Lyft Metric Investigation: Root-Causing a Sudden Jump in Cancelled Trips

You are a product manager at Lyft. Ride cancellations rose sharply this week compared with a normal week. What could explain the jump, and how would you work out which explanation is the real one? ```hint Who cancelled, and at what point A ride can be cancelled by the rider or the driver, before a driver is matched, while the driver is on the way, or at pickup. Break the metric down along these lines before you guess at causes. ``` ### Clarifying Questions - Is the increase in the number of cancellations or in the cancellation rate per ride request, and did requests themselves change this week? - Does the metric count rider cancellations, driver cancellations and requests that ended with no driver found together, or separately? - Is the rise concentrated in certain cities, ride types, platforms or app versions, and did it start on one day or build through the week? ### What a Strong Answer Covers - Confirming the metric's definition and ruling out a logging or data-pipeline cause before anything else - A decomposition by who cancelled, the stage of the trip, and segments such as market, ride type, platform, app version and rider or driver tenure - Hypotheses spanning internal changes (releases, matching, pricing, policies, driver pay) and external ones (weather, events, competitors, driver supply, outages), each with the data that would confirm or reject it - A sensible order for the checks, and how the impact on completed rides is sized - Next steps once the cause is found, with guardrail metrics and monitoring ### Follow-up Questions - The breakdown shows the rise comes from drivers cancelling after they accept, mostly on long pickups in two cities. What do you do next? - Cancellations rose but completed rides did not fall. Is this still a problem? - How would you set up monitoring so the next spike is detected and attributed within a day?