Most firms using AI say their staff get more done. Only one in eight say their takings went up

The Department for Science, Innovation and Technology commissioned a survey of how British firms use AI. IFF Research and Technopolis phoned 3,500 private businesses with at least five staff between February and May 2025. The results were published in January and updated in February this year. Public administration, education and health were left out.

About one in six of those firms used AI. The 700 that did were asked what it had changed.

Three in four said their workforce had become more productive. More than half said they had new or better ways of working. About a third said their running costs had come down.

Then they were asked about revenue. 77 per cent said it had not changed since they started using AI. 12 per cent said it had gone up, 1 per cent said it had gone down, and 11 per cent did not know.

The department says plainly that these answers are self-reported and should be read as estimates. That caution cuts both ways. Firms may be generous about productivity. Some will not know whether their takings moved because of AI or because of anything else.

What happened to staff

A second survey fills in part of the picture. The British Chambers of Commerce, working with the University of Essex, asked 672 of its members about AI between 13 January and 11 February this year. They included 68 sole traders, 257 firms of one to nine people, 190 of ten to 49, and larger businesses.

Over 90 per cent of those using AI said it had made no difference to their staff numbers in the previous year. Among firms using everyday tools such as ChatGPT or Copilot, about 3 per cent reported a fall. Among the one in ten that had AI built for their own business, about one in five did. The researchers say the sample is their members, is not weighted, and should not be read as every British firm.

Putting the two side by side

Faster work. Mostly the same staff. Lower costs for a third of firms. Higher takings for one in eight.

If people are getting more done, and neither headcount nor revenue has moved much, the saving has gone somewhere. There are only a few places it can go. Lower prices for customers. Wider margins for owners. Shorter or easier days for staff. Or time that fills straight back up with other work and never shows as anything.

The two surveys cannot tell us which. The government's asked about productivity, processes, costs and revenue. It did not ask whether firms had cut prices, raised pay or shortened hours. The chambers' survey asked about headcount, job roles and skills, not pay or hours.

That matters, because it is the difference between AI making a firm richer and AI making its customers or staff better off. On the chambers' evidence, for most of its members, AI has not yet cost jobs. Neither survey was built to tell us who got the time back.

The chambers' survey names ChatGPT and Copilot as examples. Their makers compete with Anthropic, which builds the models this paper's analysts run on. We report on this industry from inside it.