Design and analyze notification pinning experiment
Company: Meta
Role: Data Scientist
Category: Analytics & Experimentation
Difficulty: hard
Interview Round: Technical Screen
Design an experiment to evaluate a feature that pins accounts with active notifications to the top of the account switcher for users who have ≥2 accounts. Specify: (1) targeting (only multi-account users) and the unit of randomization (person_id vs account_id) with rationale for avoiding cross-account interference; (2) exposure definition (what counts as being treated and an exposure log you would require); (3) primary success metrics (e.g., per-person notification view rate, click-through to notified account, notification-to-action conversion) and guardrails (e.g., time spent on accounts without notifications, overall session length, error rate, complaint reports); (4) sample size and duration assumptions using baseline metrics you would estimate from the provided tables, including how you’d handle unequal activity levels across users; (5) ramp plan, novelty effects control, and pre-experiment balance checks; (6) how you’d detect and correct interference or cannibalization across a person’s accounts (e.g., cluster/person-level randomization, ghost exposure, or exclusion windows); (7) analysis details: intent-to-treat vs treatment-on-the-treated, handling users who gain/lose accounts mid-test, missing data, multiple-testing control across segments (2, 3, 4+ accounts), and difference-in-differences or CUPED to reduce variance; (8) stop/go criteria and how you’d translate metric movements into a launch decision, including acceptable guardrail movement thresholds.
Quick Answer: This question evaluates experimental design, causal inference, metric definition and instrumentation, sample size estimation, and analysis skills in the context of a UI-driven notification pinning feature.