← DoorDash Interview InsightsMachine Learning Engineer

DoorDash·Machine Learning Engineer·Onsite - System Design / Architecture·Senior

SeniorPrefer not to say
May 2026Remote

Summary

DoorDash ML/product case round, open-ended and pretty broad. The whole thing was one big system design question dressed up as a business case, which I wasn't fully expecting.

Questions Asked (1)

Q1

Design a system to send promotional offers like discounts and coupons to drive consumer growth. Walk through your goals and metrics, how you'd pick the target audience, how you'd assign treatments using uplift modeling, how you'd handle budget constraints, how you'd run A/B tests, and how you'd distinguish long-term retention from short-term redemption effects.

A/B Testing & ExperimentationProduct Analytics & MetricsSystem Design
Author's notes

This one sprawled in a way that made it hard to stay organized.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by framing the problem around incremental impact: define success metrics (e.g., incremental orders, ROI, long-term retention) and then walk through the pipeline from audience selection to treatment assignment, budget allocation, experimentation, and long-term effect measurement. Emphasize uplift modeling to target persuadable users and explain how you'd validate incrementality via A/B tests and holdout groups.

Pro tip: Mention that you'd use a two-model uplift approach (e.g., T-learner) and validate with a Qini curve, but also stress the importance of a long-term holdout to separate true incrementality from promotion-induced pull-forward effects.

1. Define Goals and Metrics

Clarify business objectives (e.g., consumer growth, retention) and select primary metrics like incremental orders, ROI, and long-term retention rate, along with guardrail metrics (e.g., margin dilution, cannibalization).

2. Target Audience Selection

Use historical data and predictive modeling to identify high-potential segments, focusing on persuadable users via uplift modeling rather than those who would convert anyway or never convert.

3. Treatment Assignment with Uplift Modeling

Train an uplift model (e.g., T-learner, X-learner) to estimate individual treatment effects, then assign offers to users with the highest predicted uplift, possibly incorporating budget constraints via optimization.

4. Budget-Constrained Allocation

Formulate budget allocation as an optimization problem (e.g., knapsack or linear programming) that maximizes total predicted uplift subject to cost constraints, and consider dynamic budget pacing.

5. A/B Testing and Long-Term Measurement

Run randomized controlled trials with holdout groups to measure incremental impact, and design long-term experiments (e.g., extended holdouts) to distinguish short-term redemption from sustained retention effects.

Key Points to Mention

  • Uplift modeling techniques (T-learner, X-learner) and evaluation via Qini curves
  • Incremental metrics vs. raw redemption rates; importance of holdout groups
  • Budget optimization under constraints (e.g., knapsack, linear programming)
  • A/B test design: randomization unit, power analysis, guardrail metrics
  • Long-term holdout to measure retention and separate pull-forward from true incrementality
  • Potential biases: selection bias, cannibalization, and margin dilution

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