Diagnosing an Overnight Halving of Instagram Feed Impressions

You are the product manager responsible for Instagram's feed. The latest daily numbers show feed impressions at half of what they were the day before. Walk through what you would do: how you would work out what happened, which actions you would take and in what order, and who you would keep informed. ```hint Ask what can change overnight Real user habits almost never halve in a single day. Think about the kinds of change that can happen that fast, including changes to how the number itself is produced, and which of them you can confirm or rule out quickest. ``` ### Clarifying Questions - How is a feed impression counted: main feed only, or also ads and suggested posts, and is it logged by the app or by the servers? - Is the previous day's data complete, and does the drop show up in more than one dashboard or data source? - Is the drop global, or already known to be concentrated on one platform, app version or region? - Have related metrics, such as daily active users, time spent or ad impressions, moved as well? ### What a Strong Answer Covers - Treating an overnight halving as an incident, with urgency, owners and early communication, including to the ads business - Confirming whether the drop is real by checking data completeness, logging and definition changes against independent sources - Breaking impressions into users, sessions and impressions per session, and segmenting by platform, app version, region and content type to locate the cause - Matching the timing to recent releases, ranking or experiment changes, outages and external events, then mitigating with rollbacks before the root cause is fully proven - Follow-through: correcting the data, a postmortem, and alerts or release checks that catch the next drop sooner ### Follow-up Questions - Daily active users and time spent are flat; only feed impressions halved. What does that point to, and what do you check next? - The drop is confined to the newest Android release. What do you do in the next hour, and what do you tell the ads team? - It turns out to be a logging change and real usage was unaffected. What would you change so that the next one costs hours rather than a day?