Did the DoorDash DS loop for a mid-level data scientist role earlier this year. The "what do they test" question for DS loops is always fuzzy online, so let me be specific.
Round breakdown (virtual onsite, 2026):
SQL: One dedicated SQL round, 45 minutes. Problem involved multiple tables (orders, dashers, restaurants, events), aggregations with window functions, and a question about funnel drop-off. Difficulty: real. Not "SELECT * FROM table." You need to be comfortable with CTEs, partitioned aggregations, and being able to explain your query out loud as you write it. They use a live shared environment. I wrote comments before each step to stay organized. The interviewer asked me to modify my query once to add a new filter condition mid-problem.
Product/case: Open-ended. "A restaurant partner says their orders dropped 20% last week. How do you investigate?" This is not a trick question. Walk through internal data checks first (is this real or a reporting issue), then external factors, then platform factors (algorithm changes, competitor activity, dasher availability). The structured diagnostic walk matters as much as your final hypothesis.
Stats / experimentation: Questions about A/B testing methodology. Specifically: when do you stop a test early, and what are the risks? Also a question about how you'd design an experiment when assignment is hard (Dashers are a limited, correlated population). If you've worked with switchback experiments or cluster randomization, mention that. They run a lot of experimentation internally and the interviewers know the tradeoffs.
Behavioral: Same as the SWE loop, cross-functional, STAR format.
What to prep: Window functions in SQL, diagnostic frameworks for metric drops, experiment design under non-IID conditions, and enough familiarity with marketplace data to sound like you've thought about delivery logistics before. The domain knowledge isn't a dealbreaker but it helps your examples land.
Comp for the level: offer I got was around $160k base plus equity, SF-area level DS, early 2026. YMMV.