Measuring Instagram's Context-Aware Search: North Star, Funnel and Guardrails

You are the PM for contextual search on Instagram. What metrics would you use to tell whether it is working? ```hint Define the context first Decide what context the search draws on and where people meet it; what counts as a successful search depends on the job the feature does for the person searching. ``` ### Clarifying Questions - Does "contextual" mean results that adapt to the searcher's situation (location, what they were just watching, their interests), or search suggestions placed inside content such as a Reel, a caption or the comments? - Which result types are in scope: accounts, Reels and posts, places, audio, hashtags, or all of them? - Is the team's goal right now to get more people using search, or to make each search more successful? - Should outcomes for creators and businesses whose content appears in results count toward success? ### What a Strong Answer Covers - A clear definition of contextual search, the user problem it solves, and why it matters to Instagram - The groups involved (people searching, and creators whose content is surfaced) and the journey from query to engagement - One north-star metric built on an explicit definition of a successful search - Supporting metrics across adoption, relevance, engagement and retention, including creator-side outcomes - Guardrails for failed searches, latency, harmful content, privacy discomfort and cannibalization of other surfaces ### Follow-up Questions - After launch, searches per user rise but time spent in Feed falls. How do you decide whether that is a win? - How would you set up an A/B test for a ranking change, given that it also changes which creators get reach? - How would you measure relevance when people rarely tell you a result was wrong?