Meta Data Analyst Case Interview: Product Analytics, Experimentation, and SQL on Consumer Apps

This 60‑minute, interviewer-led case mirrors Meta’s real product analytics interviews for Data Analysts. It emphasizes product sense, metric design, causal inference via experiments, and pragmatic SQL—framed around Meta surfaces such as Instagram Reels, Facebook Feed, Groups, or WhatsApp. You will be expected to move fast, focus on impact, and communicate directly while showing sound judgment on integrity and privacy trade‑offs. What the case covers (and how it flows): 1) Problem framing (≈5 min) - Scenario prompt (e.g., “Reels retention dipped 3% WoW in India on Android” or “Groups notifications are underperforming for new users”). - Clarify objective, target users, platform, and constraint trade‑offs (growth vs. integrity vs. revenue). - State a hypothesis tree and the decision you’re driving toward (ship/iterate/rollback/investigate). 2) Metric design and success criteria (≈10–15 min) - Define a clear North Star (e.g., 7‑day retained creators or daily watch time per viewer) and supporting metrics (activation, engagement depth, creator supply, quality). - Guardrails common at Meta: integrity/quality (policy violations rate, spam reports), user experience (crashes, latency, notification disable rate), and business (ads or time‑spent cannibalization across surfaces). - Specify metric formulas, units, and windows (DAU/WAU/MAU; p50/p95 latency; session‑level vs. user‑level; cohort‑based retention). 3) Analysis plan and SQL (≈10–15 min) - Propose data sources and joins at a high level (event logs in Hive/Presto; user/device/region dims). - Walk through a brief SQL sketch to compute core metrics (e.g., 7‑day retention by cohort with region/device splits; watch time per user with winsorization; notification CTR and disable rate). - Call out instrumentation assumptions (event names, deduping, timezone, late events/backfills) and how you’d validate data quality. 4) Experiment/causal approach (≈15–20 min) - Formulate hypotheses and variants; define randomization unit (user vs. page/account), exposure, and holdouts. - Power/MDE thinking, test duration, and seasonality. Name common pitfalls: SRM checks, novelty effects, population imbalance, Simpson’s paradox, and interference/network effects (e.g., creator‑viewer ecosystems). - Decision rubric: primary metric wins with guardrails respected; use CIs, not only p‑values; discuss ramp strategy and blast radius. 5) Readout and decision (≈10 min) - Interpret a noisy or mixed set of results (e.g., +1.2% watch time, −0.4% retention in new markets; integrity reports flat). Provide a crisp recommendation with next steps: iterate, segment, or rollback. - Communicate like a Meta PM/DS partner: be direct, quantify impact, and tie to long‑term value. What interviewers evaluate (Meta‑specific): - Product sense grounded in Meta surfaces and ecosystems (Feed ↔ Reels ↔ Stories trade‑offs; Groups vs. Notifications; WhatsApp integrity/latency constraints). - Practical experimentation and metrics rigor with an owner mindset. - SQL comfort sufficient to translate questions into queries and spot data pitfalls quickly. - Integrity/privacy awareness and bias mitigation in metrics. - Communication that is structured, concise, and impact‑oriented. Typical prompts you might face: - Diagnose a sudden drop in Instagram Reels retention in one region/platform. - Design success metrics and an experiment for a new Groups notification variant. - Evaluate whether to launch a ranking tweak in Facebook Feed given mixed guardrails. Suggested time split: 5 min clarify; 15 min metrics; 15 min SQL/analysis; 15–20 min experiment & decision; 5 min wrap. Deliverables during the case: - A clear metric stack with guardrails and rationale. - A lightweight SQL outline to compute the key metrics/cohorts. - An experiment plan with checks (SRM, power, interference) and a go/no‑go rule. - A concise, leadership‑ready recommendation linked to user and business impact.