Nobody can quote you a price for AI, because AI isn’t a thing you buy. It’s a category, like “vehicles”. A van and a forklift are both vehicles and you wouldn’t ask what a vehicle costs.
So the question owners are really asking is narrower than it sounds: what will it cost me to stop doing this particular job by hand, and will that be less than what the job costs me now. That one has an answer.
This post breaks the bill into its five lines, puts a UK figure against each, and gives you the arithmetic to produce your own number by the end of the afternoon. If you’re at an earlier stage and still working out what AI does inside a company your size, start with the complete guide and come back to the money.
What UK businesses are actually spending
Start with the market rather than a supplier’s slide.
Lloyds’ Business Barometer found two thirds of UK businesses have now invested in AI, and a third of them spent under £25,000. Another 18% spent between £25,000 and £100,000, 8% between £100,000 and £250,000, and 7% went past a quarter of a million. On the return side, 87% reported higher productivity and 48% reported higher profits.
Two things worth pulling out of that. The first is that the biggest band is the smallest spend, so a business your size is not unusual for keeping this in four figures or low five. The second is that “invested in AI” covers everything from a company ChatGPT account to a bespoke system, and those sit at opposite ends of what you get back.
The adoption numbers say the same thing from another angle. The ONS puts AI use among UK businesses with 10 or more employees at around 35% in June 2026, up from roughly 12% in late 2023. Wide adoption, shallow use. Most of those businesses bought seats and stopped.
The five lines on the bill
1. Seats
The per-person subscriptions. This is the line everyone knows about because it looks like every other software bill.
Microsoft 365 Copilot lists at £23.10 per user per month paid yearly on top of what you already pay for Microsoft 365. For 20 people that’s £5,544 a year. Add a few ChatGPT or Claude seats for the people who live in them and you’re comfortably past £7,000 before anything in your business works differently.
Seats are easy to approve and easy to justify. They’re also the line with the weakest link to any outcome, because what you’ve bought is a faster typist for each member of staff, not a process that runs without one. Useful. Just don’t confuse it with the thing that takes work off the payroll.
2. Somebody’s time to build the thing
When you want a process to run on its own, you’re buying engineering time, and that’s priced like engineering time.
YunoJuno’s 2026 freelancer rates report, built on more than 182,000 workforce data points, puts the average UK developer day rate at £438, with the top 10% of contracts averaging £654. UK agencies publishing their own ranges land in a similar place: ExpertSure’s 2026 guide puts typical SME automation projects at £5,000 to £30,000, and Sharp Code puts a focused first build at roughly £15,000 to £60,000.
Those are wide ranges for a reason. The same job costs very different amounts depending on how tidy your data is and how many systems it has to touch. Anybody quoting you inside 10 minutes has picked a number, not worked one out.
The bigger question on this line isn’t the rate, it’s who carries the risk of it running long. Day rates put that on you: every unforeseen problem becomes another invoice. How to choose an AI consultant goes through what to ask before you sign either kind of agreement.
3. Running costs
Software that uses AI models costs money every time it runs. Model usage, hosting, and whatever tools sit underneath it. For a small business system this typically lands somewhere around £150 to £400 a month, and it moves with volume rather than sitting flat like a subscription.
This line catches people out because it doesn’t behave like anything else in the budget. KPMG’s Q2 2026 AI Pulse survey of 2,145 senior leaders found a third cite limited understanding of AI cost structures, including how token-based pricing works, as a major problem when deploying agents. If a proposal doesn’t state the expected monthly running cost and what makes it go up, you’ve been handed a budget with a hole in it.
4. Your team’s time
The line that never appears on any quote and always gets spent.
Somebody has to explain how the process really works, find the login for the system nobody has touched since 2023, and check the first month of output. On a sensible project that’s a handful of hours a week for a few weeks from the people closest to the work. It’s the cheapest way to spend your team’s time on this and the most common thing to underestimate.
Price it properly. Using the loaded hourly cost we work through in the cost of manual work, an hour of an average UK employee’s time runs about £26.55 once you add employer NI and pension. 20 hours across a project is roughly £530 of real cost. Small next to the build, but it’s real, and pretending it’s free is how projects end up “over budget” when they were only ever under-counted.
5. Keeping it working
Your business changes. Prices move, a supplier changes their form, someone rewrites the approval rule. Software that was right in March is wrong by October unless somebody maintains it.
UK agencies commonly budget 15% to 20% of the build cost a year for this. Whether you pay that to a supplier on a monthly agreement or absorb it internally, you’re paying it. Anyone selling you a system with no line for upkeep is selling you something that works for one year. We wrote about why AI agents rot and what keeps them running.
The line that actually ruins budgets
None of the five above is the reason AI spend goes wrong. The reason is that almost nobody counts.
Flexera’s 2026 State of ITAM report found only 31% of organisations have accurate visibility into their AI software spend, and 59% say wasted AI spend has risen year on year. KPMG’s numbers point the same way: 42% of leaders report only partial visibility into what AI costs them, and 49% have scaled back, narrowed, delayed, or paused agent deployments once expected costs outran expected value.
The most useful figure in that survey is one nobody quotes. Organisations with full visibility into their AI running costs reported established returns at 15%, against 3% for those without. Same technology, five times the hit rate, and the difference is that somebody was counting.
For a business your size the practical version of that is easier than it sounds. One spreadsheet. Every AI subscription, who approved it, what it costs a month, and what it was bought to do. Most owners we do this with find two or three lines they’d forgotten and at least one nobody uses.
Why “just buy seats” is often the expensive answer
Seats look like the cautious choice. Low monthly cost, cancel any time, nothing to build.
The trouble is what they leave in place. A subscription helps the person doing the job go faster. The job still needs that person, still happens at their pace, still stops when they’re on holiday, and still costs you a salary. You’ve bought a discount on the work, not a way out of it.
That’s most of the gap the ONS numbers describe: wide adoption, shallow use, and very little of it showing up in anybody’s accounts. Gartner has predicted that 30% of generative AI projects get abandoned after proof of concept, and the pattern behind pilots that never ship is nearly always the same: nobody owned it, and nobody had costed the before.
Seats are worth having. They’re a poor answer to “this process costs too much”.
Work out your own number this afternoon
The honest cost of AI isn’t the quote. It’s the quote measured against what the manual version already costs you. Here’s the arithmetic.
Step 1. Pick one process. The one your team complains about. Not the most interesting one, the most repeated one. Which processes to automate first sets out how we rank them, and there’s a ranking tool if you want to work through several.
Step 2. Count the hours. Ask the person who does the job, not the person who manages it, how many times a week it happens and how long it really takes. The gap between the two answers is the point, and it’s nearly always in the same direction.
Step 3. Put a real hourly cost on it. Salary plus employer NI plus pension, divided by actual working hours. The manual work cost calculator does it in a couple of minutes.
Step 4. Multiply out over 3 years. That’s the number the build is competing with.
Say 2 people each spend 6 hours a week on quoting. That’s 12 hours a week, 552 hours across a 46-week working year, and at £26.55 an hour it’s about £14,700 a year. Over 3 years, roughly £44,000. Those are illustrative figures, but run your own and you’ll usually find the manual version is the expensive option by a distance nobody had noticed.
Now compare that against whatever you get quoted. Hold it next to the published market ranges above, add the monthly running cost and the annual upkeep, and total the same 3 years. On numbers like the ones in that example, a first build at the lower end of the ranges those agencies publish comes in under the manual version inside the 3 years, and the hours come back every year after that. A build that costs more than the process it replaces is one you should refuse, and a decent supplier will tell you before you ask.
What brings the number down
Do one process properly instead of five badly. The first system carries all the setup cost. The second and third are cheaper because the plumbing exists.
Get it costed before you commit. An AI audit exists to turn “we should do something with AI” into a specific process, a specific number, and a specific first job. Check whether the fee is credited against the build, because that changes the real cost of finding out.
Check whether somebody else will pay for part of it. If you’re a UK manufacturer with between 10 and 249 staff and turnover under £50 million, Made Smarter Adoption offers match-funded grants of up to £20,000 towards digital technology, alongside funded advice and training. It’s regional and it’s manufacturing-specific, so it won’t apply to everyone, but it’s worth 10 minutes of checking before you spend your own money.
Buy a fixed price. Not because it’s always cheaper on paper, but because it moves the risk of the project running long onto the person who controls how long it runs.
When the honest answer is “spend nothing”
Sometimes the sums don’t work, and a supplier worth hiring will say so.
The process changes substantially every month, so anything built for it gets rebuilt at your expense. Or it runs 4 times a year and the hours don’t add up to anything. Or the fix is a feature of a tool you already pay for that nobody ever set up.
We turn work down on all three. What AI can actually do for a small business covers where the line sits, and build vs buy for AI is the decision itself: which parts to rent, which to own, and what both cost once you count the hidden half. If a supplier has never once told a client not to build, ask them why.
What we charge
We sell this, so read the following as a description of how we work rather than neutral advice.
We don’t publish prices, because a number without your process behind it is a guess dressed up as a quote. What we commit to instead is fixed: one price agreed in writing before anything starts, no day rates, no scope-creep invoices, no paid discovery phase. Half to start, a working prototype inside 60 days, and a full refund if you don’t approve it. The AI audit fee comes off a build in full if you start one within 90 days.
The proof that the sums work: Founderise took 12 hours a week off its founder and lifted margin 3.5x, with an MVP live in 9 weeks. MidShift has served more than 20,000 professionals and moved people through career progression 92% faster.
You get your number on the call, and it holds.
The short version
There’s no price for AI, only a price for a specific job you want it to stop doing by hand. The bill has five lines: seats, build, running costs, your team’s time, and upkeep. Most businesses only ever pay the first, which is why most of them have nothing to show for it.
Two thirds of UK businesses have spent money on AI and a third of those stayed under £25,000. The ones getting a return are the ones who counted the before, picked one process, and knew what the monthly running cost would be before they signed.
Count first. The manual version is usually the expensive option, and until somebody adds it up, nobody in the building knows that.
Related reads
- Shadow AI: what your team is already doing with company data
- AI for small business: the complete guide
- The cost of manual work: what repetitive admin really costs
- Which processes to automate first (and how to rank them)
- What an AI audit is and what you should get from one
- How to choose an AI consultant (and the red flags)
- Build vs buy for AI: when off-the-shelf is genuinely fine
- Why most small business AI pilots never ship
- What AI can actually do for a small business (and what it can’t)
- AI agents for business: what they actually do
- Case study: how MidShift built an AI career guidance engine for 20,000+ professionals
- Case study: how Founderise automated delivery
Want your own number before anybody quotes you one?
Run the manual work cost calculator on your worst process and walk into every supplier conversation holding a figure. Then book the call and bring it. We’ll tell you what we’d build, what we wouldn’t touch, and whether the sums justify either.
Book a free call, or read what the AI audit covers in detail.