DoorDash·Data Scientist·Technical Phone Screen·Senior

Senior

Apr 2026

Summary

DoorDash DS interview with a meaty marketplace diagnostics question. The whole session was basically one long case study broken into four parts, and it went deeper than I expected for a single question.

Questions Asked (1)

DoorDash sees a 10% week-over-week drop in completed deliveries in Los Angeles. How would you identify, attribute, and quantify the key drivers behind this decline? Walk through potential root causes across supply, demand, merchant, product, external, and measurement dimensions, then outline how you'd test each hypothesis, decompose the drop quantitatively, and handle causal inference when A/B testing isn't an option.

Root Cause AnalysisProduct Analytics & MetricsA/B Testing & Experimentation

Author's notes

This one sprawled in ways I didn't anticipate.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by validating the metric and ruling out measurement artifacts, then systematically decompose the 10% drop across supply, demand, merchant, product, external, and measurement dimensions using quantitative methods like funnel analysis and cohort segmentation. For each hypothesis, outline a test (e.g., observational study, quasi-experiment) and prioritize based on potential impact and ease of validation, culminating in a causal inference approach when A/B testing isn't feasible.

Pro tip: Always begin by confirming the drop is real and not a data pipeline issue; then use a 'peel the onion' approach—segment by geography, time, user type, and device to localize the problem before diving into causes.

1. Validate and Scope the Decline

Verify the 10% drop is accurate by checking data quality, logging, and metric definitions. Segment the decline by time, geography (e.g., neighborhoods), platform, and user cohorts to identify where the drop is concentrated.

2. Decompose the Funnel and Identify Affected Stages

Break down the delivery funnel (app open → search → add to cart → checkout → delivery) and compare conversion rates week-over-week to pinpoint which stage(s) drive the decline. Quantify the contribution of each stage to the overall drop.

3. Generate and Test Hypotheses Across Dimensions

List potential root causes across supply (e.g., Dasher availability), demand (e.g., user churn), merchant (e.g., closures), product (e.g., bug), external (e.g., weather), and measurement (e.g., tracking). For each, design a test: compare affected vs. unaffected segments, use time-series analysis, or leverage natural experiments.

4. Quantify Drivers and Attribute Impact

Use decomposition methods (e.g., additive or multiplicative models) to estimate how much each factor contributes to the 10% drop. Prioritize drivers by impact and validate with sensitivity analysis.

5. Apply Causal Inference When A/B Testing Isn't Possible

If randomization is infeasible, use quasi-experimental methods like difference-in-differences, synthetic control, or instrumental variables to establish causality. Ensure assumptions are met and triangulate with qualitative insights.

Key Points to Mention

Metric validation: check for data pipeline issues, logging errors, or definition changes before attributing to real-world causes.
Funnel decomposition: isolate which stage (e.g., order completion) drives the drop and quantify its contribution.
Segmentation: analyze by geography, time, user demographics, device, and Dasher cohorts to localize the issue.
Root cause dimensions: systematically evaluate supply (Dasher supply), demand (user behavior), merchant (store availability), product (app changes), external (weather, traffic), and measurement (tracking).
Causal inference techniques: difference-in-differences, synthetic control, propensity score matching, and instrumental variables when A/B tests are not feasible.
Quantification: use contribution analysis, regression, or Shapley values to attribute the drop to specific drivers.

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