Retention curves bend for reasons that are not product failure: holiday patterns, campaign-heavy acquisition weeks, or a quiet change in how active users are counted.
Freeze the definition of active before diagnosing. Then compare cohorts from the same channel and first feature used — not every signup in a noisy month.
When a real drop remains, inspect the first session. Instrumentation audits often reveal a missing activation fire that masquerades as churn.
Keep a short memo of known seasonal effects for your category in Korea. It saves hours of false alarms after long weekends.