Talentee
Amazonadvanced

Amazon Data Scientist Interview Questions

Amazon's DS role splits across Data Scientist (analytics-heavy), Business Intelligence Engineer (SQL + dashboards), and Applied Scientist (ML research). Each has a different interview. This guide covers the Data Scientist track.

Process length
6-10 weeks
Rounds
7
Questions
10
Mid-level TC
$210k–$280k (L5)
Practice Amazon questions with AI

The Amazon Data Scientist interview process

What to expect, in order.

  1. 1Online assessment (work simulation + 1-2 stats/SQL problems)
  2. 2Phone screen (60 min — SQL + 1 Leadership Principle question)
  3. 3Onsite — typically 5 rounds
  4. 4Stats / Experimentation round (60 min)
  5. 5SQL deep-dive round (60 min)
  6. 6ML round (60 min — model selection, evaluation, business framing)
  7. 7Bar Raiser (60 min — behavioral with deep LP probing)

What Amazon actually evaluates

Amazon DS work is famously ownership-heavy: you'll write your own SQL, build your own dashboards, run your own A/B tests, and present results to leadership directly. The interview reflects this — no specialization handoffs, everyone has to do it all.

Customer Obsession — analytics that drive customer wins
Dive Deep — data dives that go beyond surface metrics
Are Right A Lot — calibrated judgment under uncertainty
Deliver Results — analyses that change business action

Process quirks worth knowing

Like SWE roles, the Bar Raiser is a critical gate. They will dig 4-5 layers deep on Leadership Principles, especially Customer Obsession and Dive Deep. Behavioral answers must follow STAR + LP mapping strictly.

10 questions Amazon actually asks

Each question includes the tip for answering and what the interviewer is actually evaluating.

Q1behavioral

Tell me about a time you used data to influence a business decision.

Why Amazon asks: Maps to Deliver Results + Customer Obsession + Are Right A Lot. Bar Raiser wants to see analysis that actually changed direction, not just informed it.
How to answer: Use STAR strictly. Show: business question, what data you pulled, what surprised you, the recommendation, the action taken, the result. Quantify the business outcome.
What they evaluate: Genuine business impact, ability to translate analysis into recommendation, ownership of the decision
Q2case

Design an experiment to measure the impact of free shipping on Amazon Prime sign-ups.

How to answer: Discuss: randomization unit (user? session? geo?), treatment (full free shipping vs current threshold), metrics (sign-up rate primary, AOV counter), sample size, duration (control for shipping seasonality), business interpretation (LTV math on lifetime margin).
What they evaluate: Experimental design rigor, awareness of business context, counter-metric thinking, ability to discuss financial implications
Q3technical

Write a SQL query to find products whose sales rank dropped the most week-over-week.

How to answer: Use LAG() window function to get last week's rank per product. Compute rank delta. ORDER BY delta DESC LIMIT N. Handle products that didn't exist last week.
What they evaluate: Window function fluency, comfort with NULL handling for new products, clean SQL
Q4behavioral

Tell me about a time you went deep into data when others wanted to move on.

Why Amazon asks: Pure Dive Deep LP. Amazon famously values diving beyond surface metrics when something looks off.
How to answer: Pick a real example where you found something surprising by going one level deeper. Show: what looked obvious at first, what made you suspicious, what you did, what you found, the impact.
What they evaluate: Genuine curiosity, willingness to push back on consensus, ability to find non-obvious insights
Q5case

You built a model that predicts customer churn with 90% accuracy. Stakeholders are excited. What do you push back on?

How to answer: Accuracy is misleading for imbalanced classes. If only 10% churn, predicting 'no churn' for everyone is 90% accurate. Discuss: precision/recall, business cost of false positives vs false negatives, calibration, actionability.
What they evaluate: Awareness of imbalanced data issues, ability to push back on misleading metrics, business framing of ML metrics
Q6behavioral

Tell me about a time you disagreed with a senior leader on data interpretation.

Why Amazon asks: Maps to Have Backbone; Disagree and Commit. Amazon wants people who push back hard with data, then commit fully if outvoted.
How to answer: Show: the disagreement, the data you brought, how you stayed respectful, the outcome. If you lost, show how you committed fully afterward.
What they evaluate: Comfort with conflict, data-driven argument, ability to commit after losing
Q7technical

Write SQL to find sessions where a user added an item to cart but didn't purchase, then purchased a similar item within 7 days.

How to answer: Self-join the events table on user_id and product category. Filter for: event_a = 'add_to_cart' without 'purchase' within session, event_b = 'purchase' in same category within 7 days. Use date arithmetic.
What they evaluate: Comfort with complex joins, date arithmetic, ability to translate ambiguous business questions into SQL
Q8technical

Explain the difference between bagging and boosting. When would you use each?

How to answer: Bagging (Random Forest): parallel weak learners, reduces variance, robust to noise. Boosting (XGBoost): sequential weak learners that fix previous errors, reduces bias, more sensitive to outliers but often higher accuracy. Use bagging when training data is noisy, boosting when accuracy is critical and data is clean.
What they evaluate: Conceptual clarity, business-relevant tradeoff framing, awareness of practical considerations like training time
Q9behavioral

Tell me about a time your analysis was wrong.

Why Amazon asks: Tests intellectual humility + Are Right A Lot (which is paradoxically about being calibrated, including knowing when you're wrong).
How to answer: Pick a real mistake. Show what you missed, how you found out, what you did to correct, and how you changed your subsequent process. Don't blame data quality without owning your role.
What they evaluate: Ownership of the mistake, structured process improvement, calibration improvement
Q10values

Why Amazon over Google or Meta for DS roles?

How to answer: Connect to Amazon's specific data scale (massive transactional + logistics data) and the ownership culture. Show you've researched the team and the LP that resonates most.
What they evaluate: Genuine LP alignment, specific team interest, signal of multi-year intent

Common ways candidates fail this interview

Specific to Amazon, not generic interview advice.

  • ⚠️Treating it like Google DS — Amazon's behavioral round is much more probing
  • ⚠️Vague stories without LP mapping in behavioral
  • ⚠️Strong stats but weak SQL — Amazon DS write a lot of SQL daily
  • ⚠️Missing the BIE vs DS vs Applied Scientist distinction when applying
  • ⚠️Generic 'why Amazon' — Bar Raiser will probe LP alignment specifically

Amazon Data Scientist compensation (2026)

Entry / Junior
$155k–$190k total comp (L4)
Mid-level
$210k–$280k total comp (L5)
Senior+
$320k–$450k total comp (L6)

Sources: levels.fyi, Glassdoor, public filings (US figures, total compensation including base + bonus + equity).

Practice these questions with a live AI interviewer

Nova is Talentee's voice AI interviewer. Speak your answer out loud, get scored on structure, clarity, and confidence, with a detailed PDF report.

Start Free PracticeAll Data Scientist questions