Diagnosing and reducing a global rise in Uber peak-hour ride cancellations
Company: Uber
Role: Product Manager
Category: Analytics & Experimentation
Difficulty: medium
Interview Round: Other
Uber has noticed that more rides are being cancelled than before, and the rise is sharpest during peak hours. The problem is not confined to particular cities or regions. As the product manager, how would you work out what is driving the increase, and what would you do to bring cancellations back down?
```hint Split before you solve
A cancellation can come from either side of the trip and at several moments between the request and the pickup. Break the number apart first, and ask what is different about peak hours.
```
### Clarifying Questions
- Is the increase coming from rider cancellations, driver cancellations or both, and is it measured as a count or as a share of requests?
- Did the rise appear suddenly or build up gradually, and did anything change around then, such as an app release, a change to cancellation fees or windows, or a change in how cancellations are logged?
- Which ride types are affected: standard rides, shared rides, or all of them?
- Has peak-hour demand grown over the same period, and has driver supply kept pace?
### What a Strong Answer Covers
- Confirming the problem is real (metric definition, logging, rate versus count) before diagnosing it.
- A breakdown by who cancels, at which stage between request and pickup, ride type, and peak versus off-peak.
- Hypotheses tied to what makes peak hours different, such as supply falling behind demand, long pickups in traffic, unreliable wait estimates, drivers weighing offers across apps, and hard pickups, each with the data that would confirm or rule it out.
- Prioritized fixes for both riders and drivers, mapped to confirmed causes, with a test plan that accounts for riders and drivers sharing one marketplace.
- Success metrics with guardrails, and an explanation of why the rise shows up everywhere rather than in one region.
### Follow-up Questions
- The data shows most of the increase comes from drivers cancelling after they see where the pickup is. What would you build, and what could go wrong?
- Raising cancellation fees would quickly cut rider cancellations. Why might you hold off?
- How would you show that a fix reduced cancellations without simply turning more requests into unfulfilled ones?
Overview: Product diagnosis question about a worldwide rise in Uber ride cancellations that is sharpest at peak hours. It tests validating the metric, splitting rider and driver cancellations across the trip funnel, forming peak-specific hypotheses, and proposing tested fixes with fulfillment and guardrail metrics.