The post-mortems usually blame the model, the data, or the vendor. In our experience, the project was already in trouble long before any of those were chosen. Most enterprise AI projects fail at the definition stage — in the gap between an exciting idea and a decision that something concrete will change.
When a project is struggling, we ask three questions. They sound simple. They are rarely answered well.
1. What decision or action changes?
“Use AI for customer support” is not a project. “Cut first-response time by surfacing the right answer to the agent before they finish reading the ticket” is. If you can't name the moment in someone's day that gets better, you don't have a product — you have a budget line.
2. Who owns the outcome?
Not the project. The outcome. If the answer is “the AI team,” it will ship a demo and stall. The owner has to be the person whose number moves — and they have to want it.
Strategy is now abundant. Execution is the scarce resource. The projects that ship are the ones where someone owns the result, not the roadmap.
3. What does “good” look like, measurably?
If you can't define success before you build, you won't recognise it after. Worse, you'll keep shipping pilots — each impressive, none in production — because there's no bar to clear.
None of this requires AI expertise. It requires the discipline to define the problem before falling in love with the solution. Get those three answers on one page, and the technical work becomes almost boring. That's the goal.