Metrics for AI Products

DAU and retention still matter — but they can't see the thing that makes AI features succeed or die: whether the output was good.

The AI-native metric stack: task success rate (did the user's job get done — measured by completion, not satisfaction), acceptance rate (draft kept / regenerated / edited / rejected), intervention rate (how often humans fix or redo the output), and time saved minus time spent verifying — the net, not the gross.