Metric Framework for Instagram's Search Bar, From Query Intent to Satisfied Session
Company: TikTok
Role: Product Manager
Category: Analytics & Experimentation
Difficulty: medium
Interview Round: Other
How would you tell whether search on Instagram is succeeding?
```hint Success depends on intent
Separate the reasons people open search, such as finding a specific account versus browsing a topic, before you pick a single number.
```
### Clarifying Questions
- Does search here include every result type (accounts, keywords, hashtags, places and audio), or mainly finding accounts?
- Is this an ongoing health assessment of search, or a readout for one change, such as a new ranking model?
- Which outcome matters most to the business: new connections between people, discovery of creators and content, or overall time in the app?
### What a Strong Answer Covers
- The role search plays in Instagram and the distinct user intents it serves
- A metric funnel from starting a search to a satisfied outcome, with precise definitions and time windows
- One north-star metric, why it was chosen, and how it could be gamed
- Guardrails for speed, integrity of results and the health of the rest of the app
- How to tell a click apart from real satisfaction, using post-click behavior, rater judgments or surveys
### Follow-up Questions
- Suppose many clicks land on results that do not match the query. Would your metrics count those as success, and how would you correct for that?
- Is it a problem if someone searches, gets results unrelated to their query and still engages with them? Does the answer depend on what they searched for?
- After a ranking change, search success rate goes up but searches per user go down. How do you interpret that?
Overview: A product analytics question about how to judge whether Instagram's search feature is succeeding. It tests separating user intents, building a metric funnel from search start to a satisfied outcome, choosing a north-star metric with guardrails, and handling clicks on results that do not match the query.