DoorDash Data Scientist Interview Experience — SQL Screen With Doubled Difficulty and Tight Time

Part 1 was product, and since I was well prepared it felt fine. Prepare a lot more for high-frequency Biker and Cold Food. Read the attached switchback experiment design, randomization, and analysis parts carefully.

SQL was not the usual four questions (someone on the forum summarized them before, but the answers weren't quite right, so correct them yourself).

I got new questions, and the difficulty was doubled. Be thoroughly familiar with the following test points:

• % calculation: the numerator is derived through a simple calculation based on a new condition; here the new condition is non-aggregation.
• % calculation: the new condition requires aggregation; pay attention to whether the denominator includes rows where the condition label is 0.
• Computing change with window functions, multiple windows, including whether you calculate up to the current row or use unbounded.
• Follow-up: consider a unit with multiple changes over time — calculate the difference between individual records versus considering the whole partition.

Overall suggestion: SQL time is very tight and the difficulty is high, and the question stem and table information are very dense.

• The interviewer specifically said there's no need to talk while writing — just write directly, because there isn't enough time.
• The first question alone is already at the difficulty of the last question in a Meta DSA phone screen, and it escalates from there.
• Consider using CTEs since later parts build on them.
• Get comfortable computing multiple window functions at the same time.