AI project ROI · 2-minute video
How do you know an AI project will pay off before you build it?
Ask about value first. Most teams choose the technology, plan the work and ask what it returns last, if at all. Flip the order, and four questions give you most of the business case before anyone builds.
First step of the €1,200 workshop. Invoiced after. Full refund if it wasn't useful.
or test it on your own numbers first ↓1 min 55 s · Captions on. Nothing loads until you press play.
Value comes last, so nobody can measure it
Someone in the business asks for a better attribution model, or a recommendation model. Everyone feels it will change how marketing spends or lift conversion. By how much? Nobody wrote it down. The request came from the business, so everyone assumed someone knew.
Nobody added up what it costs to build, either, or what it will cost to keep running for years. Six months later someone asks what it returned, and there is nothing to check against, because nobody set a number.
The problem isn't the team or the idea. It's the order: technology first, work plan second, value last. A demo shows what got built. It can't show whether anything changed.
Flip the order: value, then a small test, then the build
A good restaurant prices a dish before it goes on the menu, and something comes off the menu to make room. Do the same with data and AI work. Before anyone builds, answer four questions.
Question 1
What will it cost to keep running?
Not just to build. Cloud, licences and the team's time on fixes and questions, every year it stays live.
Question 2
What has to stop to make room?
The team is already full. If nothing stops, the new project runs on overtime and something else quietly slips.
Question 3
What will it return?
In extra revenue or savings, in euros. A rough number with an honest confidence level beats no number at all.
Question 4
Does it solve the real problem?
Or does it only feel like it will? Check that the problem was measured, not assumed.
Then a small test. Build the smallest version that could make a difference, run it for a few weeks with the people who will use it, and write down what happened. That is an Impact Check: value first, then the test, then the build.
Run the four questions on one project
Pick one project someone wants to build: an AI pilot, a model, a dashboard. See which questions you can answer, what it returns after running costs, and how long it takes to pay back the build. Rough numbers are fine.
Runs in your browser. Nothing you type is sent, saved or tracked.
Your result appears here.
Fill in what you know. Gaps are part of the answer.
Bring the project you're being asked to approve
A 30-minute call. We agree whether the session is worth it and who joins. A data lead can join if you want one.
The working session, on your numbers. This project's four questions, next to what you already pay to keep running, so you can name what stops to make room for it.
What you leave with. A clear name for what's costing you, one number you can defend to the board, and a first list of what to stop, with the reason next to each.
In one e-commerce company, working around data was costing €930,000 a year in people's time and tools. €611,000 of it could be stopped. It took three weeks to measure, and the first cancellations were made before the engagement ended.
€1,200, invoiced after the session. Full refund if it wasn't useful. The fee is credited toward the Data Capital Audit if you continue.
Thirty minutes, no slides. If the session isn't worth it, you'll hear that on the call.
What happens in the workshopAI business case: common questions
How do you calculate the ROI of an AI project?
Take what it returns each year, minus what it costs to keep running, and set that against the build cost. The check above does exactly that. A rough return, with a note on how sure you are, is enough to choose a small test.
What costs are missing from most AI business cases?
Usually two: what it costs to keep running for years, and what has to stop to make room. Your template probably covers the return. These two are what the session adds.
I need a yes or no on this project, not a workshop about stopping things.
The yes or no depends on question 2. If nothing stops, the project runs on overtime. The session answers both: this project, and what makes room for it.
Is this different from approving an AI tool purchase?
Yes. This page is for a project you build. If you're buying a vendor tool, start with Should we approve this AI purchase?
The right time is before the budget is signed.
First step of the €1,200 workshop. Invoiced after. Full refund if it wasn't useful.
Already live? Check if it's paying off