← DoorDash Interview Insights

DoorDash·Data Scientist·Technical Phone Screen·Senior

Senior
Aug 2025Remote

Summary

DoorDash data science case focused entirely on a real operational scenario: a 12% drop in completed orders in LA. The whole thing was one big structured problem, broken into diagnosis, experiment design, and quasi-experimental fallback. Dense and technical, no fluff.

Questions Asked (3)

Q1

Completed orders in a major market dropped 12% week-over-week. Walk through how you'd build a funnel-based diagnosis to determine whether the issue is supply-side or demand-side, including what cuts, plots, and thresholds you'd use.

Root Cause AnalysisProduct Analytics & Metrics
Author's notes

This is the kind of question where you can talk for 20 minutes and still feel like you barely scratched it.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by decomposing the 12% drop into a funnel (sessions → product views → add-to-cart → checkout → order completion) and compare week-over-week to isolate where the drop occurs. Then segment by supply-side (e.g., store availability, dasher supply) and demand-side (e.g., traffic, conversion) factors to determine the root cause. Use thresholds and plots to validate hypotheses and prioritize next steps.

Pro tip: Always validate data quality first—check for tracking issues, seasonality, or one-off events (e.g., holidays, outages) that could explain the drop before diving into funnel analysis. Also, consider both sides: a supply issue can manifest as demand-side symptoms (e.g., low conversion due to out-of-stock).

1. Validate the data and define the funnel

Confirm the 12% drop is real and not due to tracking errors, seasonality, or known events. Define the key funnel stages: sessions, product views, add-to-cart, checkout initiation, and order completion.

2. Analyze funnel conversion rates week-over-week

Calculate conversion rates between each stage for the current and previous week. Identify which stage(s) show significant drops (e.g., >2% relative change) and plot trends to visualize the impact.

3. Segment by supply and demand factors

Split the funnel metrics by dimensions like market, store, dasher supply, user cohort, and device. Compare supply-side metrics (e.g., store open rate, dasher availability) and demand-side metrics (e.g., sessions, search volume) to see which aligns with the drop.

4. Apply thresholds and statistical tests

Use thresholds (e.g., >5% drop in conversion) and statistical significance testing to distinguish noise from real changes. Check if the drop is concentrated in specific segments or broad-based.

5. Synthesize findings and recommend actions

Based on the funnel stage and segment where the drop occurs, determine if it's supply-side (e.g., fewer stores open, dasher shortage) or demand-side (e.g., lower traffic, reduced intent). Propose next steps like deeper dives or A/B tests.

Key Points to Mention

  • Funnel stages: sessions, product views, add-to-cart, checkout, order completion
  • Week-over-week comparison and conversion rate calculations
  • Segmentation by market, store, dasher, user cohort, and device
  • Supply-side metrics: store availability, dasher supply, delivery times
  • Demand-side metrics: traffic, search volume, user engagement
  • Thresholds for significance (e.g., >5% drop) and statistical tests
  • Data quality checks and external factors (holidays, outages)

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

Q2

If supply shortage during the dinner peak is your leading hypothesis, design an experiment to validate it. Cover the randomization unit, primary and guardrail metrics, power assumptions, sample size, duration, and how you'd handle spillover and seasonality.

A/B Testing & ExperimentationProduct Analytics & Metrics
Author's notes

I went with ZIP cluster randomization to limit interference between treatment and control, which felt right.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clearly defining the hypothesis and the intervention (e.g., increasing Dasher supply during dinner peak). Then design a randomized experiment at the appropriate unit (e.g., city or zone) with well-chosen primary and guardrail metrics, power analysis, and duration. Address spillover and seasonality through design choices like cluster randomization, buffer zones, or time-based controls.

Pro tip: In marketplace experiments, interference is the norm, not the exception. Consider using a switchback or cluster-randomized design and pre-register your analysis plan to avoid p-hacking.

1. Define hypothesis and intervention

Clearly state the hypothesis (supply shortage during dinner peak causes issues) and the intervention (e.g., incentives to increase Dasher supply during peak hours). Specify the expected outcome (e.g., reduced delivery times, increased completion rate).

2. Choose randomization unit and metrics

Select the randomization unit (e.g., city, zone, or time-based switchback) to minimize spillover. Define primary metric (e.g., order completion rate) and guardrail metrics (e.g., delivery time, Dasher utilization, customer satisfaction).

3. Power analysis and sample size

Estimate baseline metrics, minimum detectable effect (MDE), power (80%), and significance level (5%). Calculate required sample size (e.g., number of cities or time periods) and duration to detect the MDE.

4. Address spillover and seasonality

Use cluster randomization or switchback designs to mitigate spillover. For seasonality, run the experiment for multiple weeks, include day-of-week and time-of-day controls, or use a difference-in-differences approach.

5. Analysis and decision

Analyze results using appropriate statistical methods (e.g., mixed-effects models for clustered data). Check guardrail metrics for unintended consequences. Decide whether to roll out, iterate, or abandon the intervention.

Key Points to Mention

  • Randomization unit: city, zone, or switchback (time-based) to handle interference
  • Primary metric: order completion rate or delivery time; guardrail metrics: Dasher utilization, customer satisfaction, cost per delivery
  • Power analysis: baseline conversion, MDE, alpha=0.05, power=0.80, sample size calculation
  • Duration: at least 2-4 weeks to capture weekly seasonality and novelty effects
  • Spillover: use cluster randomization, buffer zones, or switchback designs
  • Seasonality: include time fixed effects, run for multiple weeks, or use control periods

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

Q3

If a randomized experiment isn't feasible, what quasi-experimental method would you use instead? Describe the setup, pre-trend checks, and robustness tests in enough detail that another analyst could run it.

A/B Testing & ExperimentationRoot Cause Analysis
Author's notes

Went with difference-in-differences using other SoCal metros as controls, San Diego and Riverside being the obvious candidates.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Choose a quasi-experimental method like difference-in-differences (DiD) or synthetic control that fits the business context, then walk through the setup, pre-trend checks, and robustness tests in a structured, step-by-step manner. Emphasize how you would validate assumptions and quantify uncertainty to make causal claims credible.

Pro tip: Always discuss how you would handle violations of parallel trends—e.g., using synthetic control or matching—and mention that you'd pre-register your analysis plan to avoid p-hacking, which shows rigor and maturity.

1. Select the quasi-experimental method

Choose a method like difference-in-differences, synthetic control, or regression discontinuity based on the intervention and data structure. Justify why it's appropriate for the DoorDash context (e.g., market-level rollout).

2. Define the setup and treatment/control groups

Specify the treatment group (e.g., markets where the feature launched) and control group (e.g., similar markets without launch). Describe the time periods (pre and post) and the outcome metric (e.g., orders per user).

3. Conduct pre-trend checks

Plot the outcome over time for both groups and test for parallel pre-trends using event-study regressions or placebo tests. If trends diverge, consider alternative methods like synthetic control.

4. Run robustness tests

Perform placebo tests (e.g., fake treatment dates), sensitivity analyses (e.g., different control groups), and check for spillover effects. Also, test for anticipation effects and heterogeneous treatment effects.

5. Estimate and interpret the effect

Use regression models (e.g., two-way fixed effects) to estimate the treatment effect, and report confidence intervals. Discuss limitations and potential biases (e.g., omitted variable bias).

Key Points to Mention

  • Difference-in-differences (DiD) with two-way fixed effects
  • Synthetic control method for constructing a counterfactual
  • Parallel trends assumption and event-study plots for pre-trend checks
  • Placebo tests and sensitivity analyses for robustness
  • Handling staggered adoption and heterogeneous treatment effects (e.g., Callaway & Sant'Anna)
  • Pre-registration of analysis plan to mitigate p-hacking

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