In short
- Only the diagnosis can be priced unambiguously — because its format and outcome are fixed in advance. A first implementation is priced after assessment, not from a price list.
- Three things invisible before assessment drive the price: process complexity, the state of the data, and the depth of integration with existing systems.
- A firm quoting a first implementation without looking at those things is either guessing or has built in a margin for its own uncertainty.
- Invoice price ≠ total cost. The most expensive item is usually a badly chosen project: a tool nobody uses, plus dependence on the supplier when the knowledge does not stay in the company.
- Staged billing with a decision point after the first stage limits the cost of a mistake to one stage rather than the whole budget.
The one number I can give you on the spot
There's one stage that prices itself before you've settled anything else: diagnosis. For me that's the AI Readiness Audit — starting at 3,200 PLN net, one to three days of work on-site depending on the company's scale. The price is transparent because the scope is fixed up front: a set format of work and a set output — a report with an implementation map and a return estimate.
The audit is also the cheapest way to avoid overpaying for everything that comes after. Without it, the first implementation gets picked by gut feel. With it, you know which process to support first and roughly what scale of effect it will bring — and that translates directly into a sensible budget.
Why an implementation can't be priced from a rate card
The first project itself is a different matter. Here, "it depends" sounds like a dodge, but it's an honest description. The price of a first implementation depends on several things you can't see before you look closer.
Process complexity is one part of it — supporting a single repeatable task takes different effort than a workflow full of exceptions. Then there's the state of the data: organized and accessible data lets you start right away, while data scattered across spreadsheets, emails and people's heads has to be gathered first. And finally, integrations — a solution wired into an existing ERP system costs more than a standalone tool. A company that quotes a price for a first implementation without looking at any of this is either guessing or has built a buffer into the price for its own uncertainty. That's why, in my case, the first implementation gets priced after a consultation, not off a price list.
One thing about the mechanics is worth knowing. The first pilot implementation is billed in stages, with a decision point after the first week. You don't put the whole budget on the table blind — after the first stage you can see whether the project is heading the right way, and only then does the decision to continue get made.
What actually costs the most in this equation
The most expensive line item in an AI implementation is rarely on the invoice. It's the cost of a bad project — the money and months sunk into a tool nobody uses. I've seen companies spend more on a license for a system used at one-tenth capacity than a dedicated project solving their actual problem would have cost.
On top of that come costs that are easy to miss when you're just looking at the offer: the internal team's time, which has to go in regardless, maintenance and further development after go-live, and finally, dependency on the vendor if the know-how never stays in the company. A cheaper offer that leaves the company with a tool it doesn't know how to run can end up more expensive over a year than a pricier one that leaves it self-sufficient. I've laid out a fuller comparison of these paths — build it yourself, hire a software house, buy an off-the-shelf product — separately.
How to cut the bill in half
For manufacturing SMEs there's a real lever here. The Dig.IT grant covers up to 50% of eligible implementation costs, and building a dedicated solution falls within its mandatory component. On a well-structured project, half the bill goes to the grant. Exactly what the grant funds, what it doesn't, and how to structure a project around it — I break down in the article on Dig.IT's eligible costs.
What we don't know yet
Before acting on this article, it is worth checking whether the answers to these questions are known on your side:
- what state the data describing the process is in — it moves the estimate up or down more than anything else;
- whether the solution has to plug into an existing system or can run alongside it;
- how much internal team time the project will consume — a line that appears on no quotation;
- who maintains the solution after go-live, and what that costs per year;
- what happens after the first stage if the result is not convincing.
Frequently asked questions
How much does an AI implementation cost?
Only the diagnosis stage has an unambiguous price, because its scope and outcome are fixed in advance. The cost of a first implementation depends on process complexity, the state of the data and the depth of integration — which is why it is set after assessment rather than from a price list.
Why will nobody quote an implementation straight away?
Because the three factors driving the effort are invisible until the process has been seen: whether it is one repetitive task or a flow full of exceptions; whether the data is accessible or scattered across spreadsheets and inboxes; and whether the solution has to plug into an ERP or can stand alone.
What costs the most in an AI implementation?
Rarely anything on the invoice. The most expensive item tends to be a badly chosen project — money and months spent on a tool nobody uses. Add the internal team's time, maintenance after go-live, and dependence on the supplier if the knowledge does not stay in the company.
Can an implementation be financed from a grant?
For manufacturing SMEs, the Dig.IT grant covers up to half of eligible costs, and building a dedicated solution falls within its mandatory component. On a well-structured project, half the bill moves to the grant.
What's next
If you want to estimate the cost of a first implementation in your own company, the order is simple: start with the diagnosis, which has a transparent price and gives you a map, then price the implementation itself based on that. I describe the full path — from diagnosis, through the first project, to what comes after — on the offer page.
→ See the full path and packages
→ Book a 30-minute consultation
