Data Engineer Interview Experience at Netflix – 2026

This recruiter screen is not a light logistics chat - the recruiter asks about experience and whether it genuinely maps to the team's needs, plus two direct culture-fit questions grounded in Netflix's public culture memo, specifically what 'freedom and responsibility' means in a past role and whether the candidate can describe a time they made a significant decision without waiting for manager sign-off.

This round is live SQL against a schema the interviewer describes verbally, covering content-viewing and subscriber-event data, with three questions of increasing difficulty. The first asks for computing 28-day retention for subscribers who started in a given month, solvable with a self-join on the subscriber and activity tables filtered by date offset. The second is a sessionization problem: given raw play events with timestamps, group them into viewing sessions where consecutive events under a 30-minute gap belong to the same session - the expected approach uses LAG to compute the gap since the previous event per user, then a running SUM to assign session IDs whenever the gap exceeds the threshold. The third asks for computing statistical lift and segment-level differences for an A/B test result, approachable with a straightforward group-by on treatment arm and segment; candidates often run out of time to fully implement a significance check and describe the approach verbally instead.

Candidates are asked to design a data pipeline tracking viewership data end to end, from raw play events to an analytics-ready table supporting both real-time dashboards and historical cohort analysis. A strong approach proposes a streaming ingestion layer (Flink) doing initial sessionization close to real time, landing raw and lightly-processed events into an Iceberg table, with a separate batch layer building slowly-changing dimension tables for subscriber attributes. Interviewers typically push on grain decisions specifically - whether the fact table should be at the individual play-event grain or the session grain - and a well-reasoned answer argues for keeping both, an event-grain table for flexibility and a pre-aggregated session-grain table for the common dashboard queries, since forcing every dashboard query to re-sessionize from raw events would be needlessly expensive at Netflix's scale.

This round focuses on working with non-engineering stakeholders (content and product analytics teams) with ambiguous or shifting requirements. Candidates are asked to describe a time a stakeholder's data request changed significantly midway through a project, and how the rework was handled - a strong answer describes reprioritizing the remaining sprint around the new requirement rather than finishing the original spec first, and communicating the tradeoff explicitly to the stakeholder rather than absorbing the schedule slip silently.

This round is specifically named and structured around Netflix's culture of high autonomy and minimal top-down oversight, distinct from a standard behavioral interview. Candidates are asked directly for a specific story about giving difficult, direct feedback to a peer or pushing back on a decision made by someone more senior than them. A common pitfall is an answer centered on receiving hard feedback well and changing approach because of it - interviewers tend to note, politely but directly, that this is a story about receiving feedback rather than initiating a difficult conversation, and ask for a different example. Not having a strong second story ready is a frequent reason candidates don't recover from this gap.