Improve biker delivery with metrics and levers

## Case: Optimize Delivery Performance for Bike Couriers You are a Data Scientist at a food-delivery marketplace such as DoorDash or Uber Eats. Your team focuses on bike couriers in dense cities, where delivery outcomes depend heavily on geography, weather, merchant operations, courier supply, and customer demand. Leadership asks: **What should we optimize for, and how would you improve biker delivery performance?** ### Constraints & Assumptions - Treat this as a three-sided marketplace problem involving customers, couriers, and merchants. - Assume bike couriers operate in dense urban zones where hills, bridges, high-rise buildings, pedestrian areas, parking or locking constraints, and weather can meaningfully affect delivery time. - Do not optimize a single metric in isolation if it creates safety, earnings, merchant, or customer-experience harm. - You may propose product, operations, routing, dispatch, merchant-experience, or incentive changes, but each should be measurable. - Focus on an analysis and experimentation plan that can handle confounding from weather, zone, time of day, demand shocks, and courier mix. ### Clarifying Questions to Ask - What is the business priority: customer reliability, courier earnings, marketplace efficiency, merchant quality, or a balanced portfolio? - Are we trying to improve all bike deliveries or a specific segment such as downtown peak hours, bad weather, long pickup waits, or short-distance orders? - What data do we already collect on courier location, route choice, pickup wait, merchant prep, weather, and customer promise times? - Are there safety, compliance, or courier fairness constraints that limit incentives or routing recommendations? - Can we randomize by courier, zone, merchant, or zone-time block? ### Part 1 - Define the Objective Propose a clear objective for "better" bike delivery. Explain what outcome you would optimize and why it is aligned with the marketplace. #### What This Part Should Cover - A primary objective such as improving reliable, on-time delivery per courier-hour while preserving safety and earnings. - Why a bike-specific objective may differ from car or scooter delivery. - A distinction between customer outcomes, courier outcomes, merchant outcomes, and marketplace efficiency. - A warning against optimizing pure speed if it worsens safety, ETA honesty, merchant prep quality, or courier earnings. ### Part 2 - Build the Metrics Framework Define primary metrics, diagnostic metrics, and guardrails. Include formulas where helpful and cover customer, courier, merchant, and marketplace perspectives. #### What This Part Should Cover - A small set of north-star or decision metrics, such as on-time delivery rate, lateness tail, completed deliveries per courier-hour, and contribution margin per zone-hour. - Diagnostic decomposition of total delivery time into assignment, travel-to-merchant, pickup wait, travel-to-customer, handoff, reassignment, and cancellation components. - Courier metrics such as earnings per active hour, utilization, safety incidents, distance or elevation burden, and acceptance rate. - Merchant metrics such as prep-time accuracy, courier wait time, handoff friction, and cancellation rate. - Guardrails for safety, refund/contact rate, food quality, ETA inflation, courier churn, merchant satisfaction, and inequitable allocation. ### Part 3 - Identify Data, Features, and Risks List the data and features needed to understand biker delivery performance, and call out data quality risks. #### What This Part Should Cover - Zone, distance, elevation, road class, bridge or plaza constraints, building type, parking/locking availability, and route features. - Weather, time-of-day, day-of-week, event, demand, and supply features. - Merchant prep-time distributions, cuisine or packaging sensitivity, handoff process, and historical pickup wait. - Courier tenure, equipment type, active time, acceptance behavior, and location pings, while respecting privacy and policy constraints. - Data risks such as noisy GPS, biased missingness, inaccurate prep timestamps, selection bias, and confounding from demand/supply conditions. ### Part 4 - Propose Levers and Validation Plan Suggest actionable ideas to test or roll out, then outline an experimentation or causal plan that can separate real impact from confounding. #### What This Part Should Cover - Levers such as bike-aware routing, better batching rules, merchant prep-time calibration, pickup handoff improvements, courier-zone recommendations, weather incentives, and building-dropoff instructions. - A recommended test design, such as zone-time switchbacks or clustered experiments, when interference is likely. - Methods such as matched comparisons, difference-in-differences, causal forests or heterogeneity analysis, and pre/post monitoring only as a weaker fallback. - Power, ramp, seasonality, novelty, and spillover considerations. - A decision rule for shipping based on primary metric movement and guardrail health. ### What a Strong Answer Covers A strong answer treats this as a marketplace optimization problem, not a simple route-speed problem. It defines a measurable objective, decomposes delivery performance into controllable drivers, names data and instrumentation risks, proposes concrete levers, and recommends an interference-aware validation design with clear guardrails. ### Follow-up Questions - How would your design change during heavy rain or extreme heat? - How would you detect whether improvements come from routing quality versus merchant prep-time changes? - What would you do if customer delivery time improves but courier earnings per hour falls? - How would you estimate impact if you cannot run a randomized experiment? - Which segments would you prioritize first and why?