코호트 분석으로 이탈 줄이기

Acquiring users is expensive; keeping them is where profit lives. Churn—users who stop returning—quietly drains growth even when installs look healthy.

This lesson uses cohort analysis to find and fix churn.

A cohort is a group of users who share a starting event—usually the week they installed. Tracking each cohort over time reveals how retention really behaves.

Aggregate numbers hide the truth; cohorts expose it.

Plot the percent of a cohort still active on day N after install.

A healthy curve flattens into a stable plateau—a loyal core remains. A cliff (sharp early drop with no plateau) means the product never reaches habit.

The goal is to lift and flatten the curve.

Standard checkpoints are D1, D7, and D30 retention. D1 reflects onboarding quality; D7 reflects early habit; D30 reflects long-term value.

A weak D1 poisons every later number.

A cohort table shows retention by signup week across days since install.

You can calculate day-N retention from event logs.

Cohort curves pinpoint where users leave. A big drop between D1 and D3 points to onboarding; a slow bleed after D30 points to fading long-term value.

Diagnose before prescribing a fix.

Compare retained vs churned users early actions to find the activation moment—the behavior correlated with sticking around (e.g., adding a friend, completing setup).

Drive new users toward that action fast.

Match the fix to the drop-off:

• Early cliff → improve onboarding and first value.
• Mid churn → re-engagement push, new content.
• Late churn → loyalty rewards, fresh features.

After shipping a fix, compare new cohorts against old ones. If later cohorts retain better at the same day-N, the intervention worked—cohorts give you a clean before/after.

You learned to fight churn with data:

• Group users into cohorts by install week.
• Read the retention curve—aim for a flattening plateau.
• Use D1/D7/D30 benchmarks to locate drop-off.
• Find the activation moment and target fixes, then verify with new cohorts.