It started with a tool, not a problem
The most common cause of failure is in the brief. A company decides it "needs AI" and only then goes looking for something to use it on. It's like buying a drill and then wandering the house for a hole to make. Without a clear problem, the project has no way to succeed — you can't even say what success would be.
Reversing the order solves most problems before launch. First name a concrete pain: "it takes the sales team half a day to process new leads." Only then ask whether and how technology can help with exactly that.
A good brief can be measured. If at the start you can't say the sentence "it'll be a success when this number changes by this much," you don't have a project yet — you have a wish. And wishes can't be finished.
Chasing hype instead of value
The second trap is shine. In the media and at conferences you hear everything AI can do, and a pressure builds to "have it too." But the technology that's in fashion isn't the same as the technology that earns or saves you money.
In practice we see companies launch a flashy project because it sounded good in a meeting — while the biggest pain sat elsewhere, in a boring task nobody wanted to touch. Hype pulls attention toward what's visible, not toward what's valuable.
The cure is uncomfortably simple: for every idea, ask how many hours or euros it actually saves or brings in. If you can't answer that, it isn't a project for now. Value, not novelty, is the right yardstick.
The data is a mess, or missing
The third trap is the quietest and the most unpleasant. Even a perfect idea rests on data — and in many companies the data is scattered across spreadsheets, emails, and people's heads, in different shapes and full of holes. A machine that learns from a mess returns a mess.
The key is not to discover this halfway through. Before anything bigger gets going, it's worth checking whether the data even exists, whether it's in one place, and whether it's clean enough to draw anything from.
The good news is that tidying the data often delivers value on its own — even if no "AI" is deployed in the end. Order in your records is an investment that doesn't get lost. Start with it, not in spite of it but precisely because of it.
No one owns it
The fourth trap isn't technical at all. The project launches, the excitement fades, and when something breaks or needs adjusting a few weeks later, nobody is in charge of it. A thing with no owner quietly rusts.
Successful projects always have a specific person inside the company who understands them and looks after them. It doesn't have to be a programmer — it has to be someone who knows the process, notices when the system starts doing something silly, and has the mandate to fix it. Without them, even a well-built thing drifts out of step with reality over time.
So for every project, decide right at the start who its owner is. Not after launch, but before it. Technology nobody watches isn't a saving — it's a deferred problem.
A big bang instead of small steps
The fifth trap is impatience. A company decides to launch everything at once — a whole new system, every department, one big day. It sounds decisive, but it's the riskiest possible path. When something breaks, everything breaks at the same time, and nobody knows what the culprit was.
It's far safer to split the project into small, self-contained pieces and roll them out one by one. Each piece delivers a visible result, can be checked, and in case of a mistake can be stopped without dragging down the rest. Small wins build the trust that a big bang only spends.
All five traps share a common denominator: they grow from a hunger for the big leap. The way out is the same for all of them — a clear problem, measurable value, clean data, a responsible owner, and small steps. It isn't exciting. But that's exactly why it works.