DAU dropped 15 % week-over-week with no planned changes. How do you diagnose it?
A metric drop investigation starts by confirming the drop is real — ruling out logging bugs and metric-definition changes — before hypothesising causes. Then segment by platform, geography, user cohort, and funnel step to isolate where the drop is concentrated, which points to the most likely root cause.
How to think about it
Decision-tree framework for metric drops
Step-by-step walkthrough
1. Confirm the data is correct. Check if the event-logging pipeline had an outage, if a new app version silently dropped an SDK call, or if the metric definition changed (e.g., a bot-filter was tightened). If DAU is computed from a different table than usual, the “drop” may be an artifact.
2. Check for external causes. Was there a major holiday, a viral news event, or a competitor launch? Compare year-over-year for the same week.
3. Segment the drop. Narrow down who is affected:
- Platform: iOS vs Android vs web
- Geography: one country or global?
- User cohort: new users, reactivated, or long-tenured?
- Feature surface: which part of the app did users stop using?
4. Walk the funnel. A drop in DAU is the output of changes earlier in the funnel. Check: new installs, first-session completion, D1 retention, D7 retention in sequence. The first funnel step that shows the drop reveals the break.
Worked example. DAU dropped 15 % globally. Logging was intact. Seasonal check clean. Segmenting: iOS only (-28 %), Android flat. Funnel: iOS D1 retention dropped from 42 % to 21 % in the same week. Root cause: an iOS release shipped a broken onboarding screen. Fix: hotfix the release; reactivation campaign for affected cohort.