Metrics question

How would you determine sleep from cellphone data?

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

Inferring a hidden behavioral signal, sleep, from indirect device data, combining creativity with metric rigor.

List available signals: screen off duration, charging patterns, ambient light or sound if sensors are used, and app usage gaps.
Define a proxy metric, such as the longest continuous period of no screen interaction overnight.
Validate the proxy against a ground truth, like a small user study comparing inferred sleep windows to self reported sleep logs.
Account for edge cases, like phones left in another room or used briefly during the night.
Refine the model iteratively, weighting signals like charging start time and motion data if available.

Proposes a specific, testable proxy, like screen off duration between 10pm and 8am, rather than a vague analyze usage plan.
Insists on validating the proxy against real sleep data before trusting it, showing rigor.
Flags a concrete edge case, like shared devices or phones left in another room, that would bias the estimate.

Assuming screen off time alone perfectly equals sleep without validating against ground truth.
Ignoring edge cases like users who do not sleep near their phone.

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