← DoorDash Interview Insights

DoorDash·Data Scientist·Technical Phone Screen·Senior

Senior
Jul 2026

Summary

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.

Questions Asked (1)

Q1

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.

A/B Testing & ExperimentationProduct Analytics & MetricsProduct Strategy
Author's notes

This is a beast of a question.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the business goal (e.g., increase order volume, improve delivery efficiency, or enter new markets) and define success metrics that align with that goal. Then design a randomized controlled experiment across cities, ensuring proper randomization, sample size, and analysis plan to detect meaningful effects while accounting for biases.

Pro tip: Consider the network effects and interference between treatment and control cities, as delivery options may spill over across city boundaries. Use a cluster randomized design with cities as randomization units and possibly a switchback or geo-based holdout to mitigate contamination.

1. Define Goal and Hypotheses

Clarify the primary objective (e.g., increase total orders, improve delivery times, or expand market share) and state null and alternative hypotheses. Identify secondary goals like customer satisfaction or courier utilization.

2. Select Metrics and Randomization Unit

Choose primary and guardrail metrics (e.g., orders per capita, delivery time, cancellation rate). Decide on randomization unit: city-level cluster randomization is typical for geo experiments to avoid contamination.

3. Design Experiment and Address Biases

Select a subset of comparable cities and randomly assign to treatment (bike delivery) or control (no bike delivery). Address potential biases: selection bias (ensure cities are similar), novelty effect, and spillover effects between cities.

4. Determine Sample Size and Duration

Calculate required sample size using power analysis, considering expected effect size, variance, and intra-cluster correlation. Determine test duration to capture stable behavior and account for seasonality.

5. Analyze Results and Make Recommendations

Use appropriate statistical methods (e.g., difference-in-differences, mixed-effects models) to estimate treatment effect. Check for heterogeneous effects and validate with robustness checks. Provide clear recommendation based on statistical and practical significance.

Key Points to Mention

  • Cluster randomization by city to avoid spillover and network effects
  • Primary metric: orders per capita or delivery time; guardrail metrics: cancellation rate, courier utilization
  • Power analysis accounting for intra-cluster correlation and minimum detectable effect
  • Potential biases: selection bias, novelty effect, seasonality, and spillover
  • Analysis techniques: difference-in-differences, CUPED, or synthetic control if needed
  • Consideration of operational constraints and cost-benefit analysis

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.