Most companies kicking off an AI initiative share one trait: they start with the solution, not the problem. Someone reads a report, gets excited, calls a vendor. Six months and €200k later, the thing sort of works — but nobody uses it.
Here’s what goes wrong.
”We want to do AI” is not a brief
The most dangerous phrase in an AI engagement is “we want to leverage AI to improve our business.” It sounds like a goal. It isn’t.
A goal is: we lose three hours per invoice on manual classification and we want that below twenty minutes. That’s something you can build toward. “We want to leverage AI” is a starting point for a product demo, not a project.
When the brief is vague, the project becomes a proof-of-concept that proves nothing in particular. Impressive in a demo. Meaningless in production.
Vendor selection happens too early
Companies often pick a vendor before they understand the problem. A vendor with a shiny chat interface becomes “the AI solution,” whether or not a chatbot is what the business actually needs. By the time the mismatch becomes obvious, the contract is signed and the internal champion has staked their reputation on it.
The vendor selection conversation should come after you’ve written down: what problem, what data, what output, what does success look like in six weeks. If you can’t answer those, you’re not ready to evaluate vendors.
The boring parts eat the project
The flashy part of an AI project — the model, the inference, the UI — is maybe 20% of the work. The rest is data access, cleaning, integration with existing systems, deployment, and monitoring.
Companies that don’t budget for this find their project stalled four months in, waiting on an IT ticket to get read access to a database. The model was ready in week two.
What the good ones look like
Start with a specific, costly problem. One process, one team, one bottleneck. Build the smallest thing that actually routes or classifies or drafts correctly in production. Then expand.
The best engagements I’ve seen spend the first two weeks arguing about whether to build anything at all. That’s exactly right. The audit phase isn’t wasted time — it’s the part that prevents you from building the wrong thing expensively.
If a consultant won’t tell you “this isn’t worth automating,” find a different consultant. If you want to know what a scoped, honest engagement looks like, here’s how we run ours.