## Diagnose a Decline in Order Acceptance Rate The order acceptance rate on a delivery marketplace has decreased. Describe how you would determine whether the decline is real, identify its drivers, and recommend a response. ### Constraints & Assumptions - Acceptance rate is a ratio, so changes in both accepted offers and eligible offers matter. - The logging or definition of an offer may have changed. - Repeated offers, timeouts, and reassignment can make the denominator ambiguous. - A lower acceptance rate is not automatically harmful if offer quality or completed deliveries improves. ### Clarifying Questions to Ask - What is the exact numerator, denominator, time window, and unit: offer, order, courier session, or courier-day? - When did the decline begin, how large is it, and is it concentrated in any market or platform version? - Were there changes to dispatch, pay, distance, batching, notifications, supply, or demand? - Is the business concerned about assignment latency, cancellations, completion, or courier experience? - Which logging changes or known incidents overlap the movement? ### What a Strong Answer Covers - Validates the metric definition, deduplication rules, eligibility, instrumentation, and statistical significance. - Decomposes the aggregate change into mix shifts and within-segment changes using appropriate denominators. - Segments by offer attributes, market, time, courier cohort, and product version while controlling the false-discovery risk of broad slicing. - Builds a timeline of candidate changes and uses experiments or defensible quasi-experimental comparisons to test mechanisms. - Connects acceptance to completed deliveries, assignment latency, earnings, customer outcomes, and supply health. - Produces a prioritized recommendation with monitoring, uncertainty, and a clear distinction between diagnosis and causality. ### Follow-up Questions 1. How would you analyze an acceptance-rate decline caused by a new dispatch policy? 2. What if acceptance falls but completion rate rises? 3. How would you account for the same order being offered multiple times? 4. How would you avoid overreacting to noisy changes in small markets?