Small business AI adoption stats: why every survey disagrees

One headline says half of UK small businesses now use AI. Another says it's barely one in six. A third says seven in ten. All of them are published in the same year, by credible organisations, about the same country. This post explains why those numbers disagree so wildly, which differences actually matter, and how to work out where your own business sits without trusting any single percentage.

The range is roughly 16% to 71%, and nobody is lying

Depending on which survey you read, reported UK small-business AI adoption lands anywhere from about 16% to about 71%. That's not a rounding difference. That's a four-fold gap on what sounds like the same question.

The instinct is to assume someone is cooking the books — usually whoever is selling AI. Occasionally that's true. Mostly it isn't. The gap comes from three boring decisions every survey has to make before it can print a number, and almost no headline tells you how they made them.

Decision one: what counts as a "small business"

In UK terms, "small business" officially stretches from a sole trader with no employees to a firm with 49 staff. Those two things behave nothing alike. A 40-person agency has someone whose job includes trying new software. A solo electrician has a van, a phone and no spare Tuesday.

If a survey panel skews towards the larger end — firms with an office, a marketing person, a software budget — adoption looks high. If it reaches genuine micro-businesses and sole traders, it looks low. Same country, same year, same honest methodology, different population.

Check this first on any number you see. "Small business" in one report can mean 10–49 employees. In another it includes the 4.1 million-odd UK businesses with no employees at all.

Decision two: what counts as "using AI"

This is the big one. Consider five real situations:

• You asked ChatGPT to rewrite a quote email once, in March.
• You use an AI tool every week to draft social posts.
• Your accounting software added an AI categorisation feature and switched it on by default.
• Your phone line is answered by an AI receptionist on every call, including 2am.
• You've built an AI step into how jobs get quoted and booked.

A survey asking "have you used any AI tool in your business?" counts all five. A survey asking "does your business use AI in a core process?" counts maybe the last two. A survey asking about paid AI subscriptions counts a different subset again. The third case — AI bundled into software you already bought — is the sneakiest, because plenty of owners genuinely don't know whether they're using AI or not.

So a 71% figure and a 16% figure can both be accurate measurements of two different things: "has touched AI" versus "runs on AI".

Decision three: who actually answered

Surveys reach people through panels, customer lists, email invitations and trade bodies. Everyone who responds has already done something: opened a business email, clicked a link, spent ten minutes on a questionnaire. That's a mildly digital act. The sole trader who never opens marketing email is systematically harder to reach, and that trader is also less likely to be using AI.

If the survey is sent by a software company to its own customer base, the skew is stronger again. Not dishonest — just a sample of people who already buy software.

What the recent UK figures actually say

Put the Simply Business figures together and you get something more useful than either on its own: about 47% using, 13% intending, which is roughly 60% of owners either in or on the way in — and about 40% with no plans at all. That last group is not small, and it isn't shrinking as fast as the headlines imply.

Why "doubled in a year" and "1% in construction" are both true

These two findings look like they contradict each other. They don't. The FSB sector split shows where the doubling happened.

Professional and technical firms — accountants, consultants, agencies, designers — do work that is largely text, analysis and admin. That's exactly what current AI tools are good at, and adoption there sits around 37%. Construction is site work, materials, scheduling and risk. Around 1% adoption isn't stubbornness; it's an honest reflection that very little of the day job is a document.

If you average a sector at 37% with a sector at 1%, you get a national figure that describes neither. That's the problem with benchmarking yourself against the headline: for a builder, "47% of small businesses use AI" is about as relevant as the average rainfall in Europe.

The most useful finding is the one nobody headlines

46% of small firms saying they lack the knowledge to use AI is, to my mind, the single most informative figure in the whole pile. It explains the variation better than any methodology note.

It means a large share of the "not adopting" group aren't making a considered decision. They haven't got far enough to make one. It also means the adoption numbers are unstable in a specific way: they'll move sharply whenever a tool arrives that requires no knowledge to use. That's why AI bundled into existing software — your accounting package, your booking system, your phone — shifts the numbers faster than standalone AI tools do. Nobody has to decide anything.

It also explains the scepticism. If half the market doesn't yet know what these tools do, the loudest voices in the room are the vendors, and vendor enthusiasm is not evidence. We've written separately about the gap between what enterprise AI agents are actually shipping and what's still a pitch deck, and about what people genuinely complain about with AI writing tools. Both are worth reading before you let a percentage make you feel behind.

Five questions to ask of any AI adoption statistic

1. How did they define a small business? Does it include sole traders, or start at 10 employees?
2. How did they define using AI? Once, ever? Weekly? In a core process? Paid for?
3. Who did they ask, and how did they reach them? A software company's own customer list behaves differently from a representative panel.
4. Is there a sector breakdown? If not, assume the average is hiding a 37%-versus-1% style split.
5. When was the fieldwork? Not when it was published. In a fast-moving area, a six-month lag matters.

If a headline won't answer at least three of those, treat it as colour, not evidence.

How to benchmark yourself instead

The honest version of "am I behind?" has nothing to do with percentages. It's three questions about your own week:

• What is definitely being lost right now? Missed calls, unreturned enquiries, quotes that go cold. These are measurable. If you want the arithmetic, we've laid it out for what missed calls cost a UK trades business.
• What of that is a repeat-shaped problem? Software of any kind is good at things that happen the same way every time. The same enquiry questions, the same reminder, the same follow-up at day three.
• What would you still want a person for? Complaints. Bad news. Anything where the customer needs to feel heard rather than processed. Plenty of customers actively don't want to talk to an AI, and that preference isn't going away.

Answer those and you'll know whether to adopt anything, regardless of whether the national figure is 16% or 71%.

Where this doesn't apply, and where AI isn't the answer

Three honest limits.

First, if your enquiry volume is genuinely low — a handful of calls a week, all of which you answer — then no adoption statistic should push you into buying anything. The maths won't work. A tool that saves two hours a month and costs £100 a month is a bad trade, and it stays a bad trade no matter what your competitors are doing.

Second, if your work is mostly physical and relationship-led, the FSB's 1% construction figure is telling you something real rather than something embarrassing. The useful AI for a site-based business is narrow: answering the phone while you're up a ladder, chasing a quote, writing the bits of text you hate. It isn't going to do the job. We cover what an AI receptionist honestly can and can't take off your plate if that's the narrow slice you're considering.

Third, if you're adopting because a percentage made you anxious, stop. Adoption driven by a headline tends to produce a subscription you forget to cancel. Adoption driven by a specific, repeating, measurable loss tends to stick. The surveys can't tell you which you've got — only your own diary can.