Gmail Metric Drop: Diagnosing a Sudden 10% Fall in Monday-Morning Traffic

As the product manager for Gmail, you check the dashboards on a Monday morning and find that traffic has fallen by 10%. Walk through how you would investigate the drop and what you would do about it. ```hint Check the baseline first Monday mornings look different from weekends and from the rest of the week, so decide what the 10% is measured against, and whether more than one source shows it, before hunting for a cause. ``` ### Clarifying Questions - Which metric is "traffic" here: active users, sessions, inbox loads, or the volume of mail sent and received? - Is the 10% measured against the same hours on previous Mondays, against the weekend, or against a forecast? - Is the drop spread evenly, or concentrated on one platform (web, Android, iOS, third-party mail clients), one region, or one account type such as consumer versus Workspace? - Were any client releases, server changes or experiments rolled out over the weekend? ### What a Strong Answer Covers - Confirming the drop is real by ruling out pipeline delays, logging or definition changes, and a bad comparison window - Breaking the metric into components and segments (platform, version, geography, account type, entry point, timing) to localize the cause - Separating internal causes such as releases, experiments, sign-in or spam-filter changes from external ones such as holidays, OS updates and network outages, with a way to confirm or rule out each - A response matched to the cause: incident escalation, rollback or hotfix, communication with users and business administrators, and a post-incident review - Recovery tracking with guardrail metrics, and a fix for why alerting did not catch the drop first ### Follow-up Questions - The drop is limited to the iOS app and began right after Apple released an OS update overnight. What do you do in the first hour? - Every user-activity dashboard shows the drop, but server-side mail volume is flat. What does that tell you, and where do you look next? - How would you set up monitoring so the team hears about a drop like this before anyone opens a Monday dashboard?