Requirements
Group requirements by domain, show scale anchors, and separate real-time, batch, and governance expectations.
On this page
Scope + pro tip
Turn business questions into freshness, correctness, and SLA contracts.
I group requirements by data domain, then map each one to freshness, correctness, consumers, and the tables or streams that answer it.
Quick jumps
Click to jumpMonthly users
200M-250M
Docs use both 200M+ and 250M assumptions
Daily active users
80M
Used for heartbeat and Kafka math
Peak concurrency
15M
Useful for QoE and live ops workloads
Map business questions to data outputs
Clarify what the data platform must produce, how fresh it must be, and which trade-offs matter.
Pro tip
Scope Netflix as a data platform.
Design events → ingestion → processing → lakehouse → serving. Skip playback APIs, CDN routing, and DRM internals unless the interviewer explicitly asks.
6 business metrics that drive the entire platform design
Watch Hours
Content Performance
Trending Titles
QoE Metrics
Recommendation Feedback
Experiment Metrics
Anchor the discussion with scale numbers
Each number should connect to a design decision — Kafka partitions, streaming parallelism, or storage layout.
Monthly users
200M-250M
Docs use both 200M+ and 250M assumptions
Daily active users
80M
Used for heartbeat and Kafka math
Peak concurrency
15M
Useful for QoE and live ops workloads
Heartbeat events/day
19.2B
80M DAU x 2h x heartbeat every 30s
Peak ingest
1M-2M events/s
Peak multiplier applied to total events/sec
Raw data/day
50-100 TB
Playback, browse, search, errors, CDN, billing, CDC
Group requirements by domain, not by table type
Playback, QoE, recommendations, experiments, and governance map directly to the data pipelines you will design.
Requirement-to-pipeline mapper
Click a requirement on the left to highlight its flow.
How many hours were streamed per title?
Freshness SLA
Near real-time + daily final
Correctness need
Very high
Consumer
Gold table + BI dashboard
Domain
User Engagement Analytics
video.heartbeat
Kafka playback.heartbeat.raw
Flink watch-time aggregator
silver.fact_watch_session
gold.rpt_content_daily_metrics
BI dashboard
Freshness SLA timeline — where each metric sits
Ask these clarifying questions first
Are we designing for real-time dashboards, ML features, or both?
What is the expected freshness for the most latency-sensitive consumer?
Do we need to support replay and correction of historical data?
Is there a compliance or data-retention policy I should design around?
What is the primary consumer — BI analysts, on-call engineers, or ML models?
Should I design for one region or multi-region with data sovereignty constraints?
DE non-functional requirements
Kafka ingestion 99.99%, critical streaming pipelines 99.9%, BI warehouse 99.5%+
No event loss for critical events, exactly-once or idempotent writes, and replay-safe correction paths
Correct watch-time calculation, deduplication, sessionization, and stream-vs-batch reconciliation
PII hashing or tokenization, encryption in transit and at rest, RBAC, row/column policies, and audit logs
GDPR deletes, regional residency, retention policies, and late-arriving data handling built into the platform
Common interview mistake
Jumping to tools (Kafka, Flink, Iceberg) before stating freshness requirements and business consumers. Interviewers want to see that you know why you need near-real-time before you name a streaming engine.
What to skip in a 45-min interview
Skip compliance details, multi-region specifics, and advanced governance unless the interviewer asks. Cover freshness, correctness, replay, and one failure mode per layer instead.
Freshness and non-functional requirements
DE interviews care about freshness, correctness, replay, and quality SLAs just as much as raw throughput.
Freshness ladder
Live concurrent viewer dashboard
5 seconds to 1 minuteTrending content detection
1 to 5 minutesFraud or anomaly alerting
Seconds to 1 minuteResume watching position
Sub-second to a few secondsReal-time recommendation features
Seconds to minutesDaily content performance report
Available by morningMonthly revenue report
Hours acceptableML training dataset
Daily or every few hoursDE non-functional requirements
Kafka ingestion 99.99%, critical streaming pipelines 99.9%, BI warehouse 99.5%+
No event loss for critical events, exactly-once or idempotent writes, and replay-safe correction paths
Correct watch-time calculation, deduplication, sessionization, and stream-vs-batch reconciliation
PII hashing or tokenization, encryption in transit and at rest, RBAC, row/column policies, and audit logs
GDPR deletes, regional residency, retention policies, and late-arriving data handling built into the platform
Requirements framing in interview language
Open requirements with domain, consumer, freshness, and correctness — not a feature list.
Say This In Interview
I group requirements by data domain, then map each one to freshness, correctness, consumers, and the tables or streams that answer it.
30-second answer
FastI group requirements by data domain, then map each one to freshness, correctness, consumers, and the tables or streams that answer it.
2-minute answer
BalancedGroup requirements by domain, show scale anchors, and separate real-time, batch, and governance expectations. Focus on FOUNDATION, 8 min, Turn business questions into freshness, correctness, and SLA contracts..
Deep answer
DeepI group requirements by data domain, then map each one to freshness, correctness, consumers, and the tables or streams that answer it. Then go layer by layer through FOUNDATION, 8 min, Turn business questions into freshness, correctness, and SLA contracts., including trade-offs, failure handling, and why this shape is right for the consumer SLAs.
Last reviewed June 2026 · By Prasoon Parashar
Numbers are interview assumptions, not real Netflix internal figures.