Ride-hailing analytics: investigating a 10% decline in cancellations by riders

You are a product manager at Uber. A metric that tracks rider cancellations is down 10%. Walk through how you would triage it: how you would confirm what changed, find the cause, and decide whether anything needs to be done. ```hint Start from the definition Write the metric as a formula before listing hypotheses, and ask which of its parts could have moved. A smaller number here is not automatically good news. ``` ### Clarifying Questions - Is the 10% a fall in the number of rider cancellations or in the cancellation rate, cancellations as a share of ride requests? - Over what window was the fall measured, and against what baseline: the previous week, the same period last year, or a forecast? - Is the fall seen everywhere, or concentrated in certain cities, ride products or app platforms? - Which cancellations does the metric count: before a driver is matched, after matching, after the free cancellation window, or all of them? ### What a Strong Answer Covers - The metric pinned down as a count or a rate, with a recognition that a fall can be good or bad depending on what drove it - Data and instrumentation problems considered and ruled out early - A structured breakdown of where and when the fall occurs, checked in a sensible order - Internal and external hypotheses, each tied to the data that would confirm or rule it out - A decision about what to do once the cause is known, including the option of doing nothing ### Follow-up Questions - Suppose the cause is a change that charges riders for canceling sooner after a match. How would you decide whether to keep it? - If cancellations fell because fewer rides were requested, where would you look next? - How would you set up monitoring so a shift like this is caught and explained before someone asks about it?