Diagnose and reduce cold-food refund costs

DoorDash faces an ongoing high cost from 100% refunds on cold-food complaints. Design a plan to reduce monetary impact while preserving customer trust. Be specific: (a) Decompose cost into volume × complaint rate × refund amount and define a primary success KPI (e.g., average cold-food refund cost per order) plus guardrails (e.g., re-order rate, CSAT, complaint re-open rate, fraud flags). (b) List at least five diagnostic analyses you would run to identify drivers (e.g., lateness vs. promised time, store prep time vs. travel time, order value, distance, courier wait at pickup, cuisine, packaging/thermal-bag usage, weather, time-of-day, repeat complainers). (c) Propose a tiered refund policy where refund percentage is a function of lateness and order value; specify exact thresholds (e.g., <10 min late = 0%, 10–29 min = 50%, ≥30 min = 100%; cap at $X; exclude alcohol/ice cream, etc.) and justify with CX and fraud trade-offs. (d) Describe an offline policy evaluation using historical orders and refund logs to estimate savings and risks: cohorting, counterfactuals for orders that would newly receive partial refunds, selection bias (who complains), mislabeling of reasons, and censoring. (e) Provide a rollout strategy (city-level holdouts, sequential ramp, kill switch) and a pre-registered decision rule to ship (e.g., ≥8% cost reduction with no degradation on guardrails beyond set MDE thresholds).