Somebody has probably rung you about AI by now. They will not have told you what it costs, because the answer in the sales call is always that it depends.
It does depend. But there are real numbers underneath it, from firms actually paying the bills rather than from anyone selling the tools. We read the primary sources on the 22nd of October 2026, so the figures below are as they stood that day.
How much should you actually pay?
The figure doing the rounds is £75.60 a quarter. It comes from Capital on Tap, one of the bigger UK small business lenders, which went through its own card transactions between early 2023 and the middle of this year.
One thing worth saying about the source before we use it. Capital on Tap sells credit cards. The chief executive's own comment alongside the research notes that a card can help when several subscriptions land at once. It is still the best data available on what small firms actually pay, because it is what they actually paid. Just notice who is holding the tape measure.
Two things get lost on the way from that research to the headline.
It is not an average, it is the midpoint. City AM was given the same data in August. The mean is £288 a quarter, up from £93 two years earlier. It is nearly four times the midpoint because the heaviest one percent of spenders pay £3,159 a quarter each, which drags the average up for everybody else.
It is not per small business, it is per small business already buying AI. Firms paying nothing are not in the sum at all.
So the honest version. Among small firms that do pay for AI, half spend under £76 in a quarter. That is about £25 a month, which is a quarter of a single shift, or roughly what you pay for one card machine terminal. A few spend thousands. Averaging those two groups gives you a number that describes neither.
If you are working out a figure for next year, start at the £76. The £288 describes a room you are not standing in.
Why your trade is at the back of the queue
Capital on Tap also split adoption across 24 industries. Nothing else in the data spreads this wide.
Publishing leads at 26 percent. Radio and television sits at 21.6 percent, as does consulting. At the bottom of all 24, Restaurants and Food, at 4.8 percent. Transport and Food Retailers are barely ahead at 4.9.
More than one in four publishers pay for AI. Fewer than one in twenty of the restaurants in that dataset do.
I spent most of my working life in kitchens before I did this, so let me put the useful part plainly. If you run a pub, a café, a restaurant or a small hotel and you have the nagging sense that everybody else has worked this out while you have not, the numbers say the opposite. Your trade is last, for a good reason. Most of a hospitality week is physical and in front of a customer. The work is not text on a screen, which is the kind these tools handle best.
Two honest limits on that 4.8 percent before you lean on it. It counts firms paying for AI on a Capital on Tap card, which is not the same as firms using AI. The company says in its own methodology that the figures should not be read as a measure of total UK adoption. The Office for National Statistics measures a broader thing, any use of any AI technology, so it gets much higher numbers across the board. Both can be true. Paying is a higher bar than using.
The bill does not grow because the tools get dearer
This is the part worth pinning above the desk, because it changes what you watch.
Capital on Tap tracked the typical amount spent on each individual AI purchase. In just over three years it moved from £16.65 to £18.10. Barely anything, well under inflation.
What did move is how many subscriptions a firm holds. In early 2023, 11.6 percent of the firms buying AI paid two or more providers. By the middle of this year it was 28.2 percent.
Put those together and the picture is obvious. The headline price barely moved. You just bought another one, then another, each one small enough that it never seemed worth querying.
Anybody who has gone properly through a card statement knows that shape. It is the same thing that happens with the booking widget, the rota app, the stocktake tool and the thing the old manager signed up for that nobody has opened since. The line never looks worth cancelling. The column does.
So the thing to put in the diary is not a price check. It is a count. Once a quarter, list what you pay for and write a name next to each one. Anything without a name against it, cancel.
The second bill nobody tracks
Scott Downes, chief executive of Supernal, quoted in TechRound in September, put it better than I would. Most firms, he said, pay for software twice. There is the subscription everyone knows about, then there is the second bill nobody tracks: the hours spent setting it up, stitching it to something else, then helping people use it.
In a kitchen that second bill is the expensive one. A tool that saves your office ten minutes a day but needs twenty minutes a week of somebody's attention has not saved you anything. It has just moved the work to a different person, usually you, usually after service.
The bill that does not hold still
Subscriptions are predictable. Usage is not.
Pax8's Q3 2026 report, published on the 12th of October, found that 49 percent of the businesses which had paid for AI in the past year got at least one bill or usage charge higher than they expected. TechRadar reported a UK figure of 39 percent from the same research. The two are not published on the same basis, so treat them as the same warning at two scales rather than as Britain against everybody else. Usage based charges are the hardest to forecast, because they grow quietly as a tool gets connected to more of what you do.
One caveat that matters for you. Pax8 surveyed businesses employing between 5 and 499 people, 350 of them in the UK. A café with three staff is outside that sample. Take the direction as real, the exact number as belonging to somebody bigger.
The same report found that of UK firms building AI into how they work, only 23 percent have a written, documented review of it. That is the lowest of the four countries surveyed. Whatever you end up paying, somebody should be checking what it actually does.
The number that puts the rest of it in proportion
Across everything Capital on Tap processes, AI accounts for 0.1 percent of the value of all card spending. City AM worked that out to roughly £4 billion a year across the UK, which it pointed out is less than a tenth of Tesco's turnover.
That share is climbing fast, almost thirtyfold since early 2023 on Capital on Tap's own figures, so it will not stay at 0.1 percent. It is still a very small number today.
Hold that against the noise about the economy being remade. It may well be, in time. It has not yet reached the card statements of ordinary British businesses. You are not behind for being one of them.
Does it solve a problem you already have?
That is the right first question. It is the one owners on the trade forums ask each other rather than the one the sales call asks you.
Start at one tool, one person, one job. About £25 a month puts you at the midpoint of what firms already buying AI actually spend. It is a real budget rather than a token one.
Give it a job with a number attached. Not "help with admin". Something like the time it takes to answer booking enquiries, or the hours that go on rotas every week, or chasing a supplier invoice.
Put a date on the review. A quarter is about right. Two questions at it. Did it save the time you hoped? Is anybody other than you using it?
Count before you add. The research says the risk is not a price rise. It is the fourth subscription nobody decided to buy.
And if the honest answer is that nothing on offer fixes your actual problem, that is a legitimate answer. Most of what slows a venue down is a rota, a supplier, a layout or a person. No subscription touches any of those.
The real cost of AI in hospitality is rarely the £25. It is the six months paying for something that was never going to fix the thing that is actually costing you. That is why we tell people to find the bottleneck before they buy anything. The bottleneck audit costs nothing and you can run it yourself in an afternoon.
If you are in Wales, there is also free support worth using before you spend. We went through what is actually on offer and where it stops. Where public money exists, we set out where it really is rather than where the press releases say.
If you would rather somebody looked at your own week first, the hospitality side of what we do starts exactly there. We cover the whole of Wales and will come to you where that helps. Get in touch for a free consultation. If the answer is that a £20 subscription and an afternoon of your own time would do it, we will say so.