Churn post-mortem from cancellation data

Here are recent cancellations: {{paste reasons given, plan, tenure, usage before churn, support history}}.

1. Cluster by the real reason, not the stated one. Stated reason "too expensive" often means "didn't get value" — use the usage data to tell them apart, and show your evidence.
2. Segment by tenure: churn in the first 30 days is an onboarding or expectation problem; late churn is a value or competitive problem. Split accordingly.
3. Identify the pre-churn signals visible in the data 30 days before cancellation, and whether they're strong enough to act on.
4. For each cluster, say whether it's preventable, and by what — product, onboarding, pricing, or better qualification at the top of the funnel.
5. Name the churn we should welcome: customers we shouldn't have sold to. Quantify it.
6. Give the two interventions with the best expected return, how to run each as an experiment, and the metric that proves it worked.
7. State what this data cannot tell us and the five customers I should call this week.

Don't propose a save-offer discount as the first answer.