At the airport, drivers must wait in a designated staging area to receive passenger ride requests. This is a product experience unique to airports and breaks from drivers' behaviors in a typical downtown setting.
One key business problem with this requirement is how to get drivers to wait for rides at the right times. If drivers sit in those staging areas when there is low demand, it's a waste of drivers' time and they are underutilized. And if drivers don't show up during high demand times, passenger waits are long and requests end up unfulfilled or canceled.
The question is: How do we get drivers to show up at the right times? Monetary incentives certainly work, but eat into the marginal profit per ride. So I looked towards equipping drivers with the right knowledge to make good decisions about their time and better balance supply and demand.
The original signal drivers' used was the size of the queue in numerical terms. But this is a misleading signal, because at times of high demand the queue may be 100 drivers long but turning over every 10 minutes.
The product I built to address this was a proper Queue Wait Time Estimate that would tell drivers how long they could expect to wait in the queue in real time, both when they are far from the airport to help aid in their decision to target the airport AND while they are waiting in the lot for a ride.
These wait time estimates were built using an ML model that processed:
1) The number of drivers in the queue
2) Average demand levels for the current time of day/day of week in historical data
3) Updated real time data on the status of flights to account for cancelations/weather disruptions/large regional events.