Facebook Feed Ad Frequency Decision: Weighing 1-in-25 vs 1-in-100 Post Spacing

You are a product manager at Meta working on the Facebook News Feed, where ads are placed among people's posts. You are asked to decide how often an ad should appear: once after every 100 posts, or once after every 25 posts. Explain how you would reach that decision and evaluate the trade-offs between the two options. ```hint Two clocks The gains from each option show up quickly, while some of the costs build slowly. Think about how you would see the slow costs before you commit. ``` ### Clarifying Questions - Must one interval apply to every user and device, or can the frequency differ by user, surface (mobile versus desktop) or ad format? - Which goal leads: near-term revenue, long-term engagement and retention, or results for advertisers, and over what time horizon should the decision be judged? - Is there enough advertiser demand to fill a slot every 25 posts with relevant ads, or would the extra slots go to weaker matches? - Does the post count reset at the start of each visit, or carry over from one visit to the next? ### What a Strong Answer Covers - Framing the decision for all three parties: people using the feed, advertisers, and Meta's revenue, plus the feed's role in engagement and retention. - The concrete trade-offs of each interval: impressions and revenue against ad fatigue, falling click-through and prices as inventory grows, relevance of the marginal ad, advertiser return, and long-term retention. - An experiment able to detect slow effects: arms, randomization unit, duration, primary and guardrail metrics, and a long-term holdout. - A decision rule that puts revenue gains and engagement losses on a common scale, ending in a clear recommendation. - Why a fixed interval is likely a stepping stone toward ad placement that adapts to the user, the session and the ad. ### Follow-up Questions - Eight weeks in, the 25-post arm earns far more per user, but 8-week retention is slightly lower and the gap is not yet statistically significant. What do you decide, and what do you do next? - How does your answer change for people who usually scroll past fewer than 25 posts in a visit? - If you replaced the fixed interval with a model that decides ad by ad, which limits would you keep around it?