Diagnose a 20% Session-Length Decline After a Facebook Feed Ranking Launch

Meta has rolled out a new ranking algorithm for the Facebook News Feed. Since the launch, average session duration has fallen by 20 percent, and the drop is being attributed to the new algorithm. You are the product manager. What would you do? ```hint Shorter is not automatically worse Average session duration is only one view of engagement. Think about which other numbers would tell you whether people are using Facebook less, or simply using it differently. ``` ### Clarifying Questions - Was the 20 percent drop visible in the A/B test before launch, or did it appear only after the full rollout? - What was the new algorithm meant to improve: relevance, meaningful interactions, content quality, or something else? - Did the drop happen all at once or gradually, and is it global or limited to certain platforms or countries? - Over the same period, how have total time spent, sessions per user and daily active users moved? ### What a Strong Answer Covers - Validating the drop first: instrumentation, metric definitions, logging or dashboard changes, seasonality, and whether the timing lines up with the rollout. - Decomposing the metric (total time = active users × sessions per user × time per session) to separate a real engagement loss from a shift in how people use the feed. - Segmenting by platform, app version, country, new versus tenured users and content type, with internal and external causes considered separately. - Judging the change against the algorithm's goal using interaction, content creation, retention and revenue metrics, not session length alone. - A clear decision (keep, fix or roll back) backed by a holdback or reverse experiment that confirms causality. ### Follow-up Questions - Total time spent per user is flat, but ad impressions per user have fallen. What would you recommend to leadership? - How would you design a long-term holdback to measure the algorithm's effect on retention months from now? - If the drop were concentrated among new users only, how would your diagnosis and next steps change?