← DoorDash Interview Insights

DoorDash·Data Scientist·Technical Phone Screen·Senior

Senior
May 2026

Summary

DoorDash data science interview that went deep on retention and churn measurement for a transactional app. The whole session was basically one long case question with five layers, and I did not see the survival analysis part coming at all.

Questions Asked (5)

Q1

How would you precisely define retention and churn for a transactional consumer app, and which definitions would you choose based on the decision you're trying to support?

Product Analytics & MetricsData Modeling
Author's notes

This is where I spent probably too long hedging.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by acknowledging that retention and churn are not one-size-fits-all metrics; they depend on the business question. Then walk through different definitions (e.g., order-based, user-based, time-based) and explain how you would choose based on the decision at hand, such as measuring product-market fit, evaluating a new feature, or forecasting revenue.

Pro tip: Tie your definitions to actionable decisions—e.g., if you're deciding whether to invest in a re-engagement campaign, define churn as a user who hasn't ordered in 30 days, but if you're measuring long-term retention, use a cohort-based approach. This shows you think like a business partner, not just a statistician.

1. Clarify the decision context

Ask what decision the metric will inform: product improvements, marketing spend, or financial forecasting. This determines the appropriate granularity and time window.

2. Define retention and churn at the user level

Retention: a user who performs a key action (e.g., places an order) within a specified time period after their first action. Churn: a user who was previously active but has not performed the key action within a defined inactivity window.

3. Choose time windows and activity thresholds

Select windows based on natural usage frequency (e.g., for DoorDash, 30 days might be typical). Consider using rolling windows or cohort-based analysis to account for seasonality and user lifecycle.

4. Align definitions with business goals

For growth, use a shorter window to capture quick re-engagement; for LTV, use longer windows. For transactional apps, consider order-based retention (repeat purchase) vs. user-based retention.

5. Validate and iterate

Test how different definitions impact metrics and decisions. Use A/B tests or backtesting to ensure the chosen definition drives the right actions.

Key Points to Mention

  • Distinguish between user churn (user stops using app) and order churn (user stops ordering but may still browse).
  • Consider time-based cohorts (e.g., weekly cohorts) to track retention over time and avoid biases from aggregate metrics.
  • Define a 'key action' that signifies value—for DoorDash, it's placing an order, not just opening the app.
  • Use survival analysis or probabilistic models to estimate churn likelihood, especially for non-contractual settings.
  • Align with stakeholders: finance may prefer revenue churn, while product may focus on user churn.
  • Mention the importance of handling edge cases like users who never ordered (not churned, but never retained).

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

Q2

Walk through the formulas for cohort retention and churn rates. How do you handle right-censoring and delayed conversion, and what statistical pitfalls should you watch for?

Product Analytics & MetricsA/B Testing & Experimentation
Author's notes

I got the basic risk-set formula right but stumbled when they pushed on right-censoring.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining cohort retention and churn rates with clear formulas, emphasizing the importance of consistent time windows and cohort definitions. Then explain how right-censoring and delayed conversion affect these metrics, and describe methods like survival analysis or adjusted denominators to handle them. Finally, highlight statistical pitfalls such as survivorship bias, seasonality, and misaligned cohorts, and suggest ways to mitigate them.

Pro tip: When discussing right-censoring, mention that using a fixed observation window (e.g., 30 days) can bias retention estimates, and propose using Kaplan-Meier curves or including only cohorts with complete data to avoid misleading conclusions.

1. Define cohort retention and churn formulas

Clearly state the formulas: Retention Rate = (Number of users from cohort active in period t) / (Initial cohort size). Churn Rate = 1 - Retention Rate, or (Users lost during period) / (Users at start of period). Emphasize that these are complementary and depend on the chosen time period.

2. Explain right-censoring and its impact

Right-censoring occurs when users have not yet had the chance to churn or convert by the end of the observation window. This leads to underestimation of churn and overestimation of retention if not handled. Mention that censoring is more severe for recent cohorts.

3. Describe methods to handle right-censoring and delayed conversion

Use survival analysis techniques like Kaplan-Meier estimation or Cox proportional hazards models to account for censored data. Alternatively, restrict analysis to cohorts with sufficient observation time or use adjusted denominators (e.g., only users who have reached the time period). For delayed conversion, consider using a longer attribution window or modeling conversion probabilities over time.

4. Identify statistical pitfalls and mitigation strategies

Watch for survivorship bias (only analyzing users who remain), seasonality (comparing cohorts across different seasons), and misaligned cohorts (inconsistent definitions). Mitigate by using consistent cohort definitions, controlling for time effects, and validating with holdout groups or A/B tests.

5. Connect to business context and experimentation

Relate retention and churn metrics to DoorDash's business, e.g., how they inform customer lifetime value and A/B testing. Emphasize that proper handling of censoring ensures accurate measurement of experiment impact on retention.

Key Points to Mention

  • Cohort retention formula: active users in period t divided by initial cohort size; churn is the complement.
  • Right-censoring: recent cohorts have incomplete data; use survival analysis or restrict to mature cohorts.
  • Delayed conversion: extend observation window or model time-to-conversion; avoid premature conclusions.
  • Statistical pitfalls: survivorship bias, seasonality, misaligned cohorts, and multiple testing.
  • Survival analysis methods: Kaplan-Meier, Cox proportional hazards, and their assumptions.
  • Business relevance: retention/churn drive LTV and inform A/B test metrics at DoorDash.

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

Q3

How would you measure the long-term retention impact of a treatment like a 20% discount using survival analysis? Define the key concepts and explain how you'd compare groups and control for confounders.

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

Completely blindsided me.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining the retention metric and the survival analysis framework, then explain how to set up the comparison between treatment and control groups, and finally discuss methods to control for confounders such as propensity score matching or stratification. Emphasize the importance of measuring time-to-event (e.g., churn) and using hazard ratios to quantify the treatment effect.

Pro tip: Mention that you would validate the proportional hazards assumption and consider time-varying effects, as the discount's impact may diminish or change over time, which is crucial for a realistic retention analysis.

1. Define retention and survival metrics

Clarify what 'retention' means in this context (e.g., continued ordering after first order) and define the event of interest (e.g., churn) and time scale (e.g., days since first order).

2. Set up survival analysis

Choose an appropriate survival model (e.g., Kaplan-Meier curves, Cox proportional hazards) to estimate survival functions and compare groups.

3. Compare treatment and control groups

Use log-rank test for initial comparison and hazard ratios from Cox model to quantify the treatment effect on retention.

4. Control for confounders

Apply propensity score matching, stratification, or include covariates in the Cox model to adjust for differences between groups.

5. Validate and interpret

Check model assumptions (e.g., proportional hazards) and interpret results in business terms, considering practical significance.

Key Points to Mention

  • Censoring: handle right-censoring for users who haven't churned by end of study.
  • Kaplan-Meier estimator: visualize survival curves for treatment vs. control.
  • Cox proportional hazards model: estimate hazard ratio and adjust for covariates.
  • Propensity score matching: balance confounders between groups.
  • Log-rank test: compare survival distributions.
  • Time-varying effects: consider if discount impact changes over time.

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

Q4

Show with made-up numbers how rolling retention and strict cohort retention can disagree, and explain how you'd reconcile the two for an executive audience.

Product Analytics & MetricsStakeholder Management
Author's notes

This was actually fun once I relaxed.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clearly defining rolling retention (e.g., active in month N or later) and strict cohort retention (e.g., active exactly in month N), then walk through a simple made-up cohort with numbers that show how the two metrics can diverge. Finally, explain how you'd reconcile them for an executive audience by focusing on the business question, using a consistent definition, and presenting both metrics with clear context and a recommendation.

Pro tip: Use a simple, relatable example like a cohort of 100 users and show the cumulative vs. exact activity to make the divergence obvious; then emphasize that the choice of metric should depend on the decision it informs, not on which number looks better.

1. Define the metrics

Briefly define rolling retention as the percentage of a cohort active in month N or any later month, and strict cohort retention as the percentage active exactly in month N. Clarify that rolling retention is cumulative and strict is point-in-time.

2. Construct a made-up example

Create a small cohort (e.g., 100 users) with monthly activity patterns that cause the two metrics to diverge. For instance, show that strict retention drops sharply after month 1, while rolling retention stays high because users return later.

3. Calculate and contrast

Compute both metrics for each month and highlight the gap. Explain why the gap occurs: rolling retention counts returning users, while strict retention does not.

4. Reconcile for executives

Recommend aligning on the business question: use strict retention for engagement frequency and rolling retention for long-term value. Suggest presenting both with clear labels and a narrative that ties to strategic goals.

5. Provide a recommendation

Advise choosing one primary metric for decision-making, while using the other as a diagnostic. Emphasize transparency about definitions and avoiding cherry-picking.

Key Points to Mention

  • Rolling retention is cumulative (active in month N or later), strict retention is point-in-time (active exactly in month N).
  • Divergence happens when users churn and return; rolling retention masks churn, strict retention highlights it.
  • Use a simple cohort example (e.g., 100 users) with clear monthly activity to illustrate the difference.
  • For executives, tie the metric choice to the business decision: strict for engagement, rolling for LTV or long-term retention.
  • Present both metrics with clear definitions and avoid switching definitions mid-conversation.
  • Recommend a primary metric and use the other as a supporting diagnostic to build trust and avoid confusion.

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

Q5

How do you set washout, observation, and attribution windows in a retention experiment, and how do those choices affect statistical power and potential bias?

A/B Testing & ExperimentationTechnical Trade-offs
Author's notes

Shorter answer from me than it deserved.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining the three windows and their purpose in a retention experiment, then explain how each choice trades off statistical power against bias. Use a concrete example (e.g., a DoorDash retention campaign) to illustrate how you would set them based on business cycles and user behavior, and discuss sensitivity analyses to validate choices.

Pro tip: Always align window lengths with the natural usage cycle of the product (e.g., weekly for food delivery) and pre-register your analysis plan to avoid p-hacking; mention that you would run sensitivity checks with different windows to ensure robustness.

1. Define the windows

Clearly define washout (time after treatment before measuring), observation (duration of measurement), and attribution (how long after an action to credit it) windows. Explain their roles in isolating treatment effects and capturing outcomes.

2. Link to business context

Choose window lengths based on the product's usage frequency and retention dynamics. For DoorDash, consider weekly ordering cycles and seasonality to set appropriate observation periods.

3. Analyze impact on statistical power

Discuss how longer observation windows increase sample size and power but may introduce noise; shorter windows reduce power but limit confounding. Washout reduces sample but can improve signal.

4. Address potential biases

Explain how misaligned windows can cause biases: too short washout may carryover treatment effects; too long observation may include unrelated events; attribution window mismatches can under/overcount conversions.

5. Recommend validation and sensitivity

Propose pre-registration, sensitivity analyses across window choices, and using holdout groups to validate that results are robust and not artifacts of window selection.

Key Points to Mention

  • Definition and purpose of washout, observation, and attribution windows
  • Trade-off between statistical power and bias when choosing window lengths
  • Importance of aligning windows with product usage cycles (e.g., weekly for DoorDash)
  • Risk of carryover effects and confounding with improper washout
  • Attribution window impact on conversion metrics and novelty effects
  • Use of sensitivity analysis and pre-registration to ensure robust results

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