How do you handle ambiguous product metrics and root-cause analysis when rider cancellations spike in a specific city at Uber Drivers?
