Building AI products: What metrics should we track?

Working on an AI writing assistant and trying to figure out the right metrics. **Currently tracking:** - User engagement (DAU/MAU) - Generations per user - Acceptance rate (did user use the AI output?) **Questions:** 1. How do you measure AI quality at scale? 2. What's a good acceptance rate benchmark? 3. How to measure "helpfulness" beyond clicks? Building AI products feels different from traditional SaaS. The feedback loops are unclear. Would love to hear from others building AI-first products!