Most failed AI projects don't fail because of the technology. They fail before the first line of code — on a wrong assumption.
1. The problem is too broad
"We want to adopt AI" is not a project. "We want quotes ready in ten minutes instead of two hours" is. The narrower the first task, the more likely it actually gets used. A broad project usually ends in meetings, not results.
2. There's no data, or it's a mess
AI is exactly as good as the data it sees. If your prices live in five people's heads and three spreadsheets, no model will help. Often the first step is simply cleaning up the data — and that pays off even without AI.
3. There's no owner
If a new tool belongs to no one, it won't stay in use. You need a person responsible for making sure the solution is actually present in daily work: giving feedback, training colleagues, and noticing when something breaks.
4. The result isn't measured
"Feels faster" is not an argument for the next budget. Before you start, fix a number: hours per week, number of enquiries, error rate. After rollout, compare. Without that, no one knows whether the project succeeded.
How to avoid it
Pick one narrow, measurable task. Make sure the data exists. Assign an owner. Fix a baseline number. Build a small solution, put it into real work for a week, look at the result — and only then expand.
It sounds boring, but it's the boring approach that gets AI actually used. If you'd like a second pair of eyes on your first task, book a short call.