Estimation question

How would you identify all the potholes in the United States?

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

Estimation combined with practical systems thinking about data collection at scale.

Clarify the question is really two parts: estimating the total number of potholes, and designing a method to identify them.
Estimate total US road miles using a general known figure, then assume an average pothole density per mile based on road condition surveys.
Multiply road miles by assumed potholes per mile to get a rough total count, stating both inputs as explicit assumptions.
Propose an identification method: using dashcam or phone accelerometer data from ride-hailing or mapping app users to detect vibration signatures consistent with potholes.
Address validation: crowdsourced reports (like Waze) combined with computer vision on existing street-view imagery to confirm and geolocate them.
Note the trade-off between coverage speed (crowdsourced) and accuracy (dedicated sensor sweeps).

Splits the answer clearly into the estimation part and the practical detection-system part, addressing the full question.
States both the road-mileage and pothole-density assumptions explicitly, e.g. 4 million miles of road, 2 potholes per mile, marked as assumptions.
Proposes a concrete, existing-technology detection method, accelerometer data from connected vehicles, rather than an abstract idea.
Flags the trade-off between crowdsourced speed and dedicated-sensor accuracy as a real design decision.

Answering only the estimation half and ignoring the 'how would you identify them' half of the question.
Proposing manual road inspection at national scale without considering cost or feasibility.

Chapters of the AI PM course that teach what this question tests.