If you have gone looking for this figure you will have found several. They do not agree.
One says more than half of UK firms are using AI. Another says about a third. A third says a quarter. They get quoted interchangeably, usually by someone trying to make you feel late.
Here is the thing worth knowing. They are all roughly correct. They are measuring different things. Nobody quoting them tends to mention that.
The two numbers you will keep meeting
In March 2026 the British Chambers of Commerce, working with Atos, reported that 54 percent of firms were actively using AI. Around 94 percent of the firms they surveyed were SMEs. That is where the "more than half" headline comes from.
In July 2026 the Office for National Statistics reported that about 35 percent of UK businesses with 10 or more employees were using at least one AI technology. That is the official statistic, drawn from the Business Insights and Conditions Survey.
Both were published this year. Both are from credible organisations. They differ by nineteen points.
There is a third number that circulates, usually presented as the cautious one: 25 percent. That figure is real, but it is the BCC's own 2024 number, from the same series that reached 54 percent this year. Their run goes 23 percent in 2023, 25 in 2024, 35 in 2025, then 54 in 2026. So if you have seen 25 quoted as the current position, you have been shown a two-year-old figure.
Why the gap exists, according to the people who publish the numbers
This is the useful part. You do not have to take my word for it. ONS addresses the discrepancy directly in its own article.
Lower estimates, it explains, tend to capture direct, bottom-up use, whether a specific business function is genuinely using a specific technology. Higher estimates reflect a broader, top-down view, whether a firm uses any AI anywhere across its operations.
Put plainly. Ask a managing director "does your business use AI" and a fair number will say yes, because someone in the office has ChatGPT open. Ask instead "which of these named technologies is deployed in your business operations" and the number falls, because that is a harder question to say yes to.
Neither is dishonest. They are answers to different questions.
Two other things widen the gap. The BCC surveys its own membership, businesses engaged enough with a chamber of commerce to answer a survey, which is not a random slice of British business. And ONS excludes firms with fewer than 10 employees from that headline figure, for a sound statistical reason, which brings us to the number that actually matters to you.
The figure for a business your size
Buried below the headline, ONS reports adoption by company size. Among businesses with 0 to 9 employees, 28 percent report using at least one AI technology. Among those with 250 or more, it is 49 percent.
If you run a firm with eight people, 28 percent is your number. Not 54. Not 35.
ONS leaves the smallest firms out of its headline because their survey weights can disproportionately influence the aggregate, which is a legitimate methodological choice. The side effect is that the most quoted figures describe businesses larger than most of the ones reading this.
Roughly a quarter of firms your size, then. Which means three quarters have not started. You are in ordinary company either way.
What "using AI" turns out to mean
The adoption figure tells you how many businesses said yes. It says nothing about depth. Depth is where the honest answer gets interesting.
ONS tracks how many AI technologies each adopting business uses. Since late 2023 that average has moved from about 1.4 to about 1.6. Nearly three years of intense noise. The typical adopting business added roughly a fifth of a tool.
Better still, ONS asked how extensively AI is used. Among businesses that report using it, only 10 percent say they use it extensively.
So when you read that half of firms use AI, the accurate mental translation is closer to this: half say yes, most of those use one tool, about one in ten of them properly.
That should be reassuring rather than deflating. The gap between the businesses that have started and the businesses getting real value from it is still enormous. Being late to the announcement is not the same as being late to the benefit.
Your industry probably matters more than your size
Adoption is wildly uneven by sector. ONS puts information and communication at 58 percent. Construction sits at 13 percent.
If you run a building firm and feel behind, look at that second number again. Nearly nine in ten of your competitors are in exactly the same position. That is also why most of our AI for field service work starts with vans, jobs and paperwork rather than with anything anyone would call a platform.
There is a detail in the ONS data worth pausing on. In manufacturing, wholesale and retail, accommodation and food services, the most common way businesses access AI is free-to-use software. Not bought systems, not bespoke builds. Free tools. Whereas construction, information and communication, plus professional services, more often buy external software or ready-made services.
That tells you something practical. In a lot of ordinary businesses, "adopting AI" so far has meant somebody trying a free tool and seeing what happened. That is a perfectly sensible way to start. It is also why the depth figures are so modest.
The jobs question, answered with numbers rather than reassurance
This is the question underneath the question, so here is what the data says.
The BCC found that 95 percent of SMEs using AI reported it had no impact on workforce size over the past year. 86 percent said job roles were unchanged.
ONS found broadly the same, with around half of businesses reporting no headcount change and small proportions reporting a decrease. Among firms specifically using AI to improve operations, 63 percent reported no change, 6 percent a decrease, 1 percent an increase. The Bank of England's Decision Maker Panel reached a similar conclusion.
The honest reading is not "AI never affects jobs". It is that, so far, for most small firms, it has not. What has changed is the mix of tasks inside roles. The BCC did find that firms adopting deeper, bespoke AI are more likely to anticipate headcount reductions. That group is currently about one in ten.
For a business of your size deciding whether to spend a Tuesday afternoon on this, the practical takeaway is that the realistic outcome is your people spending less time on the jobs nobody enjoys.
What is actually stopping people
41 percent of businesses with 10 or more employees told ONS they had experienced no barrier at all to adopting AI. They simply have not.
Among those who did report a barrier, lack of expertise leads, running to around 18 percent in firms of 100 to 249 people. Cost features consistently but more modestly, affecting somewhere between 7 and 14 percent. Earlier ONS work found the most common blockers were difficulty identifying a use case, cost, then lack of expertise.
Difficulty identifying a use case is the interesting one. It is not a technology problem, it is a "which of my jobs would this actually help" problem. It is answerable in an afternoon by someone who knows where to look.
The expertise gap has an obvious response. The data shows firms taking it. Among businesses citing lack of expertise, 62 percent report training or retraining existing staff, against about 26 percent of those reporting no barriers. Buying skills in is far rarer than building them. That is exactly why our practical AI training is built around your own jobs rather than a generic syllabus.
So, are you behind?
On the numbers, probably not.
If you employ fewer than ten people, roughly a quarter of firms like yours have started. If you are in construction, hospitality or the trades, it is fewer. Of those who have started, most are using one tool, lightly. Only one in ten is using it seriously.
The businesses genuinely ahead are not ahead because they adopted early. They are ahead because they picked one job that was costing real hours and fixed it properly. That is a completely different thing to being an early adopter. It is still entirely available to you.
The risk is not that you are late. It is spending money on a tool because a survey made you anxious, without first knowing which of your jobs is worth the money.
Where to start
Start by finding the hours, not the technology.
If you want to do that yourself before speaking to anybody, our bottleneck audit walks through the method. It costs you nothing but an hour of paying attention.
Book a free consultation. No pitch, no obligation. We will look at where your hours actually go, then tell you honestly where AI would help and where it would not. If it makes sense after that, we run a formal efficiency analysis and put real numbers against it.
If it does not make sense, you will hear that as well. That costs you an hour and saves you a great deal more.
Figures in this piece come from the ONS article "Artificial intelligence in UK businesses: 2023 to 2026", published 20 July 2026, plus the British Chambers of Commerce and Atos report "Future of Work: AI in the Workplace", published 18 March 2026.