What an AI tool really costs to run (not the sticker price)

The price on an AI tool's pricing page is not what it costs you. These tools charge for the work they do, not just for the login, so the real bill lands after you have committed, and it is that running cost, not the monthly fee, that decides whether the tool ever pays for itself. If you only compare sticker prices, you are comparing the wrong number.

Why the subscription price hides the real cost

Most business owners size up a new AI tool the way they size up any other software: what is the monthly fee, and can we afford it. That instinct is fair, and for ordinary software it works, because the cost is basically fixed no matter how much you use it.

AI tools do not behave that way. Under the surface, most of them meter usage. Every document summarised, every email drafted, every question answered burns a little bit of compute, and you pay for that compute. A tool with a modest licence can quietly cost far more once it is genuinely running all day across your team, and a tool that looks "cheaper" up front can turn out dearer per useful result.

So the sticker price tells you the floor. It does not tell you the bill.

Even Microsoft got caught by running cost

If that sounds like a trap only small businesses fall into, it is not. Earlier in 2026 Microsoft cancelled access to an AI coding tool that its own developers had come to prefer, and moved them onto a cheaper in-house option instead. This is a company that helps build these tools, reporting to shareholders that a large share of its code is now written with AI assistance. The tool it cut was not worse. Its developers liked it more. But the running cost did not hold up once real, everyday usage was counted, and the cancellation landed right on the company's financial year-end.

The lesson for a smaller business is blunt. If the running cost caught Microsoft, with all its buying power and all its data, it can catch you, and you are the one who only saw the headline price on the way in.

What running cost actually decides

Here is the reframe we use. The useful question is not "what does this cost to buy." It is "what does this cost to run for the work we would actually give it." Those are different numbers, and the gap between them only shows up after you have signed on.

Sticker price is a one-off decision you make once. Running cost is a decision you keep making every single day, whether you meant to or not, every time the tool does a job. A tool you can run predictably and sustainably is usually a better business decision than a flashier one whose bill moves around with how busy you are.

This is the same logic behind why the return on AI shows up in weeks, not months when a tool is chosen and scoped properly. Get the running cost right and the payback is clear and quick. Get it wrong and even a capable tool becomes a line item you resent.

The one question to ask before you commit

Before you sign up for any AI tool, ask the vendor the thing they usually do not lead with: what does this cost to run at our expected volume, not what does it cost to subscribe. Ask them to walk you through a realistic month for a team your size. If they can give you a straight, specific answer, that is a good sign. If they cannot, or they steer you back to the plan price, treat that as your answer.

At Handiwork we run that test before we ever put a tool in front of a client, because protecting the budget is part of the job, not an afterthought. Choosing tools you can afford to actually use is also the most sensible next step for most businesses adopting AI: not more tools, but the right one, run within its means. If you want a hand working out where AI genuinely earns its keep in your business, that is exactly what our advisory work is for.

Ready to find out where you stand?

If you want a straight read on what your current AI tools actually cost to run, and where AI would genuinely pay off, start with our free AI Readiness Check. No cost, no pitch.