Retention curves bend for many reasons that are not product failure: onboarding cohorts from a campaign, holiday usage in Korea, or a change in how you count active users.
Start by locking the definition of "active". Then compare like cohorts — same acquisition channel, same first feature used — before declaring a crisis.
When a real drop appears, ask what changed in the first session. Instrumentation audits often reveal that a new screen stopped firing the activation event, which looks like churn but is simply a silent tracking break.
Keep a short list of leading indicators (time-to-value, second-day return) beside lagging ones (day-30 retention). Leading indicators give you room to act.