Instacart — Core Delivery — A/B Test & Causal Inference

Role context: Data Scientist, Core Delivery · Est. study time: 55 min · 5 questions

How these map to the role

These are the causal problems of a coupled grocery-logistics marketplace: choosing a unit under interference, geo quasi-experiments, switchbacks for real-time systems, ratio/tail metrics, and heterogeneous effects. Each answer is a coaching walkthrough: a Sample answer (clarify → approach → simulated back-and-forth → clear call), a Deep dive with the math, then a Grading rubric.

The recurring instinct: shared shoppers and batching couple orders, delivery is geospatial, and effects are heterogeneous — so design around interference and geography, and measure the tail and all sides.

Questions (5)

Pick the unit of randomization for a delivery experimentExperiment designHardMeasure a market-wide rollout with no clean controlCausal inferenceHardDesign a switchback test for a real-time dispatch changeTest selectionHardMeasure delivery-time quality — mean, or the tail?Metric designMediumA change that nets to zero but helps some, hurts othersHeterogeneous effectsHard