35% of UK businesses use AI. The average adopter uses 1.6 tools.

The Office for National Statistics asked 38,637 businesses what they are actually doing with AI. Fieldwork ran through June this year; the article came out on 20 July. As UK evidence goes, that is about as solid as it gets — not a vendor survey of 200 self-selected respondents, but the national statistics office with a sample most research firms could only dream of.

The headline is the one everybody quoted: adoption among businesses with 10 or more employees has gone from around 12% in late 2023 to around 35%. Roughly tripled in under three years.

The number underneath it is the interesting one, and almost nobody quoted that.

1.4 to 1.6

That is the change in the average number of AI technologies used per adopting business over the same period. Not 1.4 to 4. 1.4 to 1.6.

So the picture is not a wave of transformation. It is an enormous number of businesses trying one thing. The ONS puts it flatly: adoption has been “relatively shallow”, which “would imply relatively limited transformative impacts to date”.

Wide, not deep. Which, if you have quietly adopted one AI tool and assumed everyone else is three steps ahead of you, should be a considerable relief. You are not behind. You are the median.

The number that is being misquoted

You will see it written that only 10% of businesses use AI extensively. Check the denominator, because that is not what the figure says.

It is 10% of adopting businesses. Adopters are about 35% of firms with 10 or more employees. So extensive users are roughly 10% of that 35% — about 3.5% of businesses, or one in twenty-nine.

Reported as “one in ten businesses”, it overstates the real figure by nearly three times. This is the commonest way a true statistic becomes a false impression, and it is worth doing the arithmetic yourself whenever a percentage arrives without its base attached.

You are less behind than you think

Two breakdowns are more useful than the national average.

By size: 28% of businesses with 0–9 employees have adopted something, against 49% of those with 250 or more. A real gap, but a much narrower one than the “big firms have all the AI” story suggests. The smallest businesses are not sitting it out.

By sector: information and communication leads at 58%; construction sits at 13%. That spread of 45 points dwarfs anything the national average tells you. A builder with one working AI tool is ahead of his sector. A marketing agency with one is behind its own. The average is nearly useless as a personal benchmark.

So what should you actually do with this?

The reflex on reading “1.6 technologies” is to go and adopt a third one. I think that is the wrong lesson.

A count of technologies is a measure of purchasing, not of value. One tool that reliably removes four hours a week beats six that each save ten minutes and need managing. The better question is not how many but whether the one you have is actually doing a job you would otherwise pay for — and whether you would notice if it stopped. We wrote about judging that honestly in cost per task rather than cost per seat, and about picking the first thing worth automating in what is actually worth automating.

What argues against reading too much into it

It is self-reported. “Do you use AI?” is a question people answer differently depending on whether they count the spam filter, the autocomplete in their email, or only the chatbot they pay for. Some of the tripling is genuine adoption and some is a broadening idea of what counts.

The ONS also cautions that “not all technologies are relevant to all businesses” — a low count is not automatically a failure. A firm that needs exactly one thing and has it is finished, not shallow.

And tripling in under three years is genuinely fast. “Shallow” is not “nothing”: a third of mid-sized British businesses doing something they were not doing in 2023 is a real shift, even if each is doing one thing.

One honest note on sourcing. I have seen the depth gap explained as a data problem — people correcting AI output without ever feeding the corrections back. It is a plausible theory and it may well be right, but the ONS release does not say it, so I am not going to put it in their mouth. What they do say is the quieter point about relevance.

The short version

A third of UK businesses of any size have adopted AI, and the typical one uses a single tool. If that is you, you are normal. Spend the effort on making that one tool genuinely load-bearing rather than on collecting a second.

Source: Office for National Statistics, “Artificial intelligence in UK businesses: 2023 to 2026”, published 20 July 2026, read directly. Figures are from the Business Insights and Conditions Survey wave 159, fieldwork 5–28 June 2026, 38,637 responding businesses. The one-in-twenty-nine figure is my own arithmetic on the ONS numbers (10% of adopters × 35% adoption), not a figure the ONS publishes.