The UK’s £400bn AI forecast and its £120bn estimate measure different things

The UK’s widely repeated £400bn AI figure came from a 2023 Google-funded study, while the government’s current central estimate is £120bn by 2030. Resultsense’s new analysis traces how the larger number travelled through official documents and explains why the two forecasts are not a simple before-and-after correction.

Watch Desk analysis

What happened

The £400bn figure originated in a Public First study commissioned by Google, which estimated potential economic value from generative AI by 2030. The government’s estimate is £55bn to £140bn in additional gross value added (GVA), with £120bn as its central figure. The two figures use different methods and describe different things, so the headline gap does not make them directly comparable.

Resultsense says Public First assessed roughly 17,000 US work tasks, using a language model to estimate whether AI could halve the time required for each task. Its analysis also assumed that the effects of a general-purpose technology would take 10 to 20 years to emerge. The government’s model instead draws on OECD productivity estimates, modelled UK adoption rates and a 2030 GVA baseline. Its medium scenario calculates a £117bn uplift, while the minister’s written answer gives the central estimate as £120bn.

Why it matters

These are modelled projections, not observed gains. The government describes its estimate as highly uncertain, and its adoption assumptions do much of the work: the model scales productivity gains according to how widely AI is expected to be used. Public First’s estimate also depends on assumptions, including US task data and a model’s assessment of which tasks AI could speed up.

The £400bn figure acquired official weight as documents cited one another. Resultsense traces it from Public First’s study to the government’s AI Opportunities Action Plan and then to a later skills announcement. That chain matters for anyone using AI forecasts in a strategy or business case: a number can look newly official while still resting on an older model.

Our read

The useful takeaway is not that one forecast has been proved wrong. It is that the forecasts ask different questions, use different assumptions and carry different uncertainty. Resultsense’s account gives readers practical questions to ask of the next confident AI number: where did it originate, what assumptions drive it, and who funded or benefits from the work?

A number in a government document is not automatically a government calculation. Following the footnotes is less glamorous than quoting the headline, but considerably more useful.

What to watch

• Whether the government updates the £400bn claim still carried in its AI Opportunities Action Plan.
• Whether future estimates report observed AI adoption alongside assumed adoption curves.
• Whether the government explains how its £120bn central estimate relates to the £117bn worked scenario.

Discussion spark: When public bodies cite economic forecasts for AI, should they be required to publish the model’s assumptions alongside the headline figure?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.