A lot gets written about AI and little gets priced. Every provider talks about possibilities, nobody names a figure. That's understandable — it depends on things — but it's no reason to stay vague. The cost of AI in a business is set by a handful of concrete things, and once you know those, you also see why one quote comes to €2,000 and another to €40,000.
What determines the price?
What you actually want to achieve
This is the biggest factor, and the most skipped question. "Doing something with AI" is not a goal. Saving two hours a week on quotes is a goal. Answering customer questions outside office hours is a goal. The sharper the goal, the smaller the project — because then there's no need to spend six months exploring first.
Licences: the layer that keeps running
Reckon on €20 to €60 per employee per month for a business AI subscription. That seems little, until you multiply it by your team and add two years. For ten people that's €5,000 to €14,000 over two years, and that's separate from anything that gets built.
The setup
Here's the real work: figuring out which data may go where, making connections with the systems you already have, and making the output reliable enough to lean on. One process that actually works costs more than five that half-work — but it's also the only one that delivers.
Getting the rules in order
Since August 2026 the AI Act requires a register, policy and demonstrable AI literacy from virtually every business that uses AI. Bought separately, that's a manageable line item. Taken along in a project you're doing anyway, it largely disappears into the whole.
Bringing your team along
The most expensive AI is the AI nobody uses. Explanation about your own processes costs a session or two and determines whether the rest of the investment pays off. This is the item that gets cut first and is most often the reason a project delivers nothing.
Rough guidelines
- Rules in order (register, policy, session): €1,500 – €3,500
- One process automated and working: €2,500 – €7,500
- Setup across multiple processes, with integrations: €7,500 – €25,000
- Site or system set up for management with AI: €6,750 – €30,000
- Licences, ongoing: €20 – €60 per person per month
Where does it go wrong?
Twenty subscriptions, no policy
The most common pattern: everyone tries something, nobody keeps track. After a year you're paying for seven tools of which two are used, customer data sits in three services whose terms you never read, and you can't explain how it's arranged. An afternoon's inventory nearly always pays for itself here.
The pilot that never ends
Exploring is comfortable. There's always another tool to try and another use to investigate. But a pilot without an agreed end point and without a measurable goal isn't a pilot, it's a subscription to research. Agree up front when you decide to continue or stop.
Paying for a demo
There are parties who build an impressive demo that doesn't survive in practice, because real data is messier than sample data. Always ask whether it can be shown on your own data. If it doesn't work there, it doesn't work.
Everything at once
Tackling five processes at once costs more than five times one process, because everything becomes uncertain at the same time. Start with the process that takes the most time and carries the least risk. What you learn there makes the second project cheaper.
What does it deliver?
In the right place the sum is simple. A process that takes two hours a week and gets automated for eighty per cent frees up well over eighty hours a year. At an internal hourly rate of fifty euros that's €4,000 a year, against a one-off investment.
But not every process is that candidate. Work that's different every time, requires judgement, or where a mistake is costly, is better left alone. The gain is in the boring, repetitive, predictable part — and that's usually exactly the part nobody wants to touch.
How do you choose a party?
Ask what they run themselves
Anyone who sells AI but deploys it nowhere themselves is selling a story. Ask about their own way of working, their own tooling, their own mistakes. That conversation says more than a portfolio.
Ask what happens when you stop
The most important question, and rarely asked. Can you carry on without them? Do you get the setup, the rules and the documentation, or does that stay in their management? A solution that only works while the builder is watching isn't a solution but a subscription.
Ask about the dependency
What happens if the AI provider doubles its prices or shuts down? If the answer is "then we have a problem", you know enough. A good setup works without the AI too — it only makes the management easier.
Watch out for anyone who starts on savings without knowing your process
Percentages named before anyone has looked at how you work come from a sales deck. An honest estimate follows a conversation about your work, not one ahead of it.