A product team asks you to define a north star metric for their feature. How do you approach it?

Start from the value the feature delivers to the user, then find the measurable behavior closest to that value. A good metric moves when the product genuinely improves, is hard to game, and is available quickly enough to guide decisions. Pair it with counter metrics that catch harm, and prefer a rate or per user measure over a raw total, since totals rise with growth regardless of quality.

I start by asking what the feature is supposed to do for the user, because the metric should be the closest observable proxy for that. Then I stress test it by asking how I would game it if I were being cynical. For a search feature, clicks per search sounds reasonable until you realize bad results also generate clicks, so a worse experience raises the number. Something like the share of searches that end in a completed action within five minutes is harder to fake. I always pair it with counter metrics, so if we optimize notifications for engagement, unsubscribe and uninstall rates sit right next to it on the same chart. And I push for rates over totals, because total sessions goes up as the company grows and tells you nothing about whether the feature is good. The last part is agreeing the definition in writing with the product manager before anyone builds a dashboard, because renaming a metric after people have targets on it is painful.