Choosing a North-Star Metric and Guardrails for Meta's Personalized Feed

You own Meta's personalized news feed: the ranked stream of posts each person sees when they open the app. How would you tell whether it is succeeding, and which metrics would you use? ```hint Value before volume A ranking change can raise clicks and time spent while leaving people feeling the feed wasted their time. Decide what a good feed visit looks like for the viewer first, then ask what the same change could do to the people who post and to advertisers. ``` ### Clarifying Questions - Which feed is in scope: the Facebook app's Feed, Instagram's feed, or the shared ranking system behind both? - Is the goal an ongoing health scorecard for the feed, or a decision metric for whether individual ranking changes should ship? - Should ads shown inside the feed count toward the feed's success, or be treated as a separate product with its own goals? - Should the value of personalization be judged against a non-personalized baseline, such as a chronological feed? ### What a Strong Answer Covers - The feed's purpose for viewers and how it supports Meta's mission, stated before any metric - The groups the feed serves (viewers, the friends and creators who post, advertisers) and which group's outcome is primary - One precisely defined north-star metric, with the reason it beats time spent, impressions or active users - Supporting and guardrail metrics covering personalization quality, integrity, poster health, ad load and reliability - How the metrics would be measured causally, including experiments, long-running holdouts and segment cuts ### Follow-up Questions - After a ranking change, your north-star metric goes up but survey responses say the feed has become less worth people's time. What do you do? - How would you separate the quality of personalization from the effect of simply having more posts available to rank? - How would you use these metrics to decide how many ads the feed should carry?