AI product analytics need more than page views. Teams should connect campaign source, landing page, topic interest, feature intent, AI interaction quality, contact starts, and booked or submitted follow-through.
For RAG and agent systems, useful events include retrieval misses, source-clicks, user corrections, escalation reasons, approval outcomes, tool failures, latency bands, token cost, and model or prompt version.
The reporting loop should answer a practical question each week: which user intent showed up, what AI capability did they try or request, where did confidence break, and what should the product or architecture team improve next?