Amazon Data Scientist Interview Guide (Updated for 2024)

The Amazon data scientist interviews include recruiter and tech screens, followed by an on-site interview. The data science interview questions asked by Amazon focus heavily on machine learning and algorithms, and to a lesser extent, SQL and Python. However, in addition to technical skills, candidates are also assessed on their critical thinking, problem-solving skills, and adherence to Amazon’s 14 leadership principles.

Data scientists at Amazon perform a variety of functions, depending on the team they work with. An Amazon data scientist might be tasked with:

• Designing, developing, and deploying data-driven models and analytics solutions
• Developing accurate predictive models
• Developing data pipelines
• Deploying automated software solutions to assist in forecasting
• Researching, designing, and improving models with business impact

Qualifications

Amazon only hires experienced and highly qualified data professionals, and the company has some of the most rigorous standards in the industry. General requirements for Amazon data science roles include:

• Masters’ degree in any quantitative field such as Statistics, Quantitative Finance, Economics, Computer Science, Mathematics, Physics, Computational Biology, Operational Research) or equivalent practical experience.
• 2+ years’ work experience (4+ years for Senior Data Scientist roles) in an analytical role involving machine learning techniques, data extraction, analysis, and communication.
• Proficiency (4+ years’ experience for Senior Data Scientist) in the use of statistical software packages and functional programming languages such as R, Stata, Matlab, Python, SQL, C++, or Java.
• Experience in designing and implementing machine learning algorithms tailored to specific business needs and tested on large datasets.
• Experience in data mining and using databases in a business environment with large-scale, complex datasets.
• Excellent verbal and written communication skills with the ability to effectively advocate technical solutions to research scientists, engineering teams, and business audiences.

Preferred Qualifications (Data Scientist)

Many Amazon data science roles have additional qualifications including:

• Ph.D. in a quantitative field (Computer Science, Mathematics, Machine Learning, AI, Statistics, or equivalent)
• Sound business and project management skills
• Experience in building complex data visualization

What to Expect in Amazon Interviews

The interview process at Amazon is similar to other tech companies. The basic difference here is that there are no take-home challenges.

The interview process usually starts with an initial phone screen by a recruiter or a team manager. After this, you will face a technical phone screening, which includes a coding challenge. Finally, you will be invited for an on-site interview which is usually done in five stages, with an informal interview over a lunch break.

1. Recruiter Screen

The initial phone call comes after you have submitted your application and you are contacted by the HR or a recruiter. This is a resume-based phone interview that normally lasts for an hour. The conversion is focused mainly on your skills, previous experience, and you may be introduced to what the position entails.

Example questions include:

• Tell me about yourself.
• Tell me a time when you did not agree with your advisor.
• Tell me about a time when you had two deadlines at the same time. How did you manage the situation?
• Tell me about a time that you faced an obstacle just before a deadline. What did you do?

2. Technical Screen

The technical screening involves coding, statistics, and machine learning. At a minimum, you can expect two coding challenges, as well as SQL and machine learning questions.

Amazon technical screens are done via a service link called “CollabEdit,” which allows interviewers to see your work in real-time. Follow with the interviewer, and provide reasoning for your answers. There is also a section on “approach”, detailing how you got to the solution and why you use the steps you used.

Example questions during the technical screen:

• Write a query to find the total number of transactions.
• We have two models, one with 85% accuracy, one 82%. Which one do you pick?
• What is p-value, and how can we use it?
• Write a Python function to find the term frequency of text within a dataset.
• Which machine learning algorithms/techniques are you familiar with?
• Write a query to identify the manager with the biggest team size.

3. On-Site Interview

During the on-site interview, you will be asked about your past/current projects, ML learning skills, predictive modeling, exploratory analysis, and some coding questions.

This stage consists of five 45-minute back-to-back interviews. They’re conducted one-on-one or with a manager and a junior data scientist. The process includes:

• A behavioral interview to assess culture-fit
• A technical interview involving data analysis
• An SQL-based interview with a data scientist
• Another data analysis and design interview
• A machine learning interview that involves lots of coding

For Amazon data science interviews, practice a lot of machine learning and algorithms questions, as these subjects are covered in depth. In particular, the most frequently asked subjects are:

• Machine learning
• Algorithms
• Python
• SQL
• Behavioral Questions

The most common types of machine learning questions asked in Amazon interviews are system design and applied model questions. Both types ask you to walk through a data model or the architecture for machine learning. You can also expect definitions questions, as well as discussions about different types of machine learning models.

1. What is the difference between XGboost and random forest?

Random forest is a bagging algorithm, and in using it, you have several base learners or decision trees, which are generated in parallel and form the base learners of the bagging technique.

However, in boosting, the trees are built sequentially such that each subsequent tree aims to reduce the errors of the previous tree. Each tree learns from its predecessors and updates the residual errors. Hence, the tree that grows next in the sequence will learn from an updated version of the residuals.

2. What is variance in a model?

Variance is the measure of how much the prediction would vary if the model was trained on a different dataset, drawn from the same population. Can be also thought of as the “flexibility” of the model.

3. Is a decision tree model best for predicting if a borrower will pay back a personal loan? How would you evaluate performance of the model?

A few questions to consider are: How would you evaluate performance of the model? And how would you compare a decision tree to other models? See a full solution in this YouTube mock interview:

4. What would you do if 20% of the 100,000 sold listings are missing square footage data. You want to predict price.

This is a classic modeling interview question. Data cleanliness is a well-known issue within most datasets when building models. Real-life data is messy, missing, and almost always needs to be wrangled with.

The key to answering this interview question is to probe and ask questions to learn more about the specific context. For example, we should clarify if there are any other features missing data in the listings. If we’re only missing data within the square footage data column, we can build models of different sizes of training data.

5. How would you design the YouTube video recommendation engine?

Machine learning system design questions are common in Amazon interviews. These questions are designed to assess how you think through a design scenario. See a step-by-step solution to this video:

In Amazon interviews, algorithm questions are designed to assess your understanding of algorithms. Although in some cases there may be coding involved, the key reason these questions are asked are to determine if you:

• Know how an algorithm works
• Can explain the mathematics behind common algorithms

6. What is gradient descent?

Gradient descent is a method of minimizing the cost function. The form of the cost function will depend on the type of supervised model. When optimizing our cost function, we compute the gradient to find the direction of steepest ascent. To find the minimum, we need to continuously update our Beta, proportional to the steps of the steepest gradient.

7. What are the assumptions of linear regression?

With a question that asks the assumptions of linear regression, know that there are several assumptions, and that they’re baked into the dataset and how the model is built. The first assumption is that there is a linear relationship between the features and the response variable, otherwise known as the value you’re trying to predict.

8. How do you detect and handle correlation between variables in linear regression?

Multicollinearity in a regression model describes a situation in which two or more independent variables are highly correlated with one another. There are many indicators you can use to detect multicollinearity. For example, when standard errors are orders of magnitude higher than coefficients, that’s usually a strong indicator.

Amazon tends to test Python more rigorously than other tech companies. In particular, Amazon Python questions assess your ability to write clean Python code, and these questions cover subjects like statistics and distribution, data structures and string parsing.

9. Write a function to generate N samples from a normal distribution and plot the histogram. You may omit the plot to test your code.

This is a relatively simple problem because we have to set up our distribution and then generate n samples from it which are then plotted. In this question, we make use of the SciPy library which is a library made for scientific computing.

10. Write a function shortest_transformation to find the length of shortest transformation sequence from begin_word to end_word through the elements of word_list.

Generally, shortest path algorithms require the solution to recursively try every possible matching path from the start to the end.

• Every word in word_list is of the same length.
• The max difference between 2 words in the path is only one letter change.
• The shortest path might require us to go back and forth in the list, rather than just go forward.
• We can’t choose the same word twice in the path.
• There might be a shorter path further along with the list.

11. Write a function to determine the TF (term_frequency) values for each term of this document.

Here’s a quick overview of how to solve this question: First, split the sentences into words. Then, use a dictionary to hold the count for each word. Then, divide each word count by the total number of words and return the result.

You can expect an Amazon SQL question on the technical screen, and one or two of the on-site interviews will focus heavily on SQL and data analysis. In general, Amazon SQL questions tend to focus on customer metrics and e-commerce cases.

12. Write a query to output a table that includes every product name a user has ever purchased.

With this question, you’re provided a table that contains data about products that a user purchased. Products are divided into categories. The column id is the primary key of table products and represents the order in which the products are purchased.

13. Write a query to get the distribution of the number of conversations created by each user by day in the year 2020.

In this question, you’re given a table that represents the total number of messages sent between two users by date on messenger.

• What are some insights that could be derived from this table?
• What do you think the distribution of the number of conversations created by each user per day looks like?

See a video solution for this question:

14. Given a users table, write a query to get the cumulative number of new users added by day, with the total reset every month.

This question first seems like it could be solved by just running a COUNT(*) and grouping by date. Or maybe it’s just a regular cumulative distribution function? But we have to notice that we are actually grouping by a specific interval of month and date. And that when the next month comes around, we want to the reset the count of the number of users.

15. Write a query to get the number of customers that were upsold after their first purchase

We’re given a table of product purchases. Each row in the table represents an individual user product purchase.

Write a query to get the number of customers that were upsold, or in other words, the number of users who bought additional products after their first purchase.

Hint: An upsell is determined by multiple days by the same user. Therefore we have to group by both the date field and the user_id to get each transaction broken out by day and user.

Behavioral questions in Amazon interviews focus heavily on the Leadership Principles. Every question is an opportunity to show how your experiences align with the principles.

Some topics you should cover include the impact of your work, how your work has benefited customers, risks you’ve taken, and your ability to innovate simply.

16. Give an example of an analysis that you did that drove business impact.

“Deliver results” is an Amazon leadership principle. A question lets you provide concrete examples of the results you delivered. You can talk about an increase in user engagement, improved marketing performance, an operations efficiency, etc. Remember to structure your answer. The STAR format works well. Highlight the problem. Talk about how you approached the problem and your plan of action. Then, cover the execution and results you delivered.

17. How do you make technical topics accessible to non-technical audiences?

To answer this question, you might talk about developing visualizations that were easily accessible, or how you created a presentation that framed your project in easily digestible parts. A question like this assesses your ability to collaborate and communicate effectively.

18. Tell me about a data project you have worked on where you encountered a challenging problem. How did you respond?

This question is a chance to talk through your approach to a challenging situation. A few Amazon principles you might consider incorporating include: Learn and Be Curious, Invent and Simplify and Ownership.

19. How do you address colleagues that don’t agree with your approach?

When interviewers ask this question, they are looking to see that you can negotiate effectively with your coworkers. Like most behavioral questions, use the STAR method. State the business situation and the task you need to complete.

State the objections your colleague had to your action. Do not try to downplay the objections or write them off as “stupid”, you will appear arrogant and inflexible.

Additional Amazon Data Science Interview Questions

Amazon has one of the most rigorous interview processes in data science. Use these Amazon interview tips to stand out:

• Focus on technical skills - Amazon’s technical interviews are multi-step and test skills rigorously. To prepare, focus on solving algorithm questions, optimizing queries, and memorize how the most common machine learning algorithms work.
• Memorize Amazon principles - You will be assessed on Amazon’s core leadership principles. Memorize all 14 of them, and use past projects and work experiences to illustrate these principles.
  
• Familiarize yourself with Amazon products - Expect case questions related to Amazon’s various projects. In many cases, you’ll be asked to apply machine learning to a business scenario. See an example solution for a Amazon case question.
• Get comfortable with whiteboarding - Practice coding on a whiteboard. Whiteboarding is common in Amazon on-site interviews for data science roles.

See the latest salary estimates for data scientists at Amazon. View salary ranges by seniority and location:

Explore the latest Amazon data science jobs from the Interview Query job board: