Metrics question

Hypothetically, there is a data point that indicates that there are more Uber drop-offs at the airport than pick-ups from the airport. Why is this the case and what would you do within the product to change that?

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Root cause analysis paired with a product response, checking you can move from a data anomaly to a hypothesis to an action.

State the observation precisely: more drop-offs than pick-ups at the airport, meaning a net outflow of riders arriving via Uber but not leaving via Uber.
Generate hypotheses: some pick-ups may fail to convert to completed trips due to surge pricing or long wait times at the airport, or riders use alternatives like taxis, ride-hailing rivals, or pre-booked transport for departures.
Consider a structural explanation: business travelers dropped off by a company car or personal vehicle, then leaving by Uber for the return, would skew the ratio the other way, so verify the direction of the imbalance carefully.
Prioritize the hypothesis most testable with existing data, like comparing wait times and cancellation rates for airport pick-ups versus other trips.
Propose a product response tied to the most likely cause, for example a dedicated airport pick-up queue or pre-scheduled pick-up feature to reduce wait-time related cancellations.
State how you would validate the fix, comparing pick-up completion rate before and after the change in a controlled test.

Generates multiple distinct hypotheses, not just one, including both demand-side (alternatives) and supply-side (wait time, surge) explanations.
Proposes a specific, testable product fix, like a scheduled pick-up slot system, tied directly to the most likely cause.
Notes how to validate the hypothesis with existing data before building anything, for example checking airport cancellation rates.
Shows awareness that correlation in the raw numbers could have a mundane explanation, like data collection quirks, worth ruling out first.

Jumping to a solution before generating and testing hypotheses for the root cause.
Assuming a single cause without considering both demand and supply side explanations.
Not proposing a way to validate the eventual fix.

How would you validate which hypothesis is the real driver before building a fix?
What data would you pull first to investigate this?