Instacart Data Scientist Interview Guide

TL;DR the data science interview process at Instacart looks like:

Step 1: Recruiter screen

Step 2: Tech screen 1 - SQL & Product Case

Step 3: Tech screen 2 - Statistics & Experimentation

Step 4: Take-home

Step 5: Onsite

• 5.1 Take-home presentation
• 5.2 Statistics and experimentation deep dive
• 5.3 Product case(s)
• 5.4 Behavioral and career

Instacart is a marketplace made up of:

• Customers who demand goods
• Retailers who provide the supply
• Shoppers who pick and/or deliver orders
• Brands which advertise on Instacart

If data scientists change the search algorithm to improve the product for customers, that directly affects how brands calculate their ad ROI. Let’s say the change shows hyper-personalized products for the end customer, but those products might be harder for shoppers to find in-store so it can take them more time to fulfill one order eating into their earnings which could potentially even lead to reputation risk for the company.

While changes and effects on the user’s side are usually thought through, shopper-side effects commonly crop up at Instacart, as this Prepfully coach who is a senior data scientist at Instacart says:

Instacart data scientists have to constantly think about the potential harms a change might create. You will be asked to reason through that same logic throughout the interview process.

The overall process has five stages.

Note: Instacart tends to hire people for a general-interest opening and then team-matches during the latter half of the interview loop by gauging where your abilities and domain-knowledge would come in handy.

The recruiter call is 15 to 30 minutes. Expect basic questions on your background and a logistics discussion. Recruiters will want to know about your A/B testing experience, and ask you to rate your Python proficiency.

Questions you'll likely be asked:

• How would you rate your Python skills on a scale of 1–10?
• How much experience do you have with A/B testing and experimentation?
• Walk me through your background and what you're looking for in your next role.
• Why Instacart?

These questions help the recruiter frame the conversation for the HM and decide the emphasis for your technical screens. Rate yourself as accurately as you can, since the technical screens and take-home are calibrated on what you say here.

This 45-to-60-minute call tests two different things in the same window.

The SQL section

Monta Shen, Staff DS at Instacart, published a post explaining that SQL coding from scratch is a poor test of the job since it forces simple questions and ignores that Instacart DSs are expected to use AI tools in their actual workflow. 

For the AI-assisted SQL interview, you'll be given a dataset schema and a business-related question. The task will likely be one of these three:

• Translate an insight into a prompt for a SQL query. You have to write the prompt you'd hand an LLM to generate the query. 
• Explain and debug a sample query, often LLM-generated: what does it do, is it correct, how would you fix it? 
• Identify ways to make a SQL query more efficient. 

Interviewers want to know if you can break down a vague task into something AI can execute, and if you can verify and improve the output you get.

Some questions candidates have been asked in the past: 

• We want to run a Thanksgiving coupon experiment where users are randomly assigned to an incentive (a $10-off coupon) or no-incentive group. How would you write the query to compute order uplift within the coupon window?
• You're given an Experiments table (user_id, experiment name, group assignment, assignment timestamp) and an Orders table (user_id, order_id, order timestamp, order amount). Users in the incentive group received a $10-off coupon expiring on a given date. How would you analyze this experiment? 
• Given orders and products tables, find the top N products by reorder rate within a rolling 30-day window.

The product case section

The product case section focuses on your product sense.

A lot of the questions will be “Do you launch or not?” decisions for new features.

You could also be asked to design ways to measure and improve the product. If you’re developing metrics, try to keep your metrics simple and measurable through SQL.

Some recently reported questions:

• What metric would you use to measure whether this change improved the customer experience?
• How would you measure the success of this notification? What counter-metrics would you use?
• Our activations are down by 10%. Figure out why.

According to this senior data scientist at Instacart, a fully fleshed out answer should include other factors.

The round is 45 minutes long, and will be run by a Senior DS. 

Instacart interviewers want to see how solid your statistics fundamentals are, and if you can design experiments that account for the four-sided marketplace's problems.

The first half will test your statistical foundations so expect questions on hypothesis testing, Type I and II errors, power analysis, sample size calculation, or even when to use which test. 

Note: Some interviewers will ask you to design an experiment and then will ask follow-up statistical questions within that context.

These stats questions have been reported by Instacart DS candidates:

• Let’s say you're measuring the differences between two groups and an assumption you made is invalid. What would be your testing approach?
• What will you do if you can’t randomize users?
• Besides increasing the sample size, what are some other ways of reducing the variance?

In the second half you’ll probably be given a new feature or change the team is making. You have to walk them through how you’d design the experiment to validate the feature.

For design questions, expect something like:

• We are launching 15-minute deliveries in one area, how would you test the results? 
• Can you design an experiment to see if showing the approx. delivery time to customers increases conversion?
• How do you measure the impact of a price change on customer retention?

Instacart's shoppers are divided into geographic zones. 

Since shoppers (in both the treatment & control groups) are picking up orders from the same customer pool, carrying out a standard user-level A/B test could result in a biased estimate. 

There are two alternatives Instacart uses:

1. Cluster randomization: You assign treatment and control by geographic zone. Shoppers & customers in Region 1 get one treatment, and the ones in Region 2 get another. The tradeoff is wider confidence intervals and fewer effective units.
2. Switchback designs: You can also propose alternating between treatment and control over time within the same region. So Region 1 will have treatment on Monday, control on Tuesday, treatment on Wednesday, control on Thursday and so on. Switchbacks are great if the effects of the treatment decay fast (like surge pricing) but could run into problems if the effects persist (like a referral feature).

Since there’s a physical supply chain involved, we recommend you take some time to brush up on concepts like CUPED, IPW, MAB, synthetic controls, delta method, HTE before your interview.

You’ll be given a take-home assignment a few days before the onsite. It’ll consist of experimental data (real or closely modeled on Instacart's) and you’ll be asked to analyze it and make a recommendation. 

The work itself will take around five hours. It’s mostly EDA (exploratory data analysis) plus statistical analysis of an experiment (in Python or R) along with a small write-up. 

The take-home tests your judgment as much as your analysis. Specifically:

• Did you define and pick the right metric for what you’re trying to measure?
• Did you handle the analysis correctly? 
• Is your recommendation clear and actionable? 
• How did you communicate the results?

Past candidates have noted you should spend extra time on your findings and your recommendation. When you’re writing out your recommendation, also explain the “why” in detail.

The onsite stage should take around four hours across four rounds.

The take-home from the previous stage is also the centerpiece of your first onsite round. You walk the interviewers through your analysis. Be ready to defend any decision you made and the tradeoffs you considered. 

You can be asked questions like:

• How did you validate the randomization?
• Can you explain why you picked this primary metric and these counter-metrics? Let’s say that data isn’t available for some reason, what metrics would you pick then?
• Can you tell me what kind of internal or external factors you will be looking at?
• Can you summarize your findings for all three questions today and put it into an elevator pitch?

We recommend you keep the walkthrough a little tight and spend more time on the methodology and your findings. 

Instacart interviewers want to see conviction in your analysis and what you’ve based your recommendation on. This round is also a test of your product and business sense. 

This round will be about an hour long. Interviewers will ask you to design an experiment or two and then ask a ton of follow-up questions to see how rigorously you can think it through.

These questions have been asked of Instacart DS candidates:

• Let’s say we want to introduce a marketplace fee for the end customers to incentivize shoppers. Come up with a framework to evaluate whether this will be successful or not. Follow-up: How would this fee affect the retailers?
• Can you tell me about the Bonferroni approach and the limitations it might have?
• You're tracking five guardrail metrics in the same experiment. How do you handle multiple comparisons?

Before this round, take some time to sketch guardrails for Instacart features, and you’ll find it easier to talk about these decisions at the rigor Instacart expects.

Depending on the question, you might have to propose tracking multiple guardrails at once. If you do that, keep in mind the multiple comparisons problem that might crop up.

A simple solution is Bonferroni, which has been brought up on Instacart's blog. You should also bring up Benjamini-Hochberg to control the false discovery rate.

There may be one or two product rounds, each an hour long. The questions might be somewhat more abstract than the tech screen product questions. A super common case: you’re given a dataset and asked to design a feature from it. 

These questions have been reported by Instacart DS candidates:

• If I give you a dataset which shows how our customers browse the Shopping Hub, what product improvements would you prioritize, and how would you measure them?
• How would you use data to optimize Instacart's delivery route system?
• We want to improve shopper retention. How would you come up with features to do that?
• We want to decide if we should expand same-day delivery to another city. What sort of data would you need to make the decision?

Basically, Instacart interviewers want to see how well-developed your product sense is and how you would collaborate with the product team.

This final behavioral round will be a walkthrough of your work history.

Your interviewer will want to know who you are at a high level and what drives you. They’ll also ask about your long-term career goals and how Instacart fits into them.

Instacart interviewers have asked data science prospects these questions:

• Can you describe a situation where you had to say no? 
• Can you tell me about a time when you spotted a business opportunity?
• Can you tell me about a situation where you had to move fast, and had to rely less on data and more on your intuition to make a call?
• Can you tell me about a time you made the wrong call? What happened? 

We suggest you prepare stories that feature decisions you made independently and where you also moved fast. If you can tie the technical result to customer or partner impact, that’s even better. 

Important note: Some senior DS candidates have reported a ‘Bar Raiser’-like round. So there will usually be a cross-functional interviewer or a senior DS from another team who will appear somewhere in the process. You could also have an independent Bar Raiser round, they will cover similar questions to what we have outlined in this guide. But, there will be a greater focus on cultural fit and they do have veto power.

That’s why we suggest you prepare for cultural fit or behavioral questions to be asked in any single round.

Once the offer comes, you’ll have a conversation about team placement. You might be able to make a case for a specific team depending on your interests.

Total comp for data scientists at Instacart ranges from $248K ($165K base) at L4 to $414K ($225K base) at L6, according to data from Levels.fyi.

RSUs vest on either a two (equal 50% split) or four year (equal 25% split) schedule.