Public sector AI & consulting intelligence
The AI consulting trap: what the UK government actually spends on AI
The UK has promised to rewire the state with AI. The contract record shows £147M of visible building in seven years — under 5% of a single year's consulting bill — and a classification system so leaky the National Audit Office says government cannot measure its own spend.
Skim · Built on Skim's live UK procurement database8 June 202611 min read
Original Skim analysis of UK public contract and award notices on the Find a Tender Service and Contracts Finder, 2018–2025. AI-delivery and generative-AI contracts were identified using a combination of CPV codes, contract titles and descriptions; ambiguous matches were excluded. Third-party figures for consulting spend and supplier concentration are drawn from Tussell and the National Audit Office and attributed inline.
The promise and the procurement record
The UK government has promised to rewire the state with artificial intelligence. In January 2025 it endorsed all 50 recommendations of the AI Opportunities Action Plan and committed to a "scan, pilot, scale" push to transform public services. The speeches are loud and the ambition is real.
The procurement record tells a quieter story. Every contract a public body signs leaves a trace — a buyer, a supplier, a value, a date, and a code describing what was bought. Skim analysed those traces on Find a Tender and Contracts Finder for the period 2018 to 2025. What we found is not a tidy ratio of talk to action. It is that the government's AI spending is mostly invisible — and the slice we can actually see is dominated by advice about AI rather than AI itself.
The most striking part is that you don't have to take our word for the invisibility. In November 2025 the National Audit Office reached the same conclusion, finding that government cannot reliably measure what it spends on consultants at all.
£147M
Skim analysis of Find a Tender + Contracts Finder
~4–5%
Skim analysis · Tussell
1 of 8
Skim analysis of published award notices
How much does the UK government spend on building AI?
Strip out the advice and look only at contracts tagged as genuine delivery — building models, deploying systems, and running the unglamorous machine-learning operations (ML-ops) that keep them alive. Across 2018 to 2025, Skim's analysis finds 155 such awards worth about £147 million in total. Five buyers account for roughly two-thirds of it.
| Buyer | Tagged AI-delivery work |
|---|---|
| Department for Education | Top-five buyer |
| Department for Work & Pensions | Top-five buyer |
| Ministry of Defence | Top-five buyer |
| Home Office | Top-five buyer |
| DSIT | Top-five buyer |
To put £147 million in context: Tussell, the procurement-data firm the NAO itself cites as a recognised source, estimates UK public-sector management-consultancy spending at roughly £3.2 billion in 2024/25, with broader "consulting" definitions reaching £3.7 billion. In other words, the entire seven-year visible footprint of the state building AI is roughly 4–5% of a single year's consulting bill.
Meanwhile, the wider AI economy is booming on paper: the government's own tracker reports UK AI firms raised billions of pounds of private investment in 2025. The money to think about AI, and to invest in AI companies, dwarfs the money the state spends building AI for itself.
Why you can't measure UK government AI spending
Here is the structural problem. Every contract is stamped with a CPV code — Common Procurement Vocabulary — the official taxonomy a buyer uses to declare what it bought. CPV has codes for IT services (72000000), software (72200000), and business and management consultancy (79000000). It has no dedicated code for AI. So when a department buys AI work, it reaches for the nearest plausible drawer — usually generic IT.
This isn't a fringe observation. The National Audit Office's November 2025 report found that government does not have consistent data on how much individual departments are spending on external consultants, partly because of different definitions of what counts as consultancy and difficulties in classifying the services that consultants deliver. The independent Open Contracting data registry that publishes the UK feed goes further, warning that some contract values appear to be misreported and that the data contains extreme outliers. When the official auditor and the open-data stewards both say the numbers can't be trusted at face value, the measurement problem is not Skim's caveat — it is the finding.
Vocabulary laundering: how "AI strategy" disappears into the filing system
This is the cleanest, most verifiable pattern in the data, and it survives contract by contract. Take work whose title announces it is advisory AI — "AI strategy," "AI readiness," "AI roadmap," "AI transformation" — then check which CPV drawer the buyer filed it under. Again and again, advisory AI is coded not as consultancy but as generic IT or business services, where it vanishes from any spend analysis that scans CPV codes. Every contract below is a published award notice you can look up yourself on Find a Tender or Contracts Finder.
| Contract (title) | Buyer | Filed under (CPV) |
|---|---|---|
| Cross-Government AI Adoption Review | DCMS | 73200000 — R&D consultancy |
| AI Roadmap | Health Education England | 80000000 — Education |
| Army AI Strategy — Implementation Planning | Ministry of Defence | 72200000 — Software |
| AI Readiness Consultancy | Sussex Partnership NHS FT | 72000000 — IT services |
| AI Transformation Assessment | Cheshire West & Chester | 72000000 — IT services |
| West Midlands AI Adoption Roadmap | West Midlands CA | 79000000 — Business & management consultancy |
| UKHSA AI Readiness Assessment & Use Case | UKHSA | 72000000 — IT services |
| AI Advisory Service | Surrey & Borders NHS FT | 72000000 — IT services |
Only one of these eight landed in the code that actually means "management consultancy," and the irony is sharp: the contract literally titled an adoption roadmap is the one filed honestly. This is rarely deliberate — there is simply no honest home in the taxonomy for AI advice. But the effect is structural: you cannot measure AI advisory spend by CPV code, because the code lies by omission.
“The system has no honest home for AI. So AI hides — inside IT, inside business services, inside frameworks. The invisibility is the finding.”
The GenAI moment: a wave of noise, a puddle of money
ChatGPT launched in November 2022. Did it change government procurement, or just generate a new vocabulary for the same advisory habit? Skim's analysis of contracts referencing generative AI shows both — but the money tells the real story.
34
Skim analysis of Find a Tender + Contracts Finder
The count exploded — from a handful of contracts to 34 in 2024. The value did not. What the GenAI moment primarily produced was not systems but language: "GenAI strategy," "LLM readiness," "foundation-model adoption" — new labels on procurement that, in CPV terms, is indistinguishable from the management-consulting work that came before it. The clearest sign that the advisory habit is intact: in late 2025, PwC was named a delivery partner for the government's national AI Skills Boost programme — an AI initiative, bought the way consulting has always been bought.
Who wins government AI and consulting contracts?
When the state does buy AI — or advice about it — it buys from a structurally narrow supplier base. The National Audit Office found that the six largest firms win roughly three-quarters of the consultancy work let through the Crown Commercial Service's cross-government agreement, while SMEs win just 5% of all government consultancy work. Tussell's analysis tells the same story from the spend side: a handful of suppliers account for the lion's share of management-consultancy spending in a single year, led by Deloitte, PA Consulting, Accenture and KPMG.
~75% / 5%
National Audit Office, Nov 2025
Concentration this durable is a governance question, not an accusation. The same names sit on nearly every major framework, across every change of government. When a department decides it needs to think about AI, the suppliers it reaches into are concentrated by design — and the firms that scope the problem are frequently the firms positioned to win the work to solve it. The procurement record shows the pattern; it cannot, on its own, show cause. Those are exactly the questions an open, well-read contract record is for.
What this means for SMEs — and where the opportunity is
There is a more hopeful reading buried in all this. The state's AI and consulting spend is concentrated partly because the opportunities are invisible to anyone without the tooling to read the record at scale. The contracts exist. The expiries are coming: Tussell found nearly £1 billion of management-consultancy contracts expiring in a single quarter, and £3.4 billion rolling off by the end of 2026. Every one of those is a re-compete — a door that opens for a smaller, sharper supplier, if they can see it in time and engage before the specification is written.
That visibility gap is precisely what Skim exists to close. Skim reads UK and EU award-notice data the way this study does — but continuously, and at the level of individual buyers and suppliers — to surface who is winning your tenders, which contracts are about to expire, and which buyers are worth engaging before a notice ever drops. The legibility problem this article describes is a market problem; for the SMEs on the right side of it, it is also a market opening.
Download the full report
Submit the form below to access the full report.
Want this analysis run on your own pipeline?
Start for free and run your real opportunities through Skim — see what the data says about your win chances.