Optimizing Multiproduct Rideshare Pricing via Quasiconvex Efficiency Objectives

Pricing is the most consequential lever for financial performance in almost any business. At Lyft, analysts make financial projections and adjust the pricing policy to track them efficiently against evolving elasticities, supply-demand conditions, etc. This process could not scale to the 300+ markets that our business has grown to, relied on manual scenario planning that lacked automation, and used ad hoc, trial-and-error heuristics that led to suboptimal tradeoffs. Starting in 2024, we developed an optimization-based system that improved pricing efficiency for the Lyft rideshare marketplace by (1) automating the planning process for strategic pricing, and (2) introducing a daily policy update process that addresses the evolution of marketplace conditions over time. Powering this system is an algorithm that solves a quasiconvex program and generates ride price levels with the optimal tradeoff between different financial metrics for multiple product offerings or ride types, across all markets. The program optimizes the tradeoff between incremental financial metrics such as long-term bookings and number of rides as a quadratic-over-linear fractional objective subject to operational and market constraints. The price levels for every market are updated daily and scaled to the 300+ markets. Its direct business impact has been a 1% increase in bookings, amounting to well over $100 million in additional bookings per year. We also observed statistically significant reductions in pickup times and increases in ride acceptance rates. Crucially, the system enhanced analyst productivity and pricing observability over manual heuristics. It is interpretable, tunable, and grounded in actual analyst workflows. Further, it bridges the gap between long-term financial planning and short-term marketplace adjustments across multiple products, allowing analysts to increase their scope to explore more sophisticated price management strategies.

History: This paper has been accepted for the INFORMS Journal on Applied Analytics Special Issue—2025 Daniel H. Wagner Prize for Excellence in the Practice of Advanced Analytics and Operations Research.