For the last two years, one number anchored almost every official statement about artificial intelligence in Britain: the claim that AI could add £400 billion to the UK economy. It was meant to calm fears that the technology would bring sweeping job losses or even an economic crash. Then, in a written parliamentary answer in late October 2026, AI minister Kanishka Narayan revealed that the government’s own projection was far smaller — less than a third of the £400 billion headline — and that the larger figure had come from a 2023 report commissioned and paid for by Google.
This article traces where the £400 billion claim came from, why it sat so far above the government’s own estimates, how long departments kept repeating it after they knew better, and what the episode should teach anyone who has to evaluate an AI business case.
A £400 Billion Promise, Repeated for Two Years
What ministers actually said
For roughly two years, the £400 billion figure sat at the centre of the government’s public story about AI. It appeared in ministerial speeches, departmental briefings and agency publications as the headline economic prize of the technology transition. Former technology minister Peter Kyle cited the £400 billion number repeatedly while promoting the benefits of AI, and the figure travelled well beyond Whitehall: multiple other organisations picked it up and repeated it as an established fact.
The figure had a rhetorical job as much as an analytical one. At a moment when anxiety about AI was loudest — fears of automation-driven unemployment and whole sectors being disrupted — a very large, very round number suggested that the disruption would pay for itself. It was offered as reassurance: the turbulence is temporary, the £400 billion prize is permanent.
The answer that changed the story
The facade cracked in late October 2026. Answering a parliamentary question, AI minister Kanishka Narayan said that AI adoption and data investment could add £55 billion to £140 billion to the UK economy by 2030, bringing the central estimate to £120 billion. He added that even this more recent assessment was a hugely uncertain estimate.
According to WIRED’s investigation, that answer was the first recorded time the government set the £120 billion and £400 billion figures side by side, and the first time it said plainly where the bigger number came from: a publicly available 2023 report commissioned by Google and produced by Public First, the British consultancy now known as PF Global. Notably, the answer did not withdraw the £400 billion claim, and it did not say the number had been wrong. The government simply stopped defending it — a quiet correction by tabled paperwork rather than a public one.
Why a number came to stand in for a strategy
Large round figures are politically convenient because they do argument work that evidence cannot. The £400 billion claim let ministers answer hard questions about jobs and public services with a single reassuring statistic, and it let suppliers cite official blessing in sales pitches. That convenience is also the hazard: once a headline number is load-bearing — quoted by departments, printed in agency materials, used to justify budgets — admitting it overstated the case becomes an act of institutional self-correction that almost nothing in the machinery rewards.
Where the £400 Billion Figure Came From
A report commissioned by Google
The £400 billion claim did not begin as government analysis. It began in 2023 as consultancy work ordered by Google and written by Public First, whose economists modelled what widespread AI adoption might eventually deliver for the United Kingdom. Long-horizon impact studies like this are a familiar instrument of technology marketing: a vendor funds research, the research produces a large number, and the number then travels on its own, quoting itself from briefing to briefing.
PF Global maintains that all of its research is independent, that every judgment and assumption in the work was its own, and that its relationship with Google has continued since the report was published. Those two facts can both be true at once: a consultancy can act honestly, and a funding relationship can still shape which questions get asked, which scenarios get modelled, and which numbers reach the press release.
Why the source matters
Where a number comes from is not a footnote to the number — it is part of the number. Industry-funded studies tend to model optimistic adoption paths, because the interesting question for a sponsor is how big the prize could be, not how slowly it would arrive. Sumedha Deshmukh of the Ada Lovelace Institute put the pattern bluntly to WIRED: figures about AI that look precise often have assumptions and incentives built into them, and industry often funds or shapes the research behind them.
Once the £400 billion figure entered government use, its origin stopped being common knowledge. Departments quoted the number, not the methodology. Agencies reproduced it in training and apprenticeship materials. And for almost two years, under two governments, no published answer explained that the figure’s starting point was a vendor’s question, not a treasury one.
The pattern is not uniquely British, and it is not unique to AI. Consulting estimates funded by the firms they celebrate have anchored technology debates from cloud computing to blockchain. What distinguishes the £400 billion case is how far the number travelled from its funding: from a Google-commissioned scenario, through consultancy modelling, into ministerial language, and finally into apprenticeship handbooks — a journey in which every stop added authority and subtracted context.
Why the Two Numbers Diverged
Same consultancy, different assumptions
The gap between £400 billion and £120 billion is not an arithmetic dispute — it is a clash of assumptions. PF Global’s head of economics Jonathan Dupont explained the difference to WIRED by pointing to the speed of change: AI is a very fast-moving technology, and as we learn more, we inevitably have to update the assumptions lying behind our models. Compared with the older model that produced the £400 billion scenario, the numbers the government now uses assume a much slower rate of adoption than the consultancy thought possible in 2023.
That distinction matters more than the headline numbers suggest. The £400 billion figure was never a forecast of what would happen by a given date; it was an estimate of what could happen if adoption raced ahead of expectations across every sector at once. The government’s £120 billion central estimate asks a narrower question: what is realistically achievable by 2030, given the adoption rates we can actually observe today.
The consultancy’s own revised view
PF Global has now updated its own thinking. Given the slower rate of adoption seen in practice, its central estimate for the share of AI’s potential likely to be achieved by 2030 is much closer to the government’s number — with growth ramping up in the years after 2030. The firm has not abandoned its view of AI’s overall potential; it has redrawn the timeline, and in doing so has conceded the heart of the criticism: potential is not a budget line.
Put simply, the distance between the £400 billion scenario and the £120 billion projection is made of adoption friction — legacy systems, procurement rules, skills shortages, change management, and the ordinary caution of large organisations. None of those variables appear in a press release. All of them decide whether an economic projection survives contact with reality.
None of this makes AI economically uninteresting. The government still expects a meaningful boost by 2030, and PF Global still expects the curve to steepen afterwards. The dispute is about slope, not destination — and policy, procurement and paychecks are decided by the slope.
The estimates in black and white
One picture of the whole dispute: what each number is, who made it, and what happened to it. Read down the attribute column and the difference stops being about arithmetic at all — it is about who asked the question, how far in the future the answer lives, and how carefully the adoption assumptions were stated.
| Attribute | The £400 Billion Figure | The £120 Billion Assessment |
|---|---|---|
| Commissioned by | UK government (internal) | |
| Produced by | Public First (PF Global) | Government analysis |
| Year | 2023 | On ministers’ desks since November 2025 |
| Headline value | £400 billion | £55–140 billion by 2030 (central £120 billion) |
| Adoption assumption | Fast, broad uptake | Observed, slower uptake |
| Public status | Repeated for ~2 years, never formally withdrawn | First disclosed in October 2026 answer |
| Author’s own caveat | None prominent at publication | “Hugely uncertain” per the minister |
Who Knew What, and When
Almost a year of knowing better
The most uncomfortable fact in the WIRED story is a timeline, not a quote. Government departments and agencies used Google’s £400 billion figure for almost two years, across two governments. Meanwhile, ministers have held the lower estimate since at least November 2025 — almost a year before this autumn’s admission. And the machinery kept printing: Skills England, the government agency responsible for apprenticeships, was still publishing the old £400 billion number four months after the new lower range had landed on ministerial desks.
This is how inflated numbers behave inside large institutions. They enter through a briefing, get repeated in publications, anchor training materials and funding pitches, and acquire the authority of repetition. Correcting them is unglamorous work — nobody issues a press release to subtract £280 billion from a promise — so the old number circulates on autopilot long after the analysis has moved on.
The question that went unanswered
Parliament tried to force the clarification much earlier. Lord Holmes of Richmond, the Conservative peer who asked ministers in January 2025 about the origins of the £400 billion claim, never received a response. Speaking to WIRED after the answer finally arrived more than a year later, he called it unfortunate that the figure used repeatedly by the government — and then, unsurprisingly, by multiple other organisations — had now been significantly and rightly adjusted, and said it would have been better for the government to clarify sooner and more directly.
His central point was about stewardship rather than partisanship: robust data is critical, not least because billions of pounds of public money, and how the government chooses to deploy it, depend on numbers like these. Both the Cabinet Office, the department responsible for AI in government, and Google were approached for comment.
A timeline of the £400 billion figure
The life of the £400 billion claim, in six dates — publication, silence, correction:
| Date | Event |
|---|---|
| 2023 | Google commissions Public First; the report circulates carrying the £400 billion scenario |
| January 2025 | Lord Holmes asks ministers where the £400 billion claim came from; no answer is given |
| November 2025 | Ministers hold the lower £55–140 billion assessment internally |
| March 2026 | Skills England still publishes the old £400 billion figure, four months into the new range |
| October 2026 | Narayan’s written answer sets £120 billion and £400 billion side by side and names Google |
| November 2026 | WIRED publishes the investigation; the adjustment becomes public record |
Hype as a Policy Risk
The committee chair’s verdict
Dame Chi Onwurah, the Labour MP who chairs the Commons Science, Innovation and Technology Committee, told WIRED that the government is right to promote the potential for technology like AI to improve public services — and then named the real obstacle: hype. It is one of the central barriers to delivering the digital transformation of public services that her committee has identified, she said. Economic projections inevitably rely on assumptions, but overstating the benefits of technology risks undermining the case for change itself.
The logic is unforgiving. When a government sells transformation with an inflated number, every later disappointment reads as betrayal rather than ordinary friction. Trust in the promise collapses, and with it the political appetite for the unglamorous investment — data cleanup, skills training, procurement reform — that actually delivers the modest, real version of the benefit.
Support versus boosterism
The spat between £120 billion and £400 billion is one episode in a wider argument about where government support ends and boosterism begins. Promoting a technology and selling its best case are adjacent activities; governments do both at once, and few people outside Whitehall notice the difference until the number changes. The Ada Lovelace Institute’s Deshmukh welcomed the revised estimates but called it concerning that they follow almost two years of ministers citing research that was paid for by Google — a pattern, she argues, in which numbers look precise while their assumptions and incentives stay invisible.
Her prescription is procedural and cheap: test the numbers before building policy on them, fund independent research, measure whether promised gains are delivered, and always ask three questions of any AI figure — where did it come from, what does it assume, and who actually benefits from it.
The costs of hype fall unevenly, which is why sober numbers matter more than optimism. Vendors can reprice a market narrative in a quarter; farmers, clinics and manufacturers operate on planning horizons where a mis-sold transformation is a ruined year. When officials repeat a vendor’s best case for long enough, the disappointment lands on the institutions AI was meant to help — and on the public trust required to fund the next, genuinely promising, change.
What the £400 Billion Figure Means for AI Investment
Budgets built on borrowed figures
Lord Holmes’s point about robust data generalises far beyond Westminster. When a headline number anchors policy, procurement and training programmes, everything downstream inherits its optimism. Public money flows toward the activities the figure justifies; private firms read the same headlines, hear the same £400 billion repeated by suppliers, and set their own expectations accordingly. If the anchor number was really a vendor-funded best case, the resulting misallocation is quiet but real: programmes designed for a speed of adoption that never arrives, and boardroom expectations calibrated to a scenario rather than an estimate.
The adoption problem is the whole problem
PF Global’s own explanation — slower adoption than assumed — names the variable every CIO and operations director fights daily. The technology matures faster than organisations absorb it, and the gap between a £400 billion potential and a £120 billion realised-by-2030 estimate is made of exactly that friction. Practical guides to deploying AI employees and autonomous agents and intelligent automation exist precisely because this friction is where projects live or die — and headline numbers never show up in the work itself.
What businesses should take from this
Treat every spectacular AI figure — your vendor’s, your competitor’s, your government’s — as a scenario, not a forecast. Before importing a return-on-investment model, ask what adoption rate it assumes, whether the horizon matches your planning cycle, and whether anyone independent has tested it. Then pilot narrowly, measure against a baseline, and scale on evidence from your own operations rather than someone else’s best case. The same discipline that data management and analytics programmes need in miniature is what the national numbers lacked at scale.
How to Evaluate an AI Economic Claim
Six questions to ask before you believe a number
Where did it come from, what does it assume, and who benefits from a given figure? Those are the three questions the Ada Lovelace Institute says any government should ask before using a number for economic or productivity gains from AI. Three more belong on any serious review: has the result been independently replicated, is the projection’s horizon matched to the decision at hand, and is delivery actually measured once the spending starts?
The table below turns the six questions into a working review sheet — the same one a press office should have applied before repeating the £400 billion claim for two years, and the same one a board should apply before anchoring its AI strategy on any spectacular figure.
| Question | What to look for | Red flag |
|---|---|---|
| Who funded the research? | Independent commissioning, disclosed mandate | Vendor-funded study quoted as neutral fact |
| What assumptions drive it? | Published adoption rates and scenario definitions | Optimistic adoption assumed silently |
| Is it potential or forecast? | Clear distinction between ceiling and expectation | Best case presented as expected outcome |
| Does the horizon fit? | Projection period matches the decision window | Distant-year gains justify near-year spending |
| Has anyone replicated it? | Independent estimates that bracket the claim | A single study stands alone |
| Is delivery measured? | Adoption statistics tracked after the announcement | Nobody counts realised gains |
What to watch next
Several threads remain open. The Cabinet Office and Google were approached for comment about how a figure from a vendor-funded report became the government’s own headline. PF Global’s work for Google continues. Skills England and other bodies that published the old number will have to reconcile their archives with the new, much smaller range. And for the first time the UK has statistical infrastructure to check the rhetoric against reality: the Office for National Statistics’ digital economy programme now tracks AI adoption in British businesses. If measured adoption stays flat while the quotes keep rising, the next correction will not take two years to arrive.
The deeper lesson is institutional. Numbers like the £400 billion figure do not fail because consultants are careless or because ministers are dishonest; they fail because nobody owns the small unglamorous job of checking whether the story still matches the world. Any organisation that uses external projections should appoint that job to someone — a desk whose mandate is to ask, every quarter, whether the assumptions underneath the number have aged, and to say so publicly when they have.
References
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
