Churn Analysis

• Compute churn rate correctly from a cohort
• Separate voluntary from involuntary churn
• Build a churn-cause matrix from data and interviews

• A churn rate without cause is a symptom; the cause is where the fixable action lives.
• Involuntary churn is often silently large — dunning failures equal 'free' revenue being flushed.
• Time-to-churn distribution shows whether users exit fast (activation issue) or slow (value decay).
• Churn by segment (plan, cohort, channel) assigns ownership: the fix is product, pricing or operations.

'It's raining' (high churn) tells you to own an umbrella; 'the roof leaks over the living room' (cause) tells you to fix the shingles. The rate signals, the cause tells you what to fix.

• Define loss for each user: inactivity cutoff (e.g. 30 days no qualifying event) or canceled/no-payment for subscriptions.
• Compute period churn = lost-in-period / active-at-start (survival-style, not simple average).
• Split voluntary vs involuntary via billing outcomes (payment declined, cancellation flow used).
• Build time-to-churn distribution and cohort churn rates.
• Triangulate reasons: exit events, cancellation reasons, support tickets, win-back interviews.

• A subscription brand discovers 18% of monthly churn is involuntary (failed cards) and builds a dunning flow — recovering most of it without any product change.
• A B2B platform reads its churn-reason matrix quarterly and re-scopes its roadmap toward the #1 reason (missing integrations).
• A wellness app segments churn by plan and finds free-plan churn is 3x paid — pooling fixes into the free-to-paid value bridge.

• Use period-start denominators for rates; dividing by gross signups misstates churn.
• Track involuntary churn as its own metric with dunning/retry dashboards.
• Maintain a churn-reason taxonomy and tag exits consistently across support and win-back channels.
• Monitor churn in cohort rows so a new-vintage churn spike is visible before it becomes a trend.

• Counting every user with no activity this month as churned when the cutoff is far too short for the product.
• Ignoring involuntary churn and 'fixing' product instead of billing.
• Computing churn as lost / total users (mixing in brands-inactive lurkers) and getting a fraction of reality.
• Reading churn rate without the reason mix — two identical rates can hide opposite causes.

• A churn jump with zero cancellation-flow use = involuntary (payment/fraud) or tracking gap — check billing before product.
• Time-to-churn histogram bimodal in shape (fast + slow) signals two different churn populations with two fixes.
• Reconcile billing churn with activity churn; when they diverge, definitional drift is the usual cause.
• For fast-exit spikes, replay the last-5-sessions of a sample of churned users before theorizing.