What 35+ AI implementations taught us.

Lesson 1The problem is rarely where the client says it is.

First conversations start with a solution. "We want a chatbot." Or: "We want our documents searchable with AI." Rarely with the problem itself.

First step in every project: walk the solution back to the question. What problem does this solve, for whom, how much time does it save, and is this the biggest problem you have?

At VAPRO, an examination institute, they asked for automation of their knowledge base. After a few questions: the real pain was candidate registration in the Stipel portal. One hour per ten candidates, manual, error-prone. That was the problem we tackled. The knowledge base came later, or never.

The question worth your time is not "can we build it" but "is this the problem we should solve".

Lesson 2Adoption doesn't start after launch.

The most common mistake: planning adoption as a separate step after implementation. "We build the tool, then we run training, then people use it."

It doesn't work. Adoption starts at problem analysis. The people who'll use the tool need to be involved in defining the problem. Not as a checkbox, but because their input makes the tool better and their involvement makes adoption real.

At FedEx, a manager asked whether his team's training could be paid by the company directly, not expensed. That sounds administrative. It was an adoption signal. He was putting his team behind it. That kind of buy-in is gold, and it starts well before launch day.

Lesson 3The champion is both bottleneck and key.

Every successful project had a champion: someone internal who really wanted to understand. Not the IT manager, not the director. An employee who saw what AI meant for their own work and pulled others along.

Projects without a clear champion stall, even when the tool is good. It doesn't get used, feedback dries up, and three months later nobody remembers what you built.

Identify the champion in week one. Give them access, time, input. The rest of the organisation follows what that person signals.

Lesson 4The platform sets the ceiling.

A kickstart at a large energy provider on Copilot Studio. Halfway through: the platform blocks external tool integrations, only offers GPT without reasoning, and behaves inconsistently between Teams and Copilot interfaces.

Those are fundamental limits no better prompt fixes.

Lesson: platform assessment belongs before kickoff, not after. That avoids discovering four weeks in that what you want to build isn't technically feasible on the platform the client picked. Sometimes the right answer is: pick a different platform. Sometimes: build this use case elsewhere.

Lesson 5Expectations are asymmetric.

Clients hear "PoC" and think "working product". They hear "four weeks" and think "all done".

That's not bad faith. It's the result of marketing that frames AI as easy, and consultants (myself included sometimes) who don't set the lower bound of expectations explicitly enough.

Approach since: in week one, write a definition of done. Not vague ("a working chatbot") but concrete: which questions does it answer, at what error rate, in which system, judged by whom?

That conversation is uncomfortable. It's also the most useful conversation you can have early in a project.

Lesson 6Small scope, big trust.

The projects that turned out best weren't the most ambitious. They were the most tightly scoped.

One problem, one team, four weeks. Then measure. Then decide whether to continue.

Clients who want to start with "a complete AI platform for the whole organisation", I slow down. Not because it can't be done, but because it's rarely a good starting point. One working tool that twenty people use daily says more than a hundred-page roadmap.

The pattern.

35+ implementations aren't proof that I get everything right. They're proof I've gotten enough wrong to know what works.

The pattern in successful projects: a concrete problem, an involved champion, honest expectations, a platform that fits the ambition, and adoption that starts on day one, not day thirty.

That sounds simple. Most companies still skip at least one of these steps.

Which one are you skipping?