I mentioned to a business adviser last week that half of UK businesses can't see how AI is relevant to them.
He looked at me as though I'd said something ridiculous.
"That's rubbish!"
I understood the reaction because, when I first read it, I thought the same. Surely not. Surely everyone can see something.
But the figure isn't mine, and it isn't a guess.
The Government's own AI Adoption Research (opens in a new tab), published by the Department for Science, Innovation and Technology in January, found that 51% of UK businesses do not see AI as relevant to their organisation.
More than half.
The research also found that the most common barrier to adoption was not cost. Alongside a shortage of skills, the biggest barrier was that businesses had not identified a use for AI at all.
So either half the country is being slow, or something else is going on.
I think it's something else. And I don't think it reflects badly on them at all.
Part one: they've been hammered into switching off
Look around and you can see it. Not just caution about AI — actual weariness. Posts that have tipped from "interested" to "here we go again."
That isn't apathy. It's the entirely human response to being sold at relentlessly by a wall of hype that never quite touches the ground.
Because that's what most businesses are getting.
Not:
Here's something that would save you three hours a week.
Just noise.
Transformation! Revolution! Agents! Disruption!
The volume turned up to eleven and the substance nowhere to be found.
The uncomfortable truth about why the hype wins: hype is easy. Waving around the newest, shiniest word costs nothing and takes ten seconds.
What's genuinely hard is showing someone the real thing.
The real thing is specific and quiet. It takes time spent understanding one particular business rather than broadcasting at everyone.
I've watched this exact film before.
Cybersecurity became "a thing" a few years ago, and almost overnight everyone was a cybersecurity expert. The "cyber" word was everywhere, the fear was everywhere, and the gap between the people using the word and the people who actually understood it became very wide, very quickly.
AI is running through the same cycle, only louder.
Part two: the ones who can see it often don't trust it
Among small businesses that are paying attention, the concerns are not vague or technophobic.
They are specific, and they are sensible: inaccurate answers, data security, the misuse of intellectual property, regulatory uncertainty and not knowing what happens to the information they provide.
That isn't resistance to technology. It is a reasonable response to being asked to trust something that's both powerful and difficult to evaluate.
There's an old line about a fool and his money being easily parted. Small-business owners aren't fools.
If you cannot evaluate something, you are right to be wary of paying for it. Caution in the face of confusion isn't stupidity. It's good sense.
So put the two groups side by side.
On one side are people who cannot see a use for AI.
On the other are people who can see its potential but don't trust it.
What do they have in common? Nobody is properly answering either question.
Nobody is showing the first group where the useful work is.
Nobody is showing the second how the risks can be controlled.
So they do the rational thing. They look at the whole business, mark it "here be dragons", and decide not to go there.
Part three: they've been shown the weakest version of the tool
Here's the part I most want business owners to understand, because it is the part the noise completely skips.
When a business tells me AI is unreliable — that it makes things up or that its information may be out of date — they're not wrong.
But they've often encountered the weakest possible use of it.
They are judging a chat window asked to answer from what it learnt during training.
You ask it a question. It constructs an answer from patterns in what it has previously absorbed. Sometimes that answer is excellent. Sometimes it is confidently wrong or quietly out of date.
That kind of general-purpose question answering is the party trick. It's easy to demonstrate, which is exactly why the bandwagon-jumpers love showing it off — and exactly why it burns through people's trust so quickly.
The more powerful use begins when you stop asking AI simply to know things and give it real information to work with.
Your quotations.
Your emails.
Your spreadsheets.
Your accounts.
Your documents.
Analyse this.
Process that.
Draft this using that.
Pull the numbers from these documents and check them against those records.
The facts no longer have to come solely from the model's general training. You have given it a defined source of truth.
That doesn't make it infallible. It can still misunderstand, omit something important or draw the wrong conclusion. Its work still needs controls and checking.
But it changes the nature of the problem.
Instead of asking a machine to remember the right answer, you are asking it to work on information you possess, can inspect and can verify.
That's a much more useful — and manageable — proposition.
They have been shown a machine that answers questions.
What they haven't been shown is a tool that can take real business information and help analyse it, classify it, transform it and move work forward.
And almost nobody has properly explained the difference.
So where does that leave the owner who has quietly concluded, "Not for us"?
I understand exactly how you got there.
You've been shouted at. You've been shown a party trick that sometimes makes things up. And you've made the sensible decision not to spend money on something nobody has properly explained.
That's not a failure on your part.
It's a failure of everyone who would rather shout about revolution than show you a better Tuesday.
But I would ask you to separate two very different statements.
"I looked properly, and there's nothing here."
That is a fair conclusion.
"I've been shouted at so much that I've stopped listening."
That's not quite the same thing.
The question was never really: is AI relevant to my business?
It's: has anyone actually looked, and shown me the real thing rather than the trick?
Start with one problem. Just one.
The thing that eats your evenings.
The thing that slows your quotations down.
The process that depends on retyping the same information three times.
The knowledge that lives inside one person's head and worries you whenever they take leave.
Then ask, not "can a chatbot answer questions about this?" but:
"Can I give a system my real information and use it to help do this work?"
That's a different question.
And it has a very different answer.
Source: Department for Science, Innovation and Technology, AI Adoption Research, published January 2026. The research covered UK businesses employing five or more people; survey fieldwork took place between February and May 2025. Figures were correct at the time of writing.
This article first appeared on LinkedIn (opens in a new tab), which is where the discussion is.