More than half of small businesses now use AI. In 2023 it was a fifth.
That is a faster change than most technologies manage in a decade, and it has happened largely without anybody deciding it should. Nobody signed off a rollout. People started using the tools because the tools were there.
The interesting number is the other one. When the Office for National Statistics asked firms what was stopping them adopting AI, the most common answer was not cost and it was not skills. It was difficulty identifying activities or business use cases, named by 39%. Cost came second on 21%. Expertise third on 16%.
So the barrier is not that it is expensive or that nobody can work the thing. It is that people genuinely cannot see where it fits.
The honest version
Most of what gets written about AI in business is one of two things. Either it will transform everything, or it will leak your data and make things up.
Both of those are sales pitches. The useful version is narrower and duller: AI is good at a specific shape of task, and most businesses have a few of those lying around.
The shape is this. Work where a competent draft saves more time than checking it costs, and where being wrong is recoverable.
That is the whole test. It rules a lot in and it rules a lot out.
Where it fits
In a business of thirty people, these are the jobs that pass the test:
Drafting the first version of something. A quote covering letter, a policy, a job advert, a tender response, a difficult email. The work of getting from nothing to something is where the time goes, and a draft you rewrite is still faster than a blank page.
Making sense of something long. A supplier contract, a specification, a spreadsheet somebody left behind. “What does this actually commit us to” is a question AI handles well, and you can check the answer against the document in front of you.
Rewriting for a different reader. The same information for a client, for a surveyor, for your own staff. Plain English on request.
The jobs nobody has time for. Tidying a product list. Writing the descriptions nobody has written. Pulling the themes out of two years of customer emails. These usually do not get done at all, so a rough version is a genuine gain.
Getting unstuck. A formula that will not work, an error message that means nothing, the start of a document you have been avoiding.
Where it does not
Anything you cannot check. If you could not tell a wrong answer from a right one, you have not saved any work. You have moved the risk somewhere you cannot see it.
Numbers that matter. It will produce a figure with complete confidence. Use the system that holds the real number.
Anything with client data in it, unless you know where that data goes. This is the part worth being careful about, and we have written separately on Shadow AI and what it looks like in a smaller business.
Work where being wrong is expensive. Advice to a customer, anything with a regulator attached, anything that goes out under somebody’s professional name without a human reading it properly.
Replacing the person who knows. It does not know your business, your customers or the reason you do that odd thing in the second week of the month.
The worry is not misplaced
92% of small businesses say they worry about AI risks. The top concerns are inaccurate responses at 54% and security at 39%.
Both are reasonable, and neither is a reason to avoid the tools. They are a reason to decide where you are willing to use them.
That decision is easier than it sounds, because it comes down to two questions about any given task: can somebody here check the answer, and would it matter if it were wrong? If the answer to the first is yes and the second is no, use it. If not, do not.
What to do first
Not a strategy. A list.
- Find out what is already being used. It will be more than you think, and some of it will be on personal accounts. You cannot make decisions about tools you do not know about.
- Write down what staff can and cannot put into them. One page. We have a free AI acceptable use policy you can use as it is.
- Pick one job from the list above and try it properly for a month. Not a pilot with a steering group. One person, one task, and an honest answer at the end about whether it helped.
- Check the tools you already pay for. Microsoft 365 and most accounting packages now include AI features, which means you may be buying something twice.
That is it. The businesses getting value out of this are not the ones with an AI strategy. They are the ones who found two or three jobs it is good at and stopped there.
Not sure where you stand?
The free AI risk check takes a few minutes and tells you what is already being used across your business and where the gaps are. No sales call attached.
Sources, accessed 17 October 2026: Federation of Small Businesses, survey of 904 small business owners, published 16 June 2026, for the adoption and concern figures. Office for National Statistics, Management practices and the adoption of technology and artificial intelligence in UK firms: 2023, published 24 March 2025, for the barriers to adoption.
