DoorDash Data Scientist Interview Loop
DoorDash's loop is short and sharp. After a recruiter questionnaire, often by email rather than a call, there is a one-hour technical screen split in two. Thirty minutes of SQL on a shared schema you cannot run, then thirty minutes of a case about the marketplace. Pass that and the final loop is two or three case rounds, a conversation with a cross-functional partner, and the hiring manager. Rejections after the screen often arrive the same day.
What decides it
Structure under time pressure. The case half of the screen rewards a clear frame stated out loud. The SQL half rewards writing the query right without running it.
Practice by round
SQL on delivery data at speed: average delivery time per restaurant, late delivery rate by city, courier tip rate, delivered orders by cuisine, early repeat customers. The screen is three or four questions in thirty minutes with no way to test your query. Clean syntax first time is the skill.
Harder SQL like busiest restaurants in a month: window functions, running totals and retention over a seven-day window are the usual shapes.
| Question | Difficulty |
|---|---|
| Avg Earnings per Weekday and Hour | Medium |
| Avg Order Cost During Rush Hours | Medium |
| Daily Top Merchants | Medium |
| First Time Orders | Medium |
| Extremely Late Delivery | Medium |
| First Ever Ratings | Medium |
| Top 2 Restaurants of 2022 | Medium |
| Average On-Time Order Value | Medium |
| Customers with Large Orders | Medium |
| Workers Who Are Also Managers | Medium |
| Busiest Restaurants in June 2025 | Medium |
| Lowest Revenue Generated Restaurants | Hard |
| Delivering and Placing Orders | Hard |
| Highest Earning Merchants | Hard |
| More Than 100 Dollars | Hard |
Marketplace cases where customers, Dashers and merchants all move at once. The data projects on delivery duration prediction, failed orders and delivery market modeling are the end-to-end version of that reasoning. In the real case you define success metrics, design the test and diagnose a change, and one prompt is deliberately vague.
| Case study | Difficulty |
|---|---|
| Customer Engagement and Disengagement | Medium |
| Product Launch Evaluation Framework | Medium |
| Diagnosing Regional Delivery Delays | Medium |
| Data project | Difficulty |
|---|---|
| Delivery Duration Prediction | Medium |
| Insights from Failed Orders | Medium |
| Delivery Market Modeling | Medium |
Experimentation on a marketplace: a test that missed significance, interference between treatment and control, and sample size for a two percent lift. There is no separate statistics round at DoorDash. These questions come up inside the cases.
| Question | Difficulty |
|---|---|
| Random Seats | Medium |
| Retraining Cadence for an ETA Model | Medium |
| Test That Missed Significance | Medium |
| Sample Size for 2% Lift | Medium |
| Interference in a Marketplace Test | Hard |