Uber Marketplace Fix: A Plan to Shrink Trip Cancellations From 25% to 5%

You are a product manager at Uber. Today 25% of trips are cancelled, and you have been asked to get that figure down to 5%. What would you do? Walk through how you would find out where the cancellations come from, which ones you would go after first, what you would change in the product or its policies, and how you would track progress toward the target. ```hint Break the number apart first A single 25% blends very different events. Split it by who cancels and by the moment in the trip when it happens, and ask whether each kind is actually a problem, before you choose what to fix. ``` ### Clarifying Questions - Does the 25% describe ride-hailing trips or Uber Eats orders, and is it a global average or concentrated in certain cities? - Who is cancelling in this figure: riders, drivers or both, and is the rate measured against all requests or only against trips a driver has already accepted? - Do cancellations inside a short free-cancellation window, or ones made after a rider fails to show up, count toward the 25%? - By when should the rate reach 5%, and are rider fees or driver penalties allowed as levers? ### What a Strong Answer Covers - A precise definition of the cancellation rate, and a challenge of whether every cancellation should count against the 5% target - A breakdown by side of the marketplace and by stage of the trip, with a reasoned choice of which buckets to attack first - Fixes tied to specific causes for riders and for drivers, prioritized by impact and effort, that plausibly add up to a five-fold reduction - Success metrics with guardrails that catch harm elsewhere, such as slower matching, more unserved requests or lower driver earnings - Risks such as gaming of the metric, drivers leaving over penalties, and riders feeling locked into trips they no longer want ### Follow-up Questions - The data shows that most cancellations are drivers cancelling soon after accepting, once they see how far away the pickup is. What do you change first, and what could go wrong? - How would you test a new matching rule when riders and drivers in a city share the same pool? - Some drivers start asking riders to cancel so that the cancellation does not count against the driver. How would you detect and respond to it? - Suppose the scope is narrowed to a single screen: the waiting page a rider sees between requesting a ride and being picked up. It covers both the "finding your driver" state and the "driver assigned, on the way" state, which shows the car's live location and ETA. How would you reduce cancellations on that page, and what changes in your metric and your plan?