AI-implementatie

How do you measure the ROI of AI?

Marc Cornelius12 November 2026 · 4 min read

Ask a business owner what AI gets them and you often hear the same answer. "It feels faster." "Everyone is enthusiastic." Nice to hear, but it is not a number. And you cannot put a feeling in your annual accounts. If you seriously want to know whether AI pays off, you have to look at the right number. And that is usually not the number people point at.

Why "we feel more productive" tells you nothing

Feeling productive and being productive are two different things. Someone can feel fantastic about a new tool while barely saving any time with it. Or the reverse: grumble about the change and still gain hours structurally. Feeling is a poor gauge because it moves with the hype, not with reality. You want to know which task takes less time, how much less, and how often that task occurs.

The sum is simpler than you think

Take a concrete task. Measure how long it used to take and how long it takes now. Multiply the difference by how often the task occurs per week and by the hourly rate of the person doing it. That is your saving per week. Set the build cost and the monthly maintenance against it, and you will see when you break even. No magic, just arithmetic with real numbers instead of percentages nobody can trace back. Important: take one clearly bounded task, not a vague feeling about the whole department. A task you can measure. A department you cannot. And pick a task that comes back often, because that is where the volume sits that earns back your investment. A job you do twice a year is not a good candidate, however annoying it may be.

Adverge: from 15+ hours of manual work to almost none

At Adverge, our own marketing agency, the pain sat in recruitment. Managers reviewed candidate tests by hand, hour after hour. We built an applicant portal that simulates a Google Ads account. Candidates show their skills without being able to touch real client data or real budget. Time to hire dropped by 80 percent, managers save 15+ hours per week, and the system screens 1000+ candidates as easily as ten.

“Highly paid staff doing low-value review work is your most expensive line item. And often the most measurable.”

Do not forget the hidden costs

An honest calculation counts more than the build cost. Maintenance and hosting run with us at around 150 to 300 euros per month. Factor in adoption time too: the hours your team spends learning to work with it. Leave that out and your ROI looks prettier on paper than in reality. And a calculation that fools you is worse than no calculation.

Adoption is the real gauge

Here is the sting. You can build a perfect tool with a beautiful ROI on paper, but if nobody uses it your return is zero. I have seen it happen: a big dashboard delivered, no adoption plan, contract cancelled six months later because simply nobody started working with it. The lesson: do not just measure the theoretical saving, measure whether people still touch the tool after two weeks. Usage is the proof, not the promise.

Where the freed-up time goes

Saving hours is not the goal in itself. The goal is what you do with those hours. At Adverge managers now only speak with candidates who have already passed the simulation. The freed-up time goes to the conversations that truly matter, the human part. That is humans-in-the-loop in practice: the machine does the boring groundwork, the human does the work that calls for judgement and feel.

When can you expect to break even?

That depends on how often the task occurs and how expensive the person doing it is. A daily task done by highly paid staff pays for itself very quickly. A task that comes round once a month, much later. Our rule of thumb: below 4000 euros of implementation it usually only starts to get interesting, and above that amount you want to be able to do the sum in advance.

Want to run your own numbers?

The most honest ROI calculation starts with your own figures: which task, how often, what rate. If you do not have those to hand, that in itself is information. In an intro chat we work through it together, even if the outcome is that AI does not pay off here yet. We would rather tell you that now than afterwards. Shall we run the numbers?