DoorDash Data Scientist Interview Questions

DoorDash·Data Scientist·Technical Phone Screen

Jul 2026

DoorDash data scientist interview with a product analytics case built around a geo-targeted feature launch. The core of it was interpreting a retention chart under data constraints, which sounds manageable until you're actually in it.

You're evaluating whether to launch a new geo-targeted feature with limited traffic data. Walk through how you'd build the business case: what data you'd need, which metrics matter, and how you'd define success.
Looking at a retention curve comparing test vs. control groups, what hypotheses can you pull from it and what would you do next?

“I started with the metrics layer and worked backward, which felt right in the moment but I think I skipped over some data availability assumptions too quickly.”

View Post

DoorDash·Data Scientist·Technical Phone Screen

Jul 2026

DoorDash data science interview with a meaty marketplace diagnosis case that turned into a full experiment design problem. The scope was broader than I expected and the second half caught me pretty flat-footed.

Completed orders in the Los Angeles market dropped meaningfully. Walk through how you'd investigate this end to end.
The investigation points to merchants being online fewer minutes per day. Design an experiment to test an incentive program meant to fix that.

“I started with the funnel breakdown which felt right, app sessions down to completed orders, but I kept second-guessing whether to lead with supply or demand.”

View Post

DoorDash·Data Scientist·Onsite - Product Sense / Strategy

Jul 2026

DoorDash data scientist interview with a product strategy case built around the bike-delivery program for dashers. Pretty open-ended, which I wasn't fully prepared for.

If DoorDash were to roll out a bike-delivery program for dashers, what operational or strategic challenges would you anticipate, and how would you approach solving them?

“I started with supply balance which felt safe, then kind of spiraled into weather and safety without a clear thread connecting them.”

View Post

DoorDash·Data Scientist·Technical Phone Screen

Jul 2026

DoorDash data scientist interview with a SQL-heavy technical screen focused on delivery metrics. The question was applied and realistic, which I appreciated, but the follow-up tripped me up a bit.

Given an Orders table with expected and actual delivery dates, write a SQL query to find the percentage of orders delivered late in the last 30 days. Follow-up: which three customers had the most late deliveries in that same window?

“The main query was fine.”

View Post

DoorDash·Data Scientist·Technical Phone Screen

Jul 2026

DoorDash data scientist interview with a SQL-heavy technical screen. One question but it packed in a lot, window functions across multiple partitions in a single query. Felt like a reasonable test of whether you actually know SQL or just know the syntax.

Given an orders table with columns for order ID, user ID, order date, city, and order value, write a single SQL query that returns each order alongside the previous order value for that same user (NULL if it's their first), the rank of that order's value within its city from highest to lowest, and the percent rank of the order value within its city.

“Three window functions in one query.”

View Post

DoorDash·Data Scientist·Technical Phone Screen

Jul 2026

DoorDash data science interview with a meaty experiment design question centered on a new bike delivery feature. The whole session was basically one long case and they pushed hard on every layer of it.

Design an experiment to evaluate whether launching a bike delivery option across a subset of cities is beneficial. Walk through your goal, metrics, randomization strategy, experiment design, potential biases, analysis approach, and how you'd size the test.

“This is a beast of a question.”

View Post

DoorDash·Data Scientist·Technical Phone Screen

Jul 2026

DoorDash data science case focused entirely on a marketplace expansion problem: should bike dashers be allowed in a city currently dominated by cars and scooters. It was a deep, multi-part question that touched on everything from experiment design to network effects, and I left feeling like I'd only scratched the surface on a few of the sub-parts.

What historical marketplace data would you look at to determine whether a city or specific zones within it are good candidates for bike dashers?
What primary, secondary, and guardrail metrics would you track across consumer, merchant, dasher, and unit economics dimensions when evaluating bike dashers?
How would you design an A/B test or marketplace experiment to measure the causal impact of enabling bike dashers in a city?
Bike dashers affect dispatch and matching for all orders in a zone, not just their own. How do you account for these spillover effects in your experiment design and analysis?
What experimental outcomes would lead you to fully launch, partially roll out, or pull back the bike dasher program?

“This part felt manageable.”

View Post

DoorDash·Data Scientist·Technical Phone Screen

Jul 2026

DoorDash data science interview focused entirely on marketplace experimentation across three fairly meaty scenarios: auto-applying credits at checkout, duplicate restaurant listings in search, and a restaurant promotions product. No coding, just product analytics and experiment design the whole way through.

How would you design an experiment to test auto-applying customer credits at checkout by default, and what should the primary metric be?
What tradeoffs exist between checkout conversion, credit burn, platform margin, and long-term retention when auto-applying credits?
When the top organic result and the top sponsored result in search are the same restaurant, is showing both listings good or bad for the marketplace?
If you decide to remove duplicate restaurant listings from search results, how would you test that change and what metrics would cover user experience, ad performance, and overall marketplace value?
How would removing a duplicate listing affect lower-ranked restaurants, and how do you account for that in your experiment design?
How would you increase restaurant adoption of the existing fixed-discount promotion feature?
If restaurants can customize their promotion format rather than using a fixed template, how would you design and test that product?
How would you compare the new customizable promotion product against the old fixed-template product, and how do you separate incremental lift from cannibalization or adverse selection?

“This one tripped me up more than I expected.”

View Post

DoorDash·Data Scientist·Technical Phone Screen

Jun 2026

A data science case interview at DoorDash focused entirely on one deep, open-ended question about diagnosing a drop in completed delivery orders in LA. It was the kind of question that sounds manageable until you're twenty minutes in and realize you've only covered half of what they wanted.

Completed delivery orders in Los Angeles have dropped noticeably over the past two weeks compared to historical trends and similar cities. Walk me through how you'd investigate this end-to-end, from confirming the drop is real all the way to proposing an experiment to test a fix.

“This one ate up the whole session.”

View Post

DoorDash·Data Scientist·Take-home Assignment

Jun 2026

DoorDash data science take-home that was way heavier on SQL than I expected. Four problems, all PostgreSQL, all requiring CTEs and window functions, and the schema was deceptively simple but the logic got gnarly fast.

Given an exposures table and an orders table, compute the percentage of orders in a 7-day window that satisfy a compound filter (biker courier, cold temp category, subtotal at least 2000 cents) under two denominator definitions: one counting only orders tied to treatment-exposed units on that day, and one counting all orders regardless of exposure status.
Define a 'power day' for each unit-date as: at least 2 biker-cold orders in the trailing 7 days AND average subtotal on those orders is at least 2000 cents. What percentage of unit-days are power days under the same two denominator conventions?
Build a complete unit-day grid (including days with zero orders) for a 7-day window. For each unit and date, compute daily order count, cumulative orders, a 7-day rolling sum, day-over-day percent change, and a non-overlapping 7-day percent change using specific window frame definitions.
For each unit, identify dates where a boolean flag (whether a qualifying high-value biker-cold order exists that day) flips value. For each flip, return the unit, the date, the direction of change, the difference in 7-day rolling order counts before and after the flip, and the percent rank of that delta across all flip events.

“The two-denominator thing tripped me up more than I want to admit.”

View Post