Defining Goals, a North-Star Metric and Guardrails for Meta's Reels Recommendations

Meta's short-form videos, called Reels, are watched in a full-screen viewer where people swipe from one video to the next, and they also appear inside other parts of the app. A recommendation engine decides which Reel each person sees next. As the product manager who owns that recommendation engine, what goals would you set for it, and how would you measure whether it is meeting them? ```hint Serve more than one audience The ranking decides outcomes for viewers, for the creators whose videos are or are not shown, and for the business that funds the product. A goal that serves only one of them can quietly harm the others. ``` ### Clarifying Questions - Is this the Reels recommendation engine on Instagram, on Facebook, or a shared system serving both? - Does the engine rank only the full-screen Reels viewer, or also the Reels inserted into the main feed and other discovery surfaces? - Is the choice of which sponsored Reels to show, and how often, part of this engine, or does a separate ads ranking team own it? - Is there a current priority behind the question, such as growing the creator base or improving retention of new viewers? ### What a Strong Answer Covers - Why the engine exists in Meta's strategy and whom it serves (viewers, creators, advertisers), leading to a clear primary goal. - A precisely defined north-star metric that reflects satisfying viewing rather than raw time spent. - Supporting metrics for creators and monetization, and guardrails for integrity, user well-being and cannibalization of other surfaces. - How the goals would be measured in practice, including online experiments, satisfaction surveys and long-term holdouts. - The trade-offs between these metrics and a rule for resolving conflicts between them. ### Follow-up Questions - A ranking change increases watch time but reduces how often people share Reels with friends. Would you ship it? - How would you detect that the engine is optimizing for compulsive viewing rather than satisfaction? - How would your goals and metrics differ for a brand-new viewer with no watch history?