but they become blocking problems when you try to deploy AI agents that need to interact with multiple systems.
Reality Check: You cannot build agentic AI workflows on top of systems that do not talk to each other. The investment in data infrastructure and legacy modernisation comes first, not after.
An agency chief executive captured the pace challenge: “The pace of change with AI is scary and exciting. We were talking in January about what agentic AI could do and it had completely changed by June.” Government planning cycles operate on years. AI capability shifts on months. A government supplier director summarised the tension: “You can’t make a five-year plan based on what AI looks like today.”
The delivery problem is the AI problem
The report’s six recommendations all orbit around one central insight: the UK public sector knows what good looks like but struggles to execute at scale. This applies to AI just as much as to housing, healthcare, or infrastructure.
| Recommendation | AI implication |
|---|---|
| Invest in citizen experience | Digital services must deliver outcomes, not just online forms |
| Supercharge major programmes | Apply existing project management knowledge to AI rollouts |
| Unleash the changemakers | Appoint CIOs at executive level to drive technology adoption |
| Update Civil Service workforce | Attract and retain people with digital and AI skills |
| Grow the Procurement Act’s potential | Fix tech procurement to consider running costs, not just upfront price |
| Be clear on AI ambition | Leaders must decide how bold they want to be and communicate that clearly |
Strategic Insight: The report recommends that government announcements about new programmes should only be made when delivery plans are in place. Applied to AI, this means no more pilot announcements without a credible plan for how successful experiments will scale.
The talent gridlock compounds the problem. The Civil Service operates a model designed to produce generalists who move between policy areas across a long career. But AI implementation needs specialists who may only stay for a few years. One senior civil servant argued: “We need to bring people in with deep expertise and accept that they’ll only work here for a few years. Instead, we’re trying to think how to attract people with digital skills for an entire career.”
Resource Reality: Government’s pay and pension structure makes it difficult to compete with the private sector for AI talent. Until the employment offer changes, the public sector will keep losing its best technologists.
What the devolved nations reveal
The Scotland, Wales, and Northern Ireland sections of the report add useful regional texture. Scotland’s leaders showed a particular zeal for reform but also acknowledged they lag England on technology investment. One senior civil servant said: “We’re lagging England on investment in technology. Instead, we employ more people, and we’re at the end of the road on the impact of doing that.”
SME Advantage: Northern Ireland’s public sector leaders were notably more positive about AI than their counterparts elsewhere in the UK, seeing it as an opportunity rather than a risk. For AI suppliers, Northern Ireland may represent a more receptive market.
Wales’s leaders expressed similar interest in AI as a basis for public sector reform, with one chief executive of a non-departmental public body noting: “The pace of change with AI is scary and exciting. We were talking in January about what agentic AI could do and it had completely changed by June. It’s the dark arts.”
Four challenges the report does not say out loud
1. The procurement system is not built for AI. The report notes that government buys technology based on upfront price rather than running costs. For AI tools that require ongoing compute, fine-tuning, and data maintenance, this purchasing model will consistently select the wrong solutions.
2. The public trust deficit may be irreversible at current pace. With 37% seeing AI as primarily risky and satisfaction with digital services barely breaking even (32% positive vs 30% negative on appointment booking), government may have a narrowing window to demonstrate that AI can genuinely improve services before public opinion hardens.
3. Shadow AI use is a governance time bomb. Civil servants using personal ChatGPT accounts to do their jobs are sending government data through systems with no oversight, no audit trail, and no data protection controls. The report notes this trend without drawing the obvious conclusion about information governance.
4. The gap between the nations will widen. Scotland, Wales, and Northern Ireland have different attitudes, different fiscal constraints, and different political incentives around AI adoption. A single UK-wide AI strategy for the public sector was never realistic, and this report shows why.
What this means for your organisation
The State of the State 2026 is ultimately a report about a public sector that knows it needs to change but is constrained by its own structures, incentives, and culture.
For technology suppliers: the message is that government wants AI that delivers measurable efficiency gains in specific, bounded processes. Grand transformation narratives are not landing. Show a pilot that reduced a specific team’s workload by a quantifiable amount, and you have an argument that will resonate.
For business leaders watching government AI policy: the public sector’s experience is a preview of the challenges any large, complex organisation will face. The technology is not the hard part. The hard part is changing how people work, what they are measured on, and how decisions get made.
The report’s concluding recommendation on AI deserves to be read in full: “Leaders need to be clear on how bold they want to be with AI. Being bold is not essential, but clarity is, because AI can require organisational change that goes beyond technology implementation.”
That is as honest an assessment as you will find in a major consulting report. The question for UK government in 2026 is whether clarity will translate into action before the window of opportunity closes.
Source: Deloitte/Re:State, “Delivery That Matters: The State of the State 2026”. Published February 2026. Research conducted December 2025 (public survey) and October 2025 to February 2026 (leader interviews).
Analysis by Resultsense — Making sense of AI in the UK.