Why I turn down AI projects with unclear success metrics

A prospective client with genuine enthusiasm and budget, but no clear answer to "how will we know if this worked," is one of the more common reasons I decline a project, and it's counterintuitive advice for a consultant to give, since enthusiasm and budget are usually exactly what a consultant wants to see.

The problem with an undefined success metric isn't that the project might fail, it's that it can't cleanly fail or succeed at all, which means six months in, the conversation about whether to continue, expand, or stop happens based on vague impressions rather than an agreed standard, and vague impressions tend to favor whoever's more persuasive in the room that day, not whoever's actually right.

Before starting any engagement, I ask for a specific, measurable definition of success, stated by the client in their own words, not proposed by me: a number, a rate, a time saved, something that can be checked against reality at a defined point. If the client genuinely can't articulate one, that's the actual first project, not the automation or tool they came in asking for.

Turning down an eager, funded client feels like leaving money on the table. It's a better outcome than six months of ambiguous work that ends in a disagreement about whether it succeeded, which is a worse outcome for the client's trust in AI projects generally, not just in this one.