Application analytics for CI/CD starts with a walk, not a spreadsheet. Sit with the person who last unblocked a failed release and ask them to narrate a typical change from pull request to production.
Useful prompts: Which stage emits timestamps you would defend in an audit? Where do retries reset the clock? Who notices when a canary stalls? Which alerts fire for humans versus which only fill a log bucket?
Write down every system that owns a piece of the story — source control, CI, artifact store, deploy orchestrator, incident channel. Gaps between those systems are where delivery metrics go missing. In our Busan engagements, the most common gap is deploy success recorded in one tool while rollback is logged only in a chat thread.
Leave the first hour with a stage list, an owner list, and a honesty rating for each timestamp. That sketch is worth more than an imported industry benchmark.