How many UK businesses actually use AI? Three answers

If you’ve been told that “most businesses are using AI now”, you’ve been told something that’s either true or wildly false depending on which survey the person read.

Three credible organisations published UK adoption figures in the last eighteen months. The Office for National Statistics says 35%. The British Chambers of Commerce says 54%. The Department for Science, Innovation and Technology says 16%. Nobody’s lying. The gap between those numbers is the most useful thing in all three reports, and it points at something worth knowing if you run a business and haven’t got round to any of this yet.

The three numbers, and where they come from

The ONS says 35%. Its Artificial intelligence in UK businesses release, published on 20 July 2026, is far and away the most robust of the three: 38,637 responding businesses, fieldwork in the second half of June 2026. It found 35% of UK businesses with ten or more employees using at least one AI technology, up from about 12% in late 2023. Almost tripled in under three years.

The BCC says 54%. Its Future of Work report, published 18 March 2026 with Atos and the University of Essex, found 54% of firms actively using AI — up from 35% in 2025, 25% in 2024 and 23% in 2023. About 94% of the firms surveyed were SMEs. It’s a Chamber-network sample, so it skews towards businesses engaged enough to answer a Chamber survey, but the trajectory is the point.

DSIT says 16%. Its AI Adoption Research was published in January 2026, but read the small print: the telephone fieldwork ran from February to May 2025. It’s a year older than it looks. Across 3,500 businesses with five or more employees, it found 16% using at least one AI technology, 5% planning to, and 80% doing neither.

So: a very large, very recent, technology-by-technology survey of firms with ten or more staff says 35%. A self-selecting sample of engaged SMEs says 54%. A telephone survey from over a year ago, reaching further down into small firms, says 16%. All three are defensible. Pick whichever suits your argument, or — better — look at what they agree on.

What they agree on is more interesting

Adoption is broad and shallow. That’s the finding, and every one of the three supports it.

The ONS number that should stop you is this: the average adopting business uses 1.6 AI technologies. Up from 1.4 in late 2023. In nearly three years of the most-hyped technology shift in a generation, the typical adopter has added two-tenths of a tool.

Only 10% of adopting businesses say they use AI “extensively”. And across all businesses with ten or more employees, only 15% say more than half their employees use AI daily, while only 11% have given AI training to more than half their workforce. The ONS draws the conclusion itself, in the sort of carefully flat language statisticians use when they’re making a point:

This would imply relatively limited transformative impacts to date for most AI-adopting firms.

DSIT’s older data says the same thing from a different angle: 77% of businesses using AI reported no change in revenue yet, against 12% reporting an increase. And 31% of businesses spent nothing at all on AI in 2024 — including 27% of the businesses that said they were using it. They were using the free tier.

That’s what most “AI adoption” is. One or two people using a chatbot, on a free plan, for drafting. It counts. It’s genuinely useful. It is not the thing the headlines are describing.

The gap between what the business does and what its staff do

The single most revealing pair of numbers in the ONS release: 35% of businesses report using AI, but 55% of employees report using AI for work or education.

The two come from different surveys and aren’t a clean like-for-like — the employee figure is self-reported, covers education as well as work, and the ONS notes it may take in some use outside formal business activity. Take a chunk off it and the gap is still there.

Some of that gap is people using AI at work whether or not their employer knows. It’s usually harmless and often quietly productive. It also means client information, supplier pricing, half-finished contracts and staff details are being pasted into tools nobody has assessed, under terms nobody has read, with no record of what went where.

You don’t fix that with a ban; people find the tools too useful, and a ban just moves it onto their phones. You fix it by picking a tool, paying for the version that doesn’t train on your data, and telling everyone which one it is. That’s an afternoon’s work and it’s the highest-return AI decision most small businesses could make this year.

Small businesses that adopt, adopt deeper

Here’s the finding that runs against expectations. The ONS asked adopting businesses whether they use AI extensively. Among businesses with 250 or more employees, 9% said yes. Among businesses with nought to nine employees, 17% said yes — nearly double.

Big firms adopt more often. Small firms that adopt, adopt harder. That makes sense once you think about it: there’s no procurement committee, no eighteen-month IT roadmap, and the person who decides is usually the person doing the job. If you run a small business, that’s your advantage, and it’s not a small one.

The BCC data hints at the same shape. Around one in ten SMEs are going beyond off-the-shelf tools into deeper, bespoke work. Worth being straight about the other side of that: the BCC found those same firms are more likely than casual users to expect headcount reductions. Going deeper is where the returns are, and it’s also where the harder decisions start.

What’s stopping everyone else

Not cost, mostly. The ONS found 41% of businesses with ten or more employees reported no barriers to adopting AI in the previous three months. Among those that did report a barrier, lack of expertise leads — around 18% for firms of 100 to 249 people — while cost lands somewhere between 7% and 14% across size bands. Regulation and staff concerns are cited far less than you’d expect from the coverage.

DSIT’s answer was blunter. Its top barrier, at 71%, was “lack of identified need”. Fully 51% of businesses said they do not see AI as relevant to their organisation at all.

Which is either a failure of imagination or a perfectly sensible read of the situation, depending on the business. Some firms genuinely have no repetitive information work worth automating. Most have more of it than they think, because it’s spread thinly across people who’ve stopped noticing they’re doing it.

The sector split makes the point. In information and communication, 58% of businesses use AI. In construction, 13%. It’s tempting to conclude construction doesn’t need it. But most construction firms are drowning in quotes, variations, subcontractor paperwork, compliance certificates and chasing — which is exactly the shape of work this stuff is good at.

What we’d actually do with this

Three things, in order.

Find out what your people are already using. Not to tell them off — to find out which jobs they’ve quietly decided are worth automating. That’s free market research into your own operation, and it’s better targeted than any consultant’s audit because it comes from the people doing the work.

Go deep on one job rather than shallow on ten. The 1.6-technologies figure is what shallow looks like, and shallow is why 77% of the AI-using businesses in DSIT’s survey saw no revenue change. Pick the single process that eats the most time — the one where someone retypes information from one system into another — and do that one properly, end to end, with the errors handled.

Train the people, not just the tools. The ONS found that 62% of businesses citing a lack of expertise are training existing staff, against 26% of those reporting no barriers. The firms that admit they don’t know what they’re doing are the ones doing something about it. Only about 10% of businesses report automating or replacing roles at all — this is overwhelmingly a story about people getting more done, not fewer people.

One last note on what the statistics can’t yet tell you. The ONS says explicitly that it is still “exploring new measures of how AI is integrated into business processes, including agentic workflows”. Official statistics don’t currently capture AI agents at all. So when you read that adoption is at 35%, understand that it’s measuring people using tools — not systems doing work on their own. That’s the next number, and nobody has it yet.