Choose the Product Bet That Cuts DoorDash's Late-Delivery Rate

You are a product manager at DoorDash, the marketplace that connects customers with restaurants and other merchants and with the couriers (Dashers) who carry the orders. The company wants fewer deliveries to arrive late. Which product or feature would you prioritize to make that happen? Commit to one investment, explain why it beats the alternatives, and say how you would know it worked. ```hint Find where the minutes are lost Trace one order from checkout to the customer's door, stage by stage, and ask at which stages the real time most often overruns the prediction, and which of those stages DoorDash can actually influence. ``` ### Clarifying Questions - What counts as late: arriving after the end of the time window shown at checkout, or after the latest estimate shown while the order was in progress? This decides whether making the promise more accurate counts as progress. - Is the target the overall share of late orders, or the smaller set of orders that arrive very late and generate refunds and support contacts? - Should the answer cover restaurant orders only, or also grocery and convenience orders, where in-store picking and substitutions add their own delays? - Are changes for Dashers and merchants (dispatch logic, merchant tools, pay incentives) in scope, or only features in the customer app? ### What a Strong Answer Covers - A precise definition of late, and the difference between delivering faster and promising more accurately, including why simply padding estimates is not a win - A breakdown of where delay arises across merchants, Dashers, dispatch and the final handoff, with a stated hypothesis about the largest source DoorDash can control - One prioritized feature, compared against alternatives on impact, effort, and which side of the marketplace bears the cost - Success metrics for how often and how badly orders are late, with guardrails on quoted delivery time, checkout conversion, Dasher earnings and merchant experience - A test plan that accounts for treatment and control orders sharing the same pool of Dashers ### Follow-up Questions - Suppose the data shows late orders are concentrated in weekend dinner peaks and bad weather. Does your priority change? - After launch, the late rate fell but the average delivery time quoted at checkout rose. Is that a success? - Why would a standard customer-level A/B test give a misleading result for a dispatch change, and what would you run instead?