Identify latent group-call demand from behavior
Company: Meta
Role: Data Scientist
Category: Analytics & Experimentation
Difficulty: hard
Interview Round: Technical Screen
Beyond 'call loops', design a concrete, data-driven framework to infer demand for a Group Call feature using user-level internal data. Provide at least four measurable signals, each with a precise definition, unit of analysis, and SQL-friendly metric. Examples to consider (select and formalize): (1) Multi-party coordination friction: number of distinct recipients a user calls within 15 minutes after posting in a shared group chat; (2) Concurrent availability: probability that ≥3 strongly connected friends are simultaneously online within 10 minutes; (3) Failed reachability cascades: sequences of ≥2 failed 1:1 call attempts followed by a successful call to a different friend in the same ego-network; (4) Message-to-call burstiness: alternating message threads among ≥3 users culminating in back-to-back 1:1 calls; (5) Overlapping 1:1 calls among a triad within 10 minutes. For each signal, specify: data sources, exact filters (time windows, tie-strength thresholds), expected direction if demand is high, and a validation plan using internal early data from similar products (e.g., prior small-group call pilots). Conclude with a triangulation plan that combines the signals into a single weekly demand index, and list two falsification/guardrail checks that would reduce confidence in the inferred demand.
Quick Answer: This question evaluates a data scientist's ability to design measurable product-analytics signals, infer latent user demand from event-level messaging and calling data, and operationalize those signals into SQL-friendly metrics within the Analytics & Experimentation domain.