Solving the ride-sharing productivity paradox : Priority dispatch and optimal priority sets

Ride-sharing platforms face a “productivity paradox”, where any efficiency gained through improved dispatch or pricing strategies will not benefit drivers or riders. We show that this is a limit of the traditional ride-hailing model, and a consequence of the Hall-Horton driver equilibrium earning hypothesis [1]. In response to this challenge, Lyft introduced Priority Mode, which allows drivers to concentrate their work during specific, prioritized hours. We prove that Priority Mode solves the productivity paradox, and is able to increase the average driver earnings, while also benefiting the platform and the riders. Implementing Priority Mode requires important changes to the platform’s dispatch and pricing policy, but most importantly requires careful control of the number of drivers that can be offered the opportunity to be prioritized at any given time. We introduce a queuing setting to model the market dynamics of priority mode and illustrate the challenges of this control problem. We then leverage this intuition to build a real time priority admission control system that is able to balance the number of drivers offered priority and achieve the desired productivity increase. Priority Mode has been successfully rolled out to various cities throughout North America, with hundreds of thousands of driving hours so far. It generates tens of millions of dollars of value that is shared between the drivers, the riders and Lyft, with the potential to generate much more when rolled out in all markets. Finally, our internal driver surveys reveal that it was well received