Should we be doing something with AI?

Almost certainly, yes. But before you act on that, notice how the question is usually framed. "Doing something" - anything - is precisely how businesses end up with a drawer full of subscriptions and nothing to show for them. The useful version of the question isn't "should we be doing something?" It's "what, specifically, would be worth doing here?" The rest of this article is about answering that second question without betting the business - and without pretending the whole thing will blow over, because it won't.

Start with the honest numbers

AI adoption among UK SMEs has climbed fast - though the figure depends heavily on how you count. The government's own research puts strategic deployments at roughly one in six firms. Broader surveys that count any use at all get well past half. Either way, the direction is clear and the pace is quick.

But here's the figure that matters more. While around three quarters of adopters report a productivity gain, only about one in eight report a revenue increase. Read that twice. AI is helping people get work done faster. It is not, by itself, making most businesses more money. The gap between those two numbers is where the good decisions get made - and where most of the wasted spend happens.

Why the gap? Because a productivity gain is easy to feel and hard to bank. An hour saved here and there across a team doesn't turn into margin on its own - someone has to decide what those hours are now for. The firms converting AI into money treat it as a capacity question, not a gadget question.

Where AI for small business actually pays off

Saving your team a couple of hours a week is real, but it isn't a strategy. The businesses getting genuine value from AI aren't the ones who bought the most tools. They're the ones who picked a specific, expensive problem - a process that eats days, a bottleneck that caps growth, a job nobody wants to do - and pointed AI squarely at it.

In an established business, those problems tend to look like this: the quote that takes two days to assemble from six sources. The invoices someone retypes into the accounts system every week. The routine customer queries that interrupt your best people a dozen times a day. The month-end report that exists mainly to be compiled, not read. Dull, repetitive, information-heavy work - that's where the current generation of AI genuinely earns its keep. None of this is glamorous. That's rather the point: the glamour is where the wasted spend lives; the payback is in the plumbing.

What doesn't pay off: buying a chatbot because a competitor has one, or licensing a tool for the whole company because the demo was impressive. The question was never "how do we use AI?" It's "where is it actually worth using?" Those are very different questions, and only one of them leads anywhere useful.

The hype tax

Move too fast and you pay what we'd call the hype tax. Tools bought on a wave of enthusiasm and quietly abandoned by month three. Subscriptions stacking up next to the ones you already weren't tracking. Sensitive data pasted into systems nobody vetted - often by well-meaning staff who started using AI on their own because nobody gave them a sanctioned option.

AI doesn't get a pass from the discipline you'd apply to any other investment: what's the problem, what's the return, what's the risk, and who owns it. Those four questions kill most bad AI ideas in ten minutes, which is exactly what they're for. Skip them and you don't get a strategy - you get a bigger software bill.

Why does a messy estate make AI harder?

Here's the part the vendors tend to skip. AI works best on top of systems and data that are already in good order. If your information is scattered across a dozen tools that don't talk to each other, AI doesn't fix that - it inherits it. Point a clever tool at a messy estate and you get fast, confident answers built on bad information - which are worse than slow ones, because they look right.

The same goes for access. Most AI tools that work across your business see whatever your permissions let them see - and in most businesses, those permissions are years out of date. The groundwork that makes AI pay off is the same groundwork that makes everything else work better: knowing what you've got, and getting it into order. We've written separately about what bolting AI onto a messy estate actually costs - it's not a small number.

What does a sensible first step look like?

You don't need an AI strategy. You need clarity on your business and your estate, and then a short list of places where AI could remove real cost or unlock real capacity.

Pick one. Run it as a proper, measured trial: a specific problem, a number you're trying to move, a date to judge it by, and one person who owns the outcome. Give it six to eight weeks, not six months. And measured means measured - before the trial starts, write down what the process costs today, in hours, errors or lead time, so "it feels quicker" never has to stand in for evidence.

If it works, you'll have proof - real hours saved, a real cost removed - and a business case for the next one. If it doesn't, you'll have spent little and learned a lot, which beats the usual alternative of spending a lot and learning nothing. Then decide where next. That's how you capture the upside without betting the business on a trend, and without quietly funding a drawer full of tools nobody opens.

The straight answer

"Should we be doing something with AI?" Almost certainly - but something specific, chosen on purpose, grounded in how your business actually runs. Not because the board asked, and not because everyone else seems to be moving. Because there's a real problem in your business that this technology is genuinely good at removing.