AI isn't sold "per licence". You pay for what you use. And that changes everything.
If you're considering AI for your company, you probably have the wrong cost model in mind. You're thinking of an annual licence, like the one for your management software or your CRM. Five, ten, twenty thousand euros — whether you use it a lot or a little, that's what it costs.
AI works in a completely different way. And that difference is some of the best news SMEs could hope for.
How AI pricing works
Every time you send a question to an AI model through an API — the technical interface that lets you build AI into your processes — you pay for the "tokens" used. Tokens are, roughly speaking, pieces of text: the words going in (your question) and coming out (the AI's answer).
How much are we talking about? For a complex question — say, "analyse this sales report and tell me the three main trends" — with a detailed answer, the cost is in the region of €0.01–0.05. Yes, cents.
That means you can make hundreds of requests a day and spend a few euros a month. Not tens of thousands of euros. Euros.
Why this changes the rules of the game for SMEs
Paying for what you use removes the biggest barrier to entry: the risk of an upfront investment.
With traditional software, you have to commit budget before you know whether it will work for you. With pay-per-use AI, you can start experimenting for the price of a coffee a day. Automate a single process — sorting incoming emails, for example, or drafting replies to customers — measure how much time it saves you, then decide whether to expand.
If it works, you scale up. If it doesn't, you've spent a few tens of euros, not a few thousand.
That's radically different from the "buy the licence and hope it works" model we're used to.
The other side of the coin: usage has to be designed
Pay-per-use has one side that needs careful handling. If a process is poorly designed — an AI agent that makes too many unnecessary calls, regenerates answers several times or processes data it doesn't need — it costs you more than it should.
The difference between a well-designed process and a poorly designed one can be 10 or 20 times the cost. We're still talking about manageable amounts — the difference between €5 a month and €100 a month, not between €5,000 and €100,000 — but the principle matters: using AI wisely is part of the strategy.
That's why it makes sense to work with someone who knows the models, knows how to optimise prompts and designs efficient processes. Not for the technology itself, but to avoid waste.
A concrete example: what it costs to automate a weekly report
Imagine a weekly sales report that someone currently spends 2 hours putting together. They pull the data from the management software, organise it into tables and write a comment on the main trends.
With AI, the same process could work like this: an agent reads the data, builds the tables and writes the comment. All in a few minutes. The cost in tokens? Around €0.10–0.30 per report, depending on complexity.
Do the maths: €0.30 × 52 weeks = €15.60 a year, against 2 hours × 52 weeks × €25/hour = €2,600 a year of people's time.
Of course there's an upfront cost to set up the agent and connect it to your data. But the recurring running cost is close to zero.
It's not machinery. It's electricity.
The best way to think about the cost of AI isn't industrial machinery. It's electricity. You pay for what you use. There are no huge fixed costs. You can switch it on and off whenever you like.
As with electricity, the secret isn't paying less per kilowatt-hour — token prices keep falling and are already very low. The secret is designing processes that use only what they need. No waste, no duplication.
For an SME, this is liberating. You don't need to persuade the board to set aside €50,000 for "the AI project". You can start tomorrow with the stationery budget.
The takeaway
AI isn't an investment to assess like a piece of machinery. It's more like electricity: you pay for what you use, and the secret is using it wisely.
If you're putting off AI because "it costs too much", you're probably using the wrong cost model. The cost of the technology is close to zero. The real cost is the time you lose by not using it.
This is the second article in Fleze.ai's AI Decoded series. In each article we explain a key concept of artificial intelligence in practical terms, without jargon, for business owners who want to really understand how AI works — and how to use it.
→ Follow us on LinkedIn so you don't miss the next one.