Root-Cause Analysis of Inaccurate Wait Estimates in Yelp's Restaurant Waitlist

You are a product manager at Yelp working on its restaurant waitlist feature, which tells diners how long they can expect to wait for a table. Diners have started posting on X (formerly Twitter) that the wait times the feature gives them are wrong. How would you investigate what is going on? ```hint Turn posts into a measurement Social posts tell you that something feels wrong, not how often or where. Work out how you would compare each party's quoted wait with the wait it actually had, and which cuts of that data would separate a problem in the estimate from a problem in what restaurants or diners do. ``` ### Clarifying Questions - Are diners saying the waits run longer than quoted, shorter than quoted (so they miss their turn), or both? - Did the complaints start recently, for example after a release, or have the estimates always been loose and only now drawn attention? - Is the problem the estimate shown before a diner joins the list, or the updates they receive after joining? - Are the posts concentrated in particular cities, restaurants or times, such as weekend dinner rushes? ### What a Strong Answer Covers - Turning social complaints into a measurable accuracy problem: how the actual wait is captured, how error is defined, and what baseline it is compared with. - Establishing whether the problem is real, new and widespread, including whether it lines up with releases or data changes, before diagnosing causes. - Hypotheses spanning the estimation logic, the inputs hosts and diners provide, and restaurant operations, narrowed through segmentation. - Fixes and short-term mitigations matched to the confirmed cause, including how affected diners and restaurants are told. - Metrics and guardrails that show the fix worked, such as accuracy, completion of the waitlist journey and restaurant adoption. ### Follow-up Questions - The estimates are accurate on average but far off for some parties. What would you change? - Hosts often forget to mark parties as seated. How would you get a trustworthy measure of the actual wait? - Would you show a range instead of a single number, and what could go wrong for diners and restaurants if you did?