Amazon Search Results Page Density: Picking and Testing the Right Page Size

You are the product manager who owns search on Amazon. You are asked to choose how many products a single page of search results should show. How would you reach that decision and validate it? ```hint Find what the number trades off Each extra result changes something for the shopper, for ranking quality further down the list, for page speed and for sponsored placements. Also ask whether one number has to serve every device and every kind of query. ``` ### Clarifying Questions - Is this for the desktop website, mobile web, the shopping app, or all of them, given how differently each screen is browsed? - Does the count have to be the same for every query, or can it vary by query type, such as a broad search for running shoes versus an exact model number? - What prompted the question: a page-speed problem, a conversion goal, a layout redesign, or room for more sponsored results? - Are there fixed constraints, such as a page-load budget or a set number of sponsored slots per page? ### What a Strong Answer Covers - The trade-offs the count controls: choice and fewer page clicks versus scanning effort, relevance decay deeper in the ranking, page load time and ad inventory - Segmentation by device, layout and query intent, with a reason one number may not fit every case - A primary metric tied to shoppers finding and buying products, with guardrails such as page latency, sponsored revenue and query reformulation - An experiment design: arms around today's count, randomization unit, duration, keeping ad load constant, separating the effect of speed, and reading results by segment - A decision rule and rollout plan, including what to ship when metrics or segments disagree ### Follow-up Questions - Your test shows that larger pages raise purchases on desktop but slow loading on low-end phones. What do you ship? - How would you keep sponsored results from driving the decision at shoppers' expense when page size changes? - Would you replace numbered pages with continuous scrolling, and how would you test that change?