Amazon Data Scientist Interview Guide

Amazon’s hiring philosophy is built around the idea that every new hire should raise the bar for the people around them, which is why even for data scientists, the interview process focuses on your technical skills and judgment simultaneously.

The role is heavy on analytics, SQL, metrics and experimentation. The role which involves heavier ML research is called “Applied Scientist” at Amazon. The interview bar for ASs means you must clear the Research Scientist (another related role) bar and have SDE-level (typically a level below the target AS level) coding skills. So if you’re interviewing for AS L5, you’ll have to clear the bar for RS L5 and SDE L4.

Amazon data scientists meanwhile spend the bulk of their time experimenting. An ideal candidate will be someone who can design experiments 0 to 100. That means they’re looking for people who have the product intuition to start with the right hypothesis and the statistical chops to understand all the little nuances of experiment design.

TL;DR data science interview loops at Amazon look like:

Step 1: Recruiter screen

Step 2: Technical screen

Step 3: Onsite loop - 5 x 45-60 mins

3.1 Data science breadth interview to test your data science knowledge.

3.2 Data science application round or case study to test if you can scope and design technical solutions around business constraints.

3.3 Data manipulation and SQL interview for more technical depth.

3.4 Data science depth interview to see how you explain a past project.

3.5 Behavioral round(s) tied to Amazon’s LPs.

There will also be a Bar Raiser interview somewhere within the other rounds (usually behavioral) to make sure new hires are better than the average Amazon employee.

The Amazon DS interview process is only somewhat consistent across candidates. While the rounds will be similar, the topics and questions can change entirely depending on the org & team which is interviewing you.

For instance, Alexa & Search DS teams work on GenAI, so you should expect questions like “how would you fine-tune this transformer model?” or “how would you evaluate LLM outputs?” Other Amazon orgs like their Forecasting & Labs team will focus on forecasting (duh) but also issues like supply chain management.  

We also want to point out that down-leveling at Amazon is quite common. And promotions at Amazon can be a little slow, so going from L4 to L5 can take 18-36 months even if you’re a strong performer. 

And unless you have a competing offer, Amazon is not open to level negotiations. Some people even suggest declining a leveled-down offer and interviewing again in 6 months.

That’s why you should set the tone and demonstrate level-appropriate capabilities in your answers from the get-go. If you want to understand what level you’d be a competitive candidate for, an Amazon Data Scientist on Prepfully can give you a candid read.

For the first screen expect straightforward questions, and recruiters overall want to see if your work background and skills fit what the role demands. But keep in mind because Amazon highly values leadership ability and cultural fit, their recruiters are already looking for evidence for their leadership principles. 

Even if you’re making small talk, recruiters will be listening to understand how you think, how you approach problems, how you work with people, how you’ve developed your career etc. 

Some questions Prepfully candidates have been asked recently:

• What do you know about how Amazon does data science?
• Tell me about your favorite past projects.
• Why Amazon? Why the [X] team?

An Amazon employee on Blind suggests candidates should reach out to recruiters and ask for some sample questions for the leadership principles they'll be gauged upon.

Some teams will have a SQL + coding online assessment, while others will jump straight into a technical screen over call. Expect the call to be long, around 60-90 mins.

Both the OA and technical call will feature similar problems. The main one will be one SQL problem on e-commerce data (15-20 mins), followed by ML conceptual questions and a test of your statistical basics (15-20 mins). And finally a few behavioral questions based on Amazon’s LPs (10-15 mins).

Some of the candidates we’ve coached at Prepfully have also noted small mismatches between recruiter expectations and what comes up in the interview. So, prepare for DS and ML fundamentals along with SQL and coding no matter what you hear from the recruiter.

Amazon DS candidates have been asked these questions:

• Given a set of user data, write a SQL query to explain the month-to-month user retention rate.
• Given a retail data set, find the best-selling item for each month, and then find the biggest invoice involving that item. Output the month, the description of the item, along with the amount paid.
• There are 4 red balls and 2 blue balls; if you pick 2 balls without replacement, what's the probability of them being different colors?
• How do you inspect missing data, and why is that important?

Amazon’s data science onsite is five back-to-back interviews. Each will take 45-60 mins, and more importantly, every interviewer gets pre-assigned 2-3 specific LPs to evaluate alongside their technical focus area. So even if you’re having a super technical round, take into account the interviewer is racking up a non-technical score as well, and bring in the non-technical and human details.

The one thing we want to stress before your onsite is that for Amazon, Leadership Principles (LPs) carry as much weight as the technical skills. 

For every interview round in the onsite, every interviewer will also have a couple of preset LPs they have to evaluate you on. Our Amazon DS coaches say it’s usually the biggest piece of the overall process, and if a candidate is strong on LPs but not so strong technically they could still be accepted.

There are 16 LPs but you don’t need 16 detailed stories. Your recruiter should tell you which LPs to prepare for and for those, you should have extremely detailed stories written down that would cover follow-up questions as well. Now, while each story should hit the primary LP, they can also touch upon and demonstrate other LPs.

A senior DS manager who coaches for us at Prepfully, and used to be a Bar Raiser at Amazon says:

For the remaining LPs, also prepare some smaller stories you can talk about so you avoid repeating stories and keep everything fresh. You can prepare different stories from the same project that focus on different aspects and situations.

The science breadth round is a wide sweep of your DS & ML basics. There’s no single focus here and likely no code involved as well.  

Some recently reported questions by Amazon DS candidates for this round: 

• How would you differentiate likelihood and probability?
• How do you calibrate an ML model? 
• Design an A/B test for a green vs. yellow buy button. 
• How do you control for FDR / FWER if your team is doing 50 test variations?
• How do you decide when to use parametric methods vs non-parametric methods?

Definitions are super important. However, for each piece of theory you reference in an interview, it’s really important to drive home your familiarity with how it's applied in a production or live environment. Your interviewers want to see you comfortable with the limitations of different approaches in different circumstances, and are actively filtering out candidates who sound like they’ve just memorized the theory.

Please note: For some candidates this round and the science breadth round are combined together.

If you’re interviewing for the GenAI team, this round will be focused on GenAI applications.

You’ll be given a business problem from Amazon's world and then be asked to talk the interviewers through your process to approach it end-to-end. 

DS candidates at Amazon have been asked these case questions:

• If I give you a customer’s purchase history how would you predict customer churn?
• We want to improve the recommendations Alexa makes. Can you design an experiment to do that?
• Tell me an approach to identify potential high-conversion customers.
• If the team is releasing a new GenAI model in prod, what kind of evals and guardrails would you put in place to make sure it adheres to policy?
• How do you design an A/B test so it will detect model regression in downstream business KPIs?

Spend some time on Amazon digging into their case studies, it’ll really help you approach these problems confidently. 

Let’s say you are told Amazon is opening a new store in XYZ country. You might then be asked related questions like “where should the fulfillment center be located, what metrics will you consider, what model will you use and how will you get the data?”.

The interviewers are watching whether and how you scope first. Think about what kind of problem this is, what other issues might it point to? You might say let’s use current metrics from existing fulfilment centres that'll help decide where to build the centre. You can ask the interviewer what these are. Spend extra time exploring available metrics and think about how to use them in your model. 

You might ask questions like: what is the primary business goal for this expansion? Is this the first store in the country? Does Amazon have any historic data available for this country? What operational or regulatory constraints exist in this country?

Throughout this process, your interviewer will also ask you questions based around Amazon’s LPs such as “how do you balance speed versus accuracy?”

Think of the depth round like a PhD defense; this entire interview revolves around one of your projects. You’ll either be asked to choose one or the interviewer will pick one for you. 

You’ll then be asked to walk through the problem, your approach, the modeling decisions you made, what happened in the end, what you’d do differently with your current knowledge, and what you learned from it. 

For every choice you made, even if it’s a small one, expect the interviewer to ask “why”?

Recently asked questions reported by Amazon DS candidates in the science depth round:

• Why did you choose this model over alternatives? What would you do differently? 
• How did you validate the model? What was the business impact?
• Why did you choose these evaluation metrics? How would this choice change if some specific business constraints were different? 

To prepare for this round, pick one project, and start with a short summary. Then ask a friend to spend 30 minutes asking you "why did you do X instead of Y?", “what tradeoffs did you make and why were those acceptable to you and to the business?” Basically, for every decision you made, have a clear answer that brings in the input and the desired output along with the constraints.

You’ll be coding on simulated Amazon e-commerce tables: orders, products, customers, categories. Some teams will also test you on Python/pandas. 

SQL interviews at Amazon feature Redshift-style syntax, so it’s possible some functions do slightly different things. We suggest you go over their SQL reference docs before your onsite.

Amazon interviewers have recently asked these questions to DS candidates:

• If speed is most important, would you use joins or subqueries?
• Can you write a query to find MoM revenue change? Also find the top-3 selling products by category and then calculate customer retention cohorts.

Make sure you practice on e-commerce data, and speed is something you’ll be evaluated on, so spend some time solving SQL problems as fast as you can. Mostly focus on common SQL problems that are prevalent in data science interviews like window functions & multi-step CTEs. 

The behavioral round is usually with the HM and/or Bar Raiser. This round is all about Amazon’s LPs. Interviewers usually have pre-assigned LPs, so you can ask your recruiter if there’s any specific ones you’ll be evaluated upon.

For this round, your answers have to be in STAR format. 

As we mentioned before, there are 16 Amazon Leadership Principles and every Amazon employee says they’re super important. The most common LPs that Amazon DS are asked about are these 6: Customer Obsession, Dive Deep, Deliver Results, Have Backbone; Disagree and Commit, Bias for Action, and Are Right, A Lot.

These questions have been asked to Amazon DS candidates: 

• Tell me about a time you made something much simpler for customers.
• If you were in charge, how would you improve the wishlists feature on Amazon?
• It sounds like you're more senior than all your co-workers. When they make mistakes, how do you help them?
• If you owned marketing for our kid’s tablets, what new things would you look into?
• That sounds like a very stressful project timeline. How did you maintain morale on your team?
• How do you manage your top performers differently?

Your behavioral round is likely with someone less technical than you so focus on the business and human factors you had to deal with.

Amazon Bar Raisers usually appear in the behavioural round (though for some candidates they might show up in the coding or case study interviews as well). 

These Bar Raisers are interviewers from a different team, and hold veto power over the entire hiring decision. 

Ex-Amazon Bar Raiser, Carlos Arguelles has 800+ interviews under his belt and described it this way: "You have veto power over all hiring decisions. A hire cannot happen without your thumbs up." But he also noted: "Using the veto highlighted a failure of leadership. I never once did." The veto power is real but rare. In actuality, the Bar Raiser works with the HM and other interviewers for a final decision. 

We’ve also compiled a guide on how to tackle Amazon Bar Raisers.

We suggest you write down 5-6 of your strongest projects or stories, and you’ll naturally be able to map different aspects of your stories to the LPs.

All interviewers will submit independent written feedback before the final debrief. To keep things unbiased, they won’t see each other's evaluations until they've submitted their own. At the debrief, the Bar Raiser drives the discussion toward consensus.

Leveling can also shift here. L5 candidates sometimes receive L4 offers. Usually, you will not be able to negotiate your level.

Down-leveling happens and the decision to accept or walk away is not obvious. 

If you want to think it through with someone who has been on the other side of that conversation, book an advice session with an Amazon data scientist.

Total compensation for Amazon DS ranges from approximately $140K at DS I (L4) to $900K at Senior Principal DS (L8), according to data from Levels.fyi. 

Amazon RSUs vest on a 5/15/40/40 schedule over four years.

There’s greater room for comp negotiations at higher levels, at lower ones you can negotiate for a higher base but Amazon might offset it by dropping your RSUs. If you have other offers, Amazon might match or beat them.