Amazon Data Scientist Interview

Prepping for an Amazon Data Scientist interview?

Grinding LeetCode, memorizing stats, and writing SQL is natural. Those skills matter. But they rarely cause rejections.

The Missing Piece: Leadership Principles I have coached over 40 candidates. The real killer is the 16 Leadership Principles. Candidates usually fail because they lack qualities like Customer Obsession or Ownership.

• The Structure: Every interviewer is assigned 2 or 3 specific principles to evaluate you on.
• The Challenge: You can answer every technical question correctly and still not get the offer if your behavioral answers do not align with these principles.

A Real Example: I worked with a Senior Data Scientist who was a brilliant coder with a strong background. He did well in the technical rounds but was ultimately turned down. The feedback was that he "seemed to wait for permission rather than taking initiative."

This guide is designed to help you bridge that gap. We will cover how to weave the Leadership Principles into your stories, the technical bar you need to meet, and the common pitfalls to avoid.

Amazon Data Scientist Role

Amazon has over 2,000 data scientists across retail, AWS, Alexa, advertising, and operations. This is an applied science role. You're not publishing papers. You're shipping models that move revenue.

A Typical Week

What the job description won't tell you: a real chunk of your week goes to data quality fires. Upstream pipelines break, schemas change without notice, and you'll debug more ETL than you ever expected.

Forecasting and optimization. Demand forecasting, delivery routes, inventory placement. A 0.5% accuracy gain moves millions here.

Customer-facing ML. Rec engines, search ranking, Alexa NLU. You're A/B testing at a scale most companies never touch.

Internal tools and measurement. Causal inference, experimentation platforms, metric definitions. Less sexy, but this is where DS has the most org-wide leverage.

Skills & What's Expected

Statistics and ML get the headlines, but day-to-day you'll lean on SQL and communication just as much. Explaining a tradeoff to a PM who doesn't know what a p-value is matters as much as running the test itself.

Levels & Career Growth

Most people get stuck at the L5 to L6 jump. L5 means you execute well. L6 means you own a problem end-to-end and influence beyond your team. Fundamentally different skill set.

You can be the best modeler on the team and still not make Senior if you can't demonstrate Leadership Principles with concrete examples. Amazon rewards people who are both technically sharp and organizationally aware.

Work Culture

Amazon's culture is not for everyone, and that's by design. The pace is fast, feedback is direct, and two-pizza teams mean real ownership. When something breaks, it's yours.

Amazon Data Scientist Compensation

Amazon pays differently than most big tech companies. The numbers are in the chart above. Here is how the structure actually works and how to negotiate.

📅 The Back-Loaded Vesting Schedule

Most tech companies vest stock equally over four years. Amazon does not.

• Year 1: 5%
• Year 2: 15%
• Year 3: 40%
• Year 4: 40%

To make up for low stock in the first two years, Amazon offers large sign-on bonuses paid out monthly. This acts as a cash bridge until your stock grants hit in Year 3.

💰 Refresher Grants

Starting around your second year, Amazon issues additional RSU grants (refreshers) on top of your original package. This is how total comp stays high after your initial four-year grant runs out. The size of your refresher depends on performance and level. Strong performers often see their Year 3+ comp exceed the original offer.

🤝 How to Negotiate

Get the recruiter on the phone. Here is what moves and what does not.

• Base Salary. Rigid. Amazon rarely moves on this.
• Sign-on Bonus. Flexible. This is your best leverage point.
• RSUs. Negotiable. Competing offers are the strongest lever here.
• Level. The most important number in your offer. Pushing for L6 over L5 is worth more long-term than any base salary negotiation.

The Golden Handcuffs

They are real. If you stay past Year 2, the payout jumps significantly. Many people who plan to leave after two years end up staying because the math changes in Year 3.

Amazon Data Scientist Interview Process

Amazon Data Scientist Interview Questions

Alright, let's get into the actual questions. I'm going to organize these by type, but remember—most rounds blend technical and behavioral. Don't expect clean separation.

1/* Daily GMS and 7-day trailing moving average by marketplace, last 30 days */
2WITH filtered_items AS (
3  SELECT
4    oi.order_id,
5    (oi.quantity * oi.item_price) AS item_revenue
6  FROM order_items oi
7  WHERE COALESCE(oi.is_refund, FALSE) = FALSE
8),
9item_revenue_by_order AS (
10  /* Collapse to order grain so later joins cannot multiply revenue */
11  SELECT
12    fi.order_id,
13    SUM(fi.item_revenue) AS order_revenue
14  FROM filtered_items fi
15  GROUP BY fi.order_id
16),
17daily_gms AS (
18  SELECT
19    DATE_TRUNC('day', o.order_ts) AS order_day,
20    o.marketplace_id,
21    SUM(iro.order_revenue) AS gms
22  FROM orders o
23  JOIN item_revenue_by_order iro
24    ON o.order_id = iro.order_id
25  WHERE o.order_ts >= DATEADD(day, -30, CURRENT_DATE)
26  GROUP BY 1, 2
27)
28SELECT
29  dg.order_day,
30  dg.marketplace_id,
31  dg.gms,
32  AVG(dg.gms) OVER (
33    PARTITION BY dg.marketplace_id
34    ORDER BY dg.order_day
35    ROWS BETWEEN 6 PRECEDING AND CURRENT ROW
36  ) AS gms_7d_trailing_ma
37FROM daily_gms dg
38ORDER BY dg.order_day, dg.marketplace_id;

1from __future__ import annotations
2
3from collections import defaultdict
4from datetime import datetime, timezone
5from typing import Iterable, List, Tuple, Set
6
7
8def longest_same_query_daily_streak(
9    events: Iterable[Tuple[str, str, int]]
10) -> int:
11    """Return the longest consecutive-day streak where a customer issued the same query.
12
13    Each input is (customer_id, query, ts_epoch_seconds). Multiple events in the same
14    UTC day count once. Streak is measured in consecutive UTC dates.
15
16    Time: O(n) expected, Space: O(n)
17    """
18
19    # Step 1: Normalize timestamps into UTC dates and dedupe within a day.
20    days_by_pair: defaultdict[Tuple[str, str], Set[int]] = defaultdict(set)
21
22    for customer_id, query, ts in events:
23        # Convert epoch seconds to a UTC date, then map to an integer day index.
24        # Using day index makes consecutive checks cheap.
25        dt = datetime.fromtimestamp(ts, tz=timezone.utc)
26        day_index = dt.toordinal()  # days since year 1-01-01
27        days_by_pair[(customer_id, query)].add(day_index)
28
29    # Step 2: For each (customer, query), compute longest consecutive run.
30    best = 0
31    for _, day_set in days_by_pair.items():
32        if not day_set:
33            continue
34
35        # Standard longest-consecutive-sequence pattern using a set.
36        for d in day_set:
37            # Only start counting from the beginning of a streak.
38            if (d - 1) in day_set:
39                continue
40
41            length = 1
42            cur = d
43            while (cur + 1) in day_set:
44                cur += 1
45                length += 1
46
47            if length > best:
48                best = length
49
50    return best
51
52
53if __name__ == "__main__":
54    sample = [
55        ("c1", "laptop stand", 1704067200),  # 2024-01-01
56        ("c1", "laptop stand", 1704070800),  # same day duplicate
57        ("c1", "laptop stand", 1704153600),  # 2024-01-02
58        ("c1", "laptop stand", 1704326400),  # 2024-01-04 (break)
59        ("c2", "headphones", 1704067200),
60        ("c2", "headphones", 1704153600),
61        ("c2", "headphones", 1704240000),
62    ]
63    print(longest_same_query_daily_streak(sample))  # c2 has 3-day streak
64

Amazon Leadership Principles (Behavioral)

Amazon is unique in a lot of ways, but their 16 Leadership Principles are the biggest hurdle for technical candidates. These are not just corporate buzzwords to memorize. They form the exact grading rubric your interviewers use to evaluate your culture fit and problem solving as a Data Scientist.

⭐️ 16 Amazon Leadership Principles

Learn these by heart.

1. Customer Obsession: Leaders start with the customer and work backwards. They work vigorously to earn and keep customer trust.
2. Ownership: Leaders are owners. They think long term and do not sacrifice long-term value for short-term results.
3. Invent and Simplify: Leaders expect and require innovation and invention from their teams and always find ways to simplify.
4. Are Right, A Lot: Leaders are right a lot. They have strong judgment and good instincts.
5. Learn and Be Curious: Leaders are never done learning and always seek to improve themselves.
6. Hire and Develop the Best: Leaders raise the performance bar with every hire and promotion.
7. Insist on the Highest Standards: Leaders have relentlessly high standards. Many people may think these standards are unreasonably high, but leaders are continually raising the bar.
8. Think Big: Thinking small is a self-fulfilling prophecy. Leaders create and communicate a bold direction that inspires results.
9. Bias for Action: Speed matters in business. Many decisions and actions are reversible and do not need extensive study.
10. Frugality: Accomplish more with less. Constraints breed resourcefulness, self-sufficiency, and invention.
11. Earn Trust: Leaders listen attentively, speak candidly, and treat others respectfully.
12. Dive Deep: Leaders operate at all levels, stay connected to the details, audit frequently, and are skeptical when metrics and anecdote differ.
13. Have Backbone; Disagree and Commit: Leaders are obligated to respectfully challenge decisions when they disagree. Once a decision is determined, they commit wholly.
14. Deliver Results: Leaders focus on the key inputs for their business and deliver them with the right quality and in a timely fashion.
15. Strive to be Earth's Best Employer: Leaders work every day to create a safer, more productive, higher performing, more diverse, and more just work environment.
16. Success and Scale Bring Broad Responsibility: We are big, we impact the world, and we are far from perfect. We must be humble and thoughtful about even the secondary effects of our actions.

💡 The No-BS Insight: You can write flawless Python and know the ins-and-outs of A/B tests, but if you fail the Leadership Principles, you will be rejected. So, don't wing it. Use this secret sauce for prep:

• Build a Story Matrix: Do not try to memorize 16 different stories. Instead, outline 8 to 12 strong, versatile projects from your past. Map each project to two or three different Leadership Principles so you can pivot on the fly.
• Format with STAR: Every answer must follow the Situation, Task, Action, Result framework. Keep the setup (Situation and Task) brief. Spend the bulk of your time detailing exactly what you did (Action) and the measurable business impact (Result).
• Say "I", Not "We": This is a critical trap. Amazon wants to know what you built, not what your team accomplished. If you say "we trained a model," the interviewer will stop you and ask exactly which part you coded.
• Quantify the Impact: Data Scientists live and die by metrics. Your results must include hard numbers. Did your model reduce latency, increase conversion by 4 percent, or save $100,000? If a project failed, quantify the cost and explain the lesson learned.
• Prepare for Pushback: The Bar Raiser will cross-examine your stories. They will ask why you chose a specific algorithm, what trade-offs you made, and what you would do differently today. Surface-level answers will fall apart here.

👎 The Bad Answer

"Users complained about irrelevant product recommendations. I retrained the model with new data. After that, the recommendations were much better and engagement went up."

Why it fails: Lacks context, avoids personal ownership, and completely ignores quantifiable metrics, offline evaluation, or A/B testing results.

👍 The Good Answer (STAR Framework)

• Situation: At my previous e-commerce startup, VIP customers experienced a 20 percent drop in engagement because the recommendation engine surfaced irrelevant, out-of-season items.
• Task: I needed to diagnose the model degradation, engineer a solution, and validate the improvement without disrupting the baseline user experience.
• Action: I conducted an exploratory data analysis and discovered seasonal trend weights were decaying too slowly. I engineered new time-decay features and trained an XGBoost model. I then designed and launched a two-week A/B test to evaluate the new model against the control group.
• Result: The A/B test achieved a statistically significant 5 percent lift in click-through rate. We launched the model globally, recovered the 20 percent engagement drop, and drove $500,000 in incremental quarterly revenue

Want to learn more? Click this video👇

Amazon Data Scientist Interview Prep

Understand the Business

Amazon interviewers expect you to know how the company makes money and where data science fits. You don't need to memorize annual reports, but you should be able to connect your work to a business segment and explain why it matters.

Before your loop, pick 2-3 segments that align with the team you're interviewing for. Be ready to propose a metric you'd track and explain why.

Leadership Principles Are Not Optional

Every single round at Amazon includes behavioral questions. The Bar Raiser round is entirely behavioral. You need 8-10 STAR stories mapped to specific Leadership Principles before you walk in.

The principles that come up most for data scientists: Customer Obsession, Dive Deep, Bias for Action, Have Backbone; Disagree and Commit, and Deliver Results. Rehearse each story out loud until you can tell it clearly in under 3 minutes.

Technical Preparation

Amazon DS technical rounds test breadth over depth. You won't build a transformer from scratch, but you will explain when logistic regression beats XGBoost and why.

SQL: Expect at least one live SQL question. Window functions (RANK, ROW_NUMBER, LAG), self-joins, and CTEs are fair game. Practice writing queries without an IDE.

Statistics: A/B testing comes up in nearly every loop. Know how to calculate sample size, pick the right test, and explain what to do when assumptions are violated.

Machine Learning: Focus on intuition over math. Be ready to walk through a full modeling pipeline: problem framing, feature selection, model choice, evaluation metrics, and how you'd deploy it.

1WITH latest_signup AS (
2  SELECT
3    user_id,
4    MAX(signup_date) AS signup_date
5  FROM prime_signups
6  GROUP BY user_id
7), post_signup_orders AS (
8  SELECT
9    o.user_id,
10    ls.signup_date,
11    o.order_id,
12    o.order_date,
13    ROW_NUMBER() OVER (
14      PARTITION BY o.user_id
15      ORDER BY o.order_date, o.order_id
16    ) AS rn
17  FROM orders o
18  JOIN latest_signup ls
19    ON o.user_id = ls.user_id
20   AND o.order_date > ls.signup_date
21)
22SELECT
23  user_id,
24  signup_date,
25  order_id AS first_order_id,
26  order_date AS first_order_date,
27  CAST(DATE_DIFF('day', signup_date, order_date) AS INTEGER) AS days_to_first_order
28FROM post_signup_orders
29WHERE rn = 1
30  AND DATE_DIFF('day', signup_date, order_date) <= 30
31ORDER BY user_id;

The Product Sense Round

This is where Amazon DS interviews differ from other companies. You'll be given a vague product scenario and asked to define success metrics, identify tradeoffs, and propose an experiment.

Framework: Start with the business goal. Define a primary metric (what you're optimizing) and guardrail metrics (what you can't break). Then propose how you'd measure impact, whether that's an A/B test, quasi-experiment, or pre/post analysis.

Alright, you know what to expect. Now let's talk about how to prepare. This is where most people either over-prepare the wrong things or under-prepare the right things.

Mock Interviews (Do Them, Seriously)

I'm obviously biased since I do interview coaching, but even if you don't hire a coach, do mocks with someone.

Why they matter:

You can know your stuff cold and still bomb under pressure. Mock interviews help you:

• Practice articulating your thoughts clearly
• Get comfortable with the pressure
• Identify gaps in your stories
• Learn to manage time
• Handle unexpected follow-ups

Who to practice with:

• Best: Interview coach who knows Amazon's process
• Good: Friend who works in DS and can push back on your answers
• Okay: Any technical friend who can hold you accountable to time limits
• Not helpful: Your mom who thinks everything you say is perfect

What to practice:

Do at least 3-4 full mocks before your real interview:

• 1-2 focused on behavioral (full loop of LP questions)
• 1-2 focused on technical (SQL + coding + stats)
• 1 full simulation (both types mixed, 5 rounds)

Record yourself if you can. It's painful to watch but super helpful.

One client's experience:

Guy I worked with did zero mocks, felt confident, bombed the interview. His feedback: "Answers were too long and lacked structure."

We did 4 mock sessions. He got used to the pressure, tightened his answers, learned when to stop talking. Second interview at Amazon (different role, 6 months later), he got the offer.

Practice matters.

The Week Before

Final prep tips for the week leading up to your interview:

Monday-Wednesday:

• Do one final mock if possible
• Review your prepared stories one more time
• Brush up on SQL with 10-15 practice problems
• Read through Amazon's Leadership Principles again

Thursday-Friday:

• Light review only, don't cram
• Get sleep
• Review the job description and team info
• Prepare questions to ask your interviewers (yes, have these ready)

Day before:

• Do nothing interview-related after 6pm
• Get a good dinner
• Sleep 8 hours
• Seriously, sleep matters

Day of:

• Test your tech setup 30 minutes early
• Have water nearby
• Take those 5-10 minute breaks between rounds

What not to do:

Don't stay up late cramming SQL. Don't try to memorize new concepts the day before. Don't have 3 cups of coffee right before (nervous energy ≠ good energy).