AI-implementatie

Why AI projects truly fail (it's not the tech)

Marc Cornelius6 October 2026 · 4 min read

The tool works. The integrations are live. Everything does exactly what was agreed. And yet six months later the project sits idle and the client has cancelled the contract. That's not a tech problem. That's an adoption problem, and it happens more often than you'd think.

The boss builds it, nobody uses it

The pattern is almost always the same. A CEO or manager sees what AI can do, gets excited and has a tool built. Only it never gets discussed internally. The tool ships and then precisely nothing happens. Nobody picks it up on their own, because the way the tool works doesn't match how people do their day-to-day work. The CEO gets frustrated that it's gathering dust. The staff get frustrated that something was dropped on them. And so a perfectly good tool sits idle within a few weeks.

A dashboard nobody opened

We built a reporting dashboard for a marketing agency that connected to all the tools they used internally. One of them is Optimizer, which watches Google Ads tasks and budget. We pushed all those actions into our own tool and into Google Sheets, so staff could work from there. On top of that we built our own actions via the Google Ads API that Optimizer didn't cover. A sizeable project. And then? No adoption plan. The odd message in the Slack channel I was in too, but nobody replied. Nobody picked the tool up on their own. Six months later they cancelled the contract and built a sort of Business OS themselves with some advice from us. Mapping out their way of working should have been step one.

You often see it in the first meeting

Usually I can feel it coming in the intro meeting. When someone says a tool needs to have a huge impact on internal work, but there's no budget, no focus, they're doing twenty things at once and nobody from the team is involved. That's when I already expect adoption to fail, and the whole project with it. And I say so straight away. We hate seeing a project sit idle as much as they do, because after six months the money is gone and nobody has anything to show for all that work. So watch for these signals: a tool meant to change the work radically without any time freed up for it, a boss who decides everything alone, and a team that only hears at delivery that something got built. Tick those three and you're building something that will grind to a halt.

“How daft is it to have something built and then do nothing with it?”

Humans-in-the-loop starts at the meeting

There are always two of us at the table, and we give honest advice based on experience. If we sense adoption becoming a risk, we say so on the spot. We'd rather see a deal shrink or fall through than a client stuck with a dead tool in six months. That's what humans-in-the-loop means in practice: a person who steps in at the right moment, even when it's awkward.

We run into ourselves too

And no, we're not immune. Take Bel je mail, a fun internal project of Maarten's that lets you call your inbox from your browser: voice chat with your mail. It works, it's fun. But the focus internally was never on that project. It still needs plenty of coding and a lot of care around security, and that simply isn't our core. So it never quite gets off the ground. The tech can do it fine; nobody just picks it up daily. If you want to make a project like that big, you have to be on it around the clock. Same law as with a client.

What to do when the pilot has already failed

Just had a pilot fail? Don't start tinkering halfway through the same loose process again. Go back to the start: what processes do you have, where's the most pain, and who on the team is going to use it in the end. Appoint an AI Lead who secures internal buy-in, and only then build. Adoption isn't a last little step after delivery. It's the first step before anything gets built.

Fancy a chat?

Not sure whether your AI plan will land or just sit idle? Run it past us in an intro call. We'll tell you honestly whether it'll work, even if that means you're better off waiting a bit. That saves you an expensive tool nobody opens.