AI for small business: what the UK numbers say, and where to start

The introduction came with a warning. Very hands on, the mutual contact said. Drops everything the moment something happens in the business.

That turned out to be accurate. It also turned out to be the whole diagnosis, and none of it was about software.

I think about that conversation whenever I read about AI for small business, because almost everything written on the subject starts in the wrong place. It starts with the tools. The owner who is already doing the heavy lifting, with no spare afternoon to learn anything, barely features.

So this starts somewhere else. With what the evidence actually says, who is telling you what, and the one small job worth trying first.

Why the numbers you read disagree

Search for this and you will be told that a small minority of UK small businesses use AI. Or most of them. Sometimes both, on the same page of results.

One guide near the top quotes fewer than 15%, attributed to the Federation of Small Businesses. The FSB's own survey of 904 small business owners, published in June, found 55% of its members using AI.

That is not even the widest gap.

The Department for Science, Innovation and Technology found 16% of UK businesses with five or more employees using AI in spring 2025. Enterprise Nation, in a report partner funded by Google, Sage, Dell and Square, found about one in five small businesses using it regularly.

The ONS, surveying more than 38,000 businesses this June, found 28% of micro businesses using at least one AI technology. And Opinium, in research commissioned by OpenAI, produced figures that imply around 81%.

So which one is right?

Probably all of them, which is less helpful than it sounds. They are asking different questions. "Have you ever used it" and "do you use it regularly" are completely different measures.

Who asks matters too. The FSB surveys its own members, who lean towards engaged firms. A survey paid for by a company that sells AI tends to find a lot of AI.

If I had to pick one figure, it would be the ONS one. It is official, the sample is enormous, and nobody involved is selling anything. But I might be wrong, and the honest answer is that nobody knows precisely.

It is worth being clear what that means for you. There is no stampede you are about to be left behind by. What there usually is, in most towns, is one or two competitors getting quietly quicker at the dull parts of the same job. That is the thing worth watching.

Who the other guides are written for

I looked at what ranks for this before writing it, which seemed only fair.

A programme page for a course, sponsored by a telecoms company. A list of ten AI tools whose first recommendation is the publisher's own product. A long and genuinely thorough guide from a firm that builds custom AI systems.

None of that makes any of them wrong. It is just worth knowing who is talking before you take their advice about what to buy.

The same applies here. I sell AI training and consultancy, so read this with that in mind. The thing I would point to is that most of what follows is about what you can do without paying anybody, including me.

What AI for a small business actually looks like

Forget the conference version for a minute. On an ordinary Tuesday, it looks like this.

An email from a customer who is annoyed about a late delivery. You paste it in with the name taken out and ask for a calm reply in your own words. You change two lines and send it yourself.

A forty page tender document. You ask for the deadlines, the mandatory requirements and anything that would rule you out, and you get them on one page. Then you check the three that matter against the original, because it will occasionally miss one.

Last month's supplier invoices, checked against the prices you think you are paying. The job advert you have been meaning to write since March. The notes from a site visit, typed up into something the office can actually use.

None of that will feature on a conference stage. That is roughly why it works.

What it should not be doing is deciding your prices, sending anything without a person pressing send, or giving you tax or employment advice that you act on without checking. It drafts. You decide. That split is the whole thing, really.

What firms your size actually get from it

Here is the part the brochures leave out.

A Bank for International Settlements study of more than 12,000 European and US firms found AI raised firm level labour productivity by 4%. But the gains were largest in medium and large firms, and larger again where firms had also invested in their data and in training their people.

So the idea that AI levels the playing field between you and a company fifty times your size is not what the best evidence shows. I would rather you heard that from me now than found it out after buying something.

To be fair to the other side, small firms using it do report benefits. In the FSB's survey of its members, 59% of those using AI reported higher productivity and 24% higher revenue, while 8% reported reducing staff.

Those are self reported impressions rather than measured results. They are still worth having, as long as you read them that way.

Why the small end gets less out of it

The reason is not the technology. Everybody has access to the same tools, at roughly the same price.

The difference is time. A larger firm can give someone a week to work out how to use it properly. In a business of six people, that someone is usually the owner, and the owner is the person who drops everything the moment something happens.

I say that as someone who ran one, the family butchery business, until it closed in January.

It is why I think the first conversation with an owner should be about their time rather than about tools. Which hour of the week would they most like back?

The answer is nearly always something small and slightly tedious. That is good news, because small and tedious is exactly what these tools are best at.

Which tool, and whether it matters

Less than you would think.

ChatGPT, Microsoft Copilot, Google Gemini and Claude will all do the jobs above. The gap between two people using the same tool is far bigger than the gap between the tools.

My honest advice is to start with whichever one you already have. If the business runs on Microsoft 365, find out which version of Copilot is already included before paying for anything new. If everyone lives in Gmail, try Gemini first.

The paid individual plans for the main assistants currently cost around twenty pounds a person a month. Check the price on the vendor's own site before you decide, because it changes, and the business versions cost more.

One thing matters more than which one you pick.

A customer's complaint email, with their name and order number in it. Personal account or business account?

Business account. Or take the details out first.

The business versions generally do not use your data to train their models. Several personal plans do by default, unless you switch it off.

Where it goes wrong at this size

Most AI failure in small businesses is not the tool failing. It is buying a project when what was needed was a skill.

A chatbot for the website, bought before anyone in the business has used AI for a week. An automation built around a process nobody had written down. A licence for every member of staff, when DSIT's research found only about 30% of staff in businesses that have adopted AI actually use it.

The mistake I would take most seriously is customer details pasted into a free personal account. It is the easiest one to make and the hardest to undo, because once personal data has left the business you cannot get it back.

A simple rule covers most of it. If you would not put it in an email to a stranger, do not put it into a personal AI account.

Do you need an AI consultant, a course, or neither

For most small businesses starting out, neither.

One afternoon, one tool you already pay for, and one real job from your own inbox will teach you more than a slide deck. That is not great news for my business. It is true anyway.

A course starts to make sense when several people in the business are going to use it and you want them doing it properly, and safely, rather than each working it out alone.

A consultant makes sense later. When the job crosses several systems, or when you have tried the simple version and hit a wall.

The warning sign in either case is anyone who wants to start with a large project and a large number. Buy one small thing. See if it works. Then decide.

What you could do on Monday

Pick the email you answer most often. The one where you could practically type the reply with your eyes shut.

Take one real example, remove the customer's name and anything personal, and paste it into whichever AI assistant you already have. Ask for a reply in a friendly, plain tone, under a hundred words.

Then add two questions to the end. What is missing from this request? What should I check before I send it? Read what comes back, fix what is wrong, and send it yourself.

Before you start, time how long that reply normally takes you. Time it again afterwards, and write both numbers down.

It is a small, boring, easy win, which is exactly what a first one should be. And those two numbers will tell you more than every statistic in this post.

The honest limit

AI will make some of the dull work in a small business quicker. On the best evidence available, it will not close the gap between you and much larger firms on its own.

And it will not give an owner who drops everything any more attention. If the business is short of time because the owner is doing the heavy lifting, a tool will make the heavy lifting slightly faster. It will not stop the owner doing it.

That conversation comes first. No software is going to start it for you.