The hardest ML projects aren't the ones with difficult data or complex models.

They're the ones where nobody can clearly define what success looks like.

"Build us an AI that makes our customer service better."

Better how? Faster responses? More accurate answers? Higher customer satisfaction? Fewer escalations to humans? Lower cost per interaction?

Each of these leads to a completely different system. And optimizing for one often hurts another (faster responses that are less accurate don't help anyone).

What I do when faced with ambiguity:

Ask "how will you know this project succeeded?" If they can't answer concretely, we're not ready to build.

Propose specific, measurable outcomes. "Reduce average resolution time from 12 minutes to 8 minutes while maintaining current satisfaction scores." Now we have a target.

Start with the metric that's easiest to measure and most aligned with business value. Once the first version ships, refine.

Build a dashboard of metrics from day one. Stakeholders who can SEE the numbers become much more precise about what they want to optimize.

The ability to translate vague business goals into specific ML objectives is worth more than any technical skill.

Because the model that solves the wrong problem perfectly is worse than a simple model that solves the right problem adequately.