Decompose and optimize delivery operational costs

Break down the operational cost of each order for a marketplace such as DoorDash, and recommend ways to reduce it without degrading the customer experience. 1) List the cost components using a driver tree — for example, courier_base_pay, surge/incentives, batching_bonus, support_costs, refunds/chargebacks, payment_processing, insurance, fraud, idle-time, failed_delivery, and packaging_subsidy — then define cost_per_order and contribution margin for each city, restaurant cohort, and hour. 2) Pinpoint the three most promising levers for lowering cost in expensive markets without hurting conversion or on-time performance; estimate their expected effect using historical elasticities or findings from earlier tests. 3) Design an experiment to reduce cost — for instance, dynamic batching thresholds, distance-based pricing adjustments, or restructuring incentives. State the randomization unit, stratification approach, power, guardrail metrics (late_rate, cancellations, complaints), planned heterogeneity analysis, and how you would avoid contamination from supply reallocation. 4) Lay out a monitoring plan for unintended consequences — such as courier churn, order acceptance, and reassignments — along with a rollback threshold. 5) If the experiment cuts $0.35 per order but raises late deliveries by 0.6 percentage points, give a decision framework for shipping versus iterating.

Overview: This question assesses a data scientist's ability to break down per-order operating costs, identify the highest-impact ways to improve unit economics, and design rigorous experiments and monitoring for a two-sided delivery marketplace. It tests cost modeling, causal inference, experiment design, and product analytics.