Decompose Operational Cost per Order and Optimize Without Harming Experience

Context: You are evaluating operational cost per order for a two-sided food delivery marketplace (e.g., DoorDash). Your goal is to break down cost drivers, identify high-impact levers to reduce cost in expensive markets while maintaining customer conversion and on-time delivery, and design a robust experiment and monitoring plan.

Tasks

1. Cost Driver Tree and Definitions

• Enumerate cost buckets in a driver tree (e.g., courier_base_pay, surge/incentives, batching_bonus, support_costs, refunds/chargebacks, payment_processing, insurance, fraud, idle-time, failed_delivery, packaging_subsidy).
• Define cost_per_order and contribution margin at the most actionable granularity (city × restaurant cohort × hour).

2. Top Levers to Reduce Cost

• Identify the top 3 levers that can reduce cost in high-cost markets while maintaining conversion and on-time rate.
• Quantify expected impact using historical elasticities or prior experiments.

3. Experiment Proposal

• Propose an experiment to lower cost (e.g., dynamic batching thresholds, distance-based pricing tweaks, incentive reshaping).
• Specify unit of randomization, stratification, power, guardrails (late_rate, cancellations, complaints), heterogeneity analysis, and how to prevent supply reallocation contamination.

4. Monitoring and Rollback

• Describe a monitoring plan for unintended effects (e.g., courier churn, order acceptance, reassignments) and a rollback criterion.

5. Ship vs Iterate Framework

• If the test saves $0.35/order but increases late deliveries by 0.6 percentage points, provide a framework to decide whether to ship or iterate.