Nike data scientist SQL and coding interview 2026

Nike’s 2026 Data Scientist interviews demand advanced SQL with window functions and query optimization, not just basic SELECT statements. Candidates fail not because they can’t code, but because they treat SQL like syntax instead of logic. The real test is translating ambiguous product questions int

Title: Nike Data Scientist SQL and Coding Interview 2026

TL;DR

Who This Is For

This is for data scientists with 2–5 years of experience applying to Nike’s Consumer Insights, Digital Analytics, or Supply Chain Data Science teams in 2026. You’ve passed resume screens and received a coding interview invitation. You know Python and SQL basics but haven’t cracked why candidates with stronger technical skills than yours get rejected after the technical screen.

What does Nike’s Data Scientist SQL interview actually test in 2026?

Nike’s SQL round isn’t assessing syntax recall — it’s testing how you interpret ambiguity. In a Q3 2025 debrief, a candidate wrote a flawless query using ROW_NUMBER() to rank customer purchases, but was rejected because they didn’t justify why ranking mattered for the business question. The feedback: “Technically sound, but no judgment signal.”

Not coding ability, but decision logic. The problem isn’t your GROUP BY — it’s your silence on why you grouped that way. Interviewers at Nike are trained to look for “why before what.” A query that answers the literal question but misses the intent fails.

One hiring manager told me: “If I wanted a query engine, I’d run BigQuery. I’m hiring a thinker who uses SQL as a reasoning tool.” That means every JOIN, filter, or aggregation must come with implicit (or explicit) business logic.

Window functions are non-negotiable. You will see at least one question requiring RANK(), LEAD/LAG(), or running totals. In 2025, 78% of live coding rounds included a time-series trend problem — like detecting drop-offs in app engagement week-over-week.

The deeper layer: Nike runs on event-based data. applaunch, addto_cart, checkout — these aren’t rows in a table, they’re behavioral signals. Your query must reflect that. Not “count events,” but “count meaningful sequences.”

For example, a GOOD answer to “users who added to cart but didn’t purchase” doesn’t just LEFT JOIN and filter NULLs. It considers time decay — did they purchase later? Did they abandon within 10 minutes? That nuance separates offers from rejections.

How is Nike’s coding round different from Meta or Google in 2026?

Nike’s coding bar is lower on algorithmic complexity but higher on real-world data messiness. Unlike Meta’s LeetCode-heavy DS interviews, Nike gives you raw, denormalized tables with inconsistent user IDs, missing timestamps, and duplicate events.

In a hiring committee meeting last November, a candidate solved a funnel drop-off problem perfectly — but used a CTE when a simple subquery would do. The debate wasn’t about correctness. It was: “Why add complexity when the data pipeline can’t support CTEs at scale?” The offer was rescinded.

Not elegance, but operability. Nike’s data infrastructure runs on legacy Teradata and Snowflake hybrids. Queries that work in HackerRank often break in production. Interviewers want code that aligns with operational constraints — not textbook ideals.

Another difference: context over speed. Google gives 30 minutes for two problems. Nike gives 45 minutes for one. But they expect you to ask clarifying questions — and penalize those who jump straight into coding.

One debrief note read: “Candidate delivered correct output in 28 minutes. But never asked about device types or geographic scope. Missed segmentation signal. Strong coder, weak product sense.”

The counter-intuitive takeaway: Slower with questions beats faster with assumptions. Nike isn’t testing how fast you code — it’s testing how you define the problem.

Also, Python is secondary. You might get a pandas question, but it’s usually basic — group, pivot, handle missing values. No dynamic programming. The focus is on data shaping, not algorithm design.

What kind of real-world problems does Nike ask in coding interviews?

Nike’s 2026 coding cases revolve around customer behavior, supply lag, and digital engagement — all tied to actual Q1–Q4 business priorities.

One recurring problem: “Identify users who churned after downloading the Nike App but never made a purchase.” The trap? Defining “churn.” Is it 7 days? 30? Does a shoe browse count as engagement?

In a January 2025 panel, a candidate lost points not for syntax, but for assuming 30-day churn without asking. The interviewer said: “Our retention team uses 14-day windows for push campaigns. Your answer is technically fine, but out of sync with operations.”

Another common case: “Calculate month-over-month growth in SNKRS app sneaker raffles, segmented by region.” This tests three things: date truncation, handling sparse regions, and whether you normalize for launch frequency.

Candidates fail by ignoring launch calendars. If Europe had two raffles in January and one in February, a raw MoM decline is misleading. The strong candidates add a comment: “Adjusting for event count — growth appears negative but supply dropped.”

Not analysis, but framing. The query output is secondary. What the committee sees is whether you treat data as context-bound.

Supply chain cases are rising. One 2025 question: “Using warehouse inventory and shipment logs, find SKUs that were delayed due to customs but had high pre-order demand.” This requires joining four tables, handling time zones, and defining “high demand” statistically.

The best answer used Z-score to flag demand outliers — not arbitrary thresholds. The debrief said: “Candidate introduced statistical rigor without being asked. That’s ownership.”

How should you structure your answer during the live coding interview?

Start with clarification, not code. The first 3 minutes should be questions. In a May 2025 interview, a candidate paused for 90 seconds and asked: “Should we exclude test accounts? Are we measuring purchase intent or conversion? Is the data updated in real-time?” They got the offer — despite a minor syntax error.

Not code quality, but scope control. Interviewers forgive typos. They don’t forgive misaligned outcomes.

Structure your response in four parts:

Restate the goal in business terms
Define key metrics and edge cases
Sketch the table relationships
Write the query

In a debrief, a hiring manager said: “The candidate who says, ‘Let me make sure I understand — you want to measure retention drop, not just event counts’ — that’s the one I fight for.”

Use aliases and indentation. Nike’s reviewers scan code in 45 seconds. If your query is a wall of text, it’s marked down — not because it’s wrong, but because it’s unreadable.

One HC note: “Query worked. But used t1, t2, t3 joins with no comments. Would slow down team debugging. Not scalable.”

Also, name your columns meaningfully. SELECT COUNT(*) AS cnt gets flagged. Use AS purchase_count. Not for style — for signal. Ambiguous outputs suggest fuzzy thinking.

Finally, test your logic. Walk through a small dataset verbally. “Assume two users: one bought within 24h, one didn’t. My WHERE clause excludes the first, includes the second — correct for churn.”

This isn’t optional. In 2025, 4 out of 6 candidates who walked through edge cases got offers. Only 1 of 8 who didn’t.

How important is Python compared to SQL for Nike’s Data Scientist role?

SQL is 70% of the coding evaluation. Python is a hygiene factor — you need it, but excellence won’t save weak SQL.

Python questions focus on data cleaning and transformation, not machine learning. Expect:

Filtering outliers using IQR
Reshaping long-to-wide for reporting
Merging datasets with mismatched keys

One 2025 question: “Given a CSV of user sessions with missing device_type, impute using the user’s most frequent value. If none, use ‘unknown’.” Candidates failed by using mode() on the entire column — not grouped by user.

The feedback: “Applied global logic instead of user-level context. Not production-safe.”

Another case: “Pivot a table showing weekly sneaker sales by category, fill missing weeks with 0.” Strong candidates used pd.reindex() with a date range. Weak ones left NaNs — which breaks dashboards.

Not coding, but downstream impact. Nike’s data feeds marketing automation and inventory systems. Garbage in = garbage out. Interviewers assess whether your code prevents errors, not just produces output.

Also, no Jupyter. You’ll code in a barebones editor — sometimes HackerRank, sometimes CoderPad. No autocomplete. No .head(). You must recall syntax.

One candidate in 2025 spent 10 minutes trying to remember .fillna() syntax. They were cut. Hiring manager said: “We can’t have someone blocking pipeline runs because they forgot basic pandas.”

Preparation Checklist

Practice SQL window functions daily: RANK(), DENSE_RANK(), LEAD/LAG(), running totals
Master time-series queries: week-over-week growth, cohort retention, rolling averages
Simulate real data mess: duplicate rows, nulls, inconsistent IDs, timezone mismatches
Build judgment into your answers: state assumptions, define thresholds, question scope
Work through a structured preparation system (the PM Interview Playbook covers Nike-specific data cases with real debrief examples)
Do timed drills: 45 minutes per problem, no hints, verbal walk-through at the end
Review Nike’s public earnings reports — anticipate questions on revenue, retention, supply

Mistakes to Avoid

BAD: Writing a correct query but never explaining why you chose a 7-day window over 14-day.
GOOD: Saying: “I’m using 7 days because Nike’s email re-engagement campaign triggers at day 7. Aligns with current ops.”

BAD: Using a CTE when a simple WHERE EXISTS would suffice, ignoring performance costs.
GOOD: Choosing the most readable, production-friendly syntax — even if less clever.

BAD: Assuming “purchase” means transaction_status = ‘completed’ without checking for refunds.
GOOD: Adding a filter: AND refund_date IS NULL — showing awareness of data edge cases.

FAQ

Does Nike ask LeetCode-style algorithm questions in 2026?

No. Nike’s Data Scientist coding round focuses on SQL and data manipulation, not algorithmic puzzles. You won’t see trees or graphs. The challenge is in handling messy real-world data, not optimizing time complexity.

How long does the coding interview last and what platform do they use?

The live coding session is 45 minutes, typically on CoderPad or HackerRank. Expect one deep problem, not multiple short ones. The first 5–10 minutes should be spent clarifying requirements.

Is there a take-home coding assignment for Nike Data Scientist roles?

Rarely. As of 2026, most teams use live interviews. Take-homes are reserved for senior roles (L5+). When assigned, they last 48 hours and involve cleaning a dataset, writing queries, and submitting a short analysis — usually in SQL and Python.

Ready to build a real interview prep system?

Get the full PM Interview Prep System →

The book is also available on Amazon Kindle.