Interview Questions › Metrics

How would you measure the quality of Google Maps ETAs?

You own arrival-time estimates (ETAs) in Google Maps driving navigation. How would you measure whether ETAs are good?

Context

ETAs are shown before a trip and updated during it. Users complain most when they arrive much later than promised.

Clarifying questions to ask first

• Initial ETA or updated ETA?
• Which trip types and regions?

How to approach it

1. Define accuracy (predicted vs actual arrival)
2. Asymmetric cost: being late hurts more than early
3. Segment by trip length, city and traffic
4. User-facing outcome: trust and navigation usage

Framework: Goal of the product -> Users and value exchange -> North Star -> Supporting metrics by funnel stage -> Guardrails and counter-metrics -> Decision or target

Sample answer

• Share of trips arriving within 10% of the initial ETA
• Share of trips arriving more than 10% late
• Absolute error by trip-length bucket

1. About 82% of trips land within ±10% of the initial ETA
2. Late trips outnumber early trips 2 to 1
3. Error is worst on trips over 45 minutes in monsoon months in Indian cities

Primary: share of trips within ±10% of the initial ETA, weighted toward late misses because they cost more trust. Track by trip length and city. Outcome guardrail: navigation sessions per user.

What interviewers look for

• Share of trips within ±10% of ETA
• Late arrival rate
• Navigation starts per user