Before the Technical Phone Screen. 5 single-table-and-basic-join problems that warm up the muscle memory you'll need when the recruiter hands you off to a senior IC.

Netflix's recruiter + hiring manager screens are conversation-heavy — Freedom & Responsibility alignment, motivation, past scope of impact. No live SQL yet, but the next call is a live editor session. If any of these problems feel slow, fix that first.

• Most relevant for: every Netflix data role (universal floor — DS-Analytics, DS-Inference, DS-Algorithms, DE, Analytics Engineer)
• Skip if: you're senior — jump to Stage 2
• Domain: Netflix streaming surface (users, subscriptions, titles, payments, playback)

Netflix's first real loop round — 45–60 min on a shared editor with a senior IC. Typical report: live SQL + light Python + applied-stats discussion. For this SQL-only collection, this stage focuses on the SQL side.

Per InterviewQuery and Prepfully, "Netflix interviews are less about trick questions and more about applied reasoning." You are expected to be comfortable with SQL, metrics, and experimentation — but interviewers focus heavily on judgment and how you reason through ambiguous problems.

Patterns tested:

• 2–3 table JOINs across the streaming surface
• Filter-with-aggregation (AVG, SUM) — Netflix's bread-and-butter watch-time metrics
• GROUP BY with HAVING (multi-device user detection)
• Conditional rate (refund rate, completion rate)
• First window function (RANK partitioned by genre)
• Multi-month date arithmetic for subscription duration

Train yourself to write each in 10–12 minutes and to narrate your approach aloud as you write — Netflix scores articulation and judgment, not just syntax.

• Most relevant for: DS-Analytics, DS-Inference, DE, Analytics Engineer
• Also mirrors: Spotify technical screen, Hulu phone interview

The dedicated 60-min SQL round inside Netflix's onsite. A peer IC walks you through 1–2 multi-step problems, then probes:

"How would you validate this metric against the experiments dashboard?" "What happens if playback events arrive 4 hours late?" "How would you handle a profile that switches devices mid-session?"

Patterns tested (the Netflix canon):

• Top-N within group — ROW_NUMBER() OVER (PARTITION BY genre ORDER BY ...)
• Per-user numbering — chronological playback session ordering
• Running totals & moving averages — SUM(...) OVER (... ROWS BETWEEN ...) for monthly watch-hour trend
• LAG diffs — payment-to-payment delta, sequential title changes
• Anti-joins — "titles never watched" / "watchers who never rated"
• Quartile / NTILE classification — engagement-tier segmentation
• Watch-time aggregations with thresholds (the Netflix-signature pattern)

This is where Netflix separates fluent from experienced — window-function depth and the ability to think about late-arriving streaming events.

• Most relevant for: DS-Analytics, DS-Algorithms, DE, Analytics Engineer
• Also mirrors: Disney+ analytics onsite, YouTube data onsite

Netflix's iconic round — and the one no other FAANG can match. Netflix is the only FAANG with a separately-titled "Senior Data Scientist — Experimentation & Causal Inference" track at scale (confirmed by current Netflix job listings on aijobs.net, startup.jobs, and Netflix Research's own page). The Netflix Tech Blog has an entire post titled "A Day in the Life of an Experimentation and Causal Inference Scientist".

Netflix is also one of the documented production users of CUPED variance reduction (Bytepawn, Microsoft Research) — alongside Microsoft, Booking, Meta, and Airbnb. Expect it to come up.

Patterns tested:

• A/B test reads — control vs. treatment lift on engagement metrics (watch time, completion rate)
• Sample-size adequacy — can we trust this test result yet?
• Weekly retention curves — first-watch cohort → return-in-week-N (Netflix's canonical retention question)
• Churn-risk segmentation — multi-signal hazard scoring
• Subscriber lifetime value — d4 capstone with multi-CTE revenue + tenure logic
• Refund rate by plan — revenue mix diagnostics

The probing follow-ups senior interviewers ask:

"How would you adjust for pre-period covariates?" (CUPED) "This is a ratio metric — how do you compute the variance correctly?" (delta method) "What if one user's behavior changes another's via shared accounts?" (interference) "How do you check for SRM (sample ratio mismatch) before reading the result?"

If you skip this stage, you skip the round that decides Netflix data offers.

• Most relevant for: DS-Inference, DS-Analytics, DS-Algorithms
• Also mirrors: Microsoft Experimentation Platform interview, Booking experimentation round, Spotify causal-inference team

The 60-min product-sense round. The interviewer hands you an under-specified business prompt:

"Engagement on Originals dropped 8% this week — why?" "Design the dashboard you'd use to evaluate a new tier launch." "What metric would you build to detect a payment-processor outage early?"

You're being graded on judgment, scope, and metric design. The SQL is downstream — the harder skill is deciding what to compute, in what order, and what to escalate.

The 6 problems are deliberately multi-CTE composite-score scorecards modeled on Netflix's analytics dashboards:

• Content performance scorecard — multi-metric composite per title
• Content engagement scorecard — d4 deep-dive (3+ CTE chain)
• Content catalog health — d4 capstone (catalog freshness × engagement × churn)
• Streaming platform user dashboard — d4 (user health composite)
• Payment health dashboard — revenue-side ops view
• Device platform analytics — multi-platform engagement breakdown

Every explainer ends with a "How a senior interviewer would extend this" section — the causal-thinking and ambiguity-tolerance probes Netflix specifically asks ("what would you investigate next?", "push back on the framing", "what's the right metric vs. the easy metric?").

• Most relevant for: all DS tracks at Netflix (Analytics, Inference, Algorithms)
• Also mirrors: Spotify product-analytics, Disney+ DSP round