Twitter (X) Home Timeline: Metrics and A/B Test Plan for a 2x Video Increase

You are a product manager working on the home timeline at Twitter (now X). The team is considering showing users twice as much video content in their home timeline as they see today. How would you decide whether this change is a good idea? Explain what you would measure, how you would test it, and which results would lead you to ship it, change it or drop it. ```hint Watch what gets displaced Video tends to raise time spent almost by construction, because a clip takes longer to watch than a post takes to read. Ask what the extra video pushes out of the timeline, and which users and authors would feel that loss. ``` ### Clarifying Questions - Does "twice as much video" mean double the share of timeline items that are videos, or double the number of video impressions per user? - Will the extra videos come from accounts the user follows, from recommended accounts, or both? - What is motivating the idea: more video ad inventory, competing with short-video apps, or growing video creators on the platform? - Should the change apply everywhere, or are some platforms and regions, such as places with slow or costly mobile data, out of scope? ### What a Strong Answer Covers - A clear goal for the home timeline and explicit hypotheses for how more video could help or hurt it. - A primary decision metric that video cannot inflate mechanically, plus secondary, revenue and guardrail metrics covering conversation, negative feedback and app performance. - An experiment design: randomization unit, more than one dose level, a duration long enough to rule out novelty, and handling of network effects between users. - Segment analysis and decision rules agreed before results arrive, including options between a blanket yes and a blanket no. - Long-term and ecosystem effects on text-first authors, video creators and advertisers. ### Follow-up Questions - Time spent rises sharply in the test, but replies and new posts fall. What do you recommend? - How would your analysis change if most of the extra video came from accounts users do not follow? - After launch, how would you catch a slow decline that a few weeks of testing could miss?