How much does it cost to implement AI in a business? What drives the price, and where the money is wasted

“How much does AI cost?” is one of the first questions a manager asks, and rightly so. The trouble is that there is no single answer, just as there is no single answer to how much a car costs.

But you can know in advance what the cost is made of, what drives it up, and how to find out before you sign a big project.

Three kinds of cost

When someone says “the cost of AI”, they usually mean three quite different things:

Cost	What it is	When you pay
Development	Connecting AI to your systems, rules, checks and testing on your data	Once, at the start
Usage	The AI service, paid per document or request	Monthly, by volume
Maintenance	Tracking accuracy, and adjustments when documents or models change	Monthly or as needed

Most of the money almost always goes into development. Not because AI is complicated, but because the real work is in the connections: AI has to read your document, enter the data into your ERP and know when to ask a person.

Running it is cheaper than you think

This surprises most managers. AI services are billed by volume, and for a typical business document that means cents, often less.

Item	How it is billed	Order of magnitude
Reading the document (PDF, scan)	per page	about a cent per page
The AI model that extracts and sorts the data	by amount of text	a fraction of a cent per document
Hosting and storage	monthly	a few euros to a few tens of euros

For a company processing 1,500 documents a month, that usually comes to a few tens of euros a month. In the savings calculation we deliberately used a generous 5 cents per document, which came to €75 a month. Even that is small next to €3,000 of manual work.

Service prices change, and mostly downwards. So always check them for your own case, but the order of magnitude stays the same.

What drives development cost up

• The number of systems to connect. AI that only reads an email is one thing. AI that reads the email, checks the customer in the CRM and enters the order into the ERP is three times the work.
• How varied the documents are. One supplier with the same form every time is easy. Two hundred suppliers, each with their own PDF, plus scans and handwritten notes, need more rules and more testing.
• How many mistakes you can afford. If AI only sorts enquiries, the odd mistake is no problem. If it enters amounts into the accounts, it needs more checks and more human approval.
• The state of your data. If guides and price lists are scattered across emails and old spreadsheets, they have to be gathered in one place first. That isn’t AI work, but it still has to be paid for.

Where companies waste money

• Licences nobody uses. AI licences are bought for every employee, and three months later a handful of people use them, for emails.
• The “AI across the company” project. A big project without one clear process has no measure of success, so it runs and costs without end.
• A custom AI model. Businesses almost never need one. Ready-made models already read documents and write text; they only need connecting to your data.
• A demo instead of a pilot. A demo on three tidy examples always works. The real cost only shows on your actual documents, and it is better to find that out in a small pilot than after a big contract.

How to know the cost before you spend

1. AssessmentSmall cost: where it pays and how much
2. PilotFixed price: does it work on your data
3. RolloutThe main cost, only once the numbers show it
4. RunningCents per document, plus maintenance

A sensible AI rollout happens in steps, and you can stop after any of them:

1. Assessment. Short and inexpensive. The processes are reviewed, manual work is counted and a savings calculation is done for each. The result is a list of processes ranked by payoff.
2. Pilot. One process, on your real data, at a fixed price. The result is numbers: how accurate AI is, how much it saves and what it costs per document.
3. Rollout. Only now is money put into full integration and daily use, with the numbers in hand rather than a promise.
4. Running. The monthly usage cost and occasional adjustments.

That way most of the money is spent only once you know it pays off.

Does it pay off?

The calculation is simple: divide the development cost by the monthly saving and you get the number of months it takes to pay back. If a process costs €3,000 of manual work a month today and €575 with AI, the saving is about €2,400 a month. Development that costs as much as four months of that saving pays back in four months, and everything after that is pure profit.

How to get those numbers for your own process is in the article on how much AI can save.

How we work

This is exactly how we work: we start with an assessment with a savings calculation in euros, then run a pilot on one process at a fixed price, and only roll it out into daily work once the numbers show it pays off. No big project in the dark and no licences nobody uses. The first step is a free 30-minute call in which we go through one of your processes and tell you what the cost would be made of.