Problem
A ride-hailing marketplace must price every ride request so that supply and demand balance and costs are covered. Drivers join and leave on their own, which moves pickup times, cancellations and rider conversion, so the right price keeps changing. The demand response to a price change is hard to estimate and non-linear across ride options. Riders also expect a stable price while they compare options, and equal ride intents must get equal prices.
Example from the article: Lyft prices rides to be reliable (consistent prices and pickup times across repeated use), fair (same price for the same ride intent unless explained, e.g. Lyft Pink membership) and competitive.
Solution in brief
Split pricing into three layers: an orchestration platform that serves and records prices, operations that give guidance, and an algorithmic layer. The algorithm sets a structural price per ride and adds a dynamic supply and demand adjustment. Off-policy evaluation warm-starts prices, then online reinforcement learning updates the policy from live transactions.
Questions this case answers
- How to set prices in a two-sided marketplace when the demand response to price is hard to estimate and changes over time
- How to update a decision policy from live transactions without losing price stability and consistency for users
- How to set ride prices that stay consistent for riders while balancing supply and demand?
- How to keep prices current when the market is volatile and demand response to price is non-linear?
Tools and methods
- Structural pricing — base price per ride from location, time, distance and ride type
- Dynamic pricing — balances riders and drivers across the network
- Off-policy evaluation — scores a policy from data logged under another policy
- Online reinforcement learning — updates prices in near real time from outcomes
- Cache layer and price quote store — keep quotes stable and log every decision
Google or a one-time email code. No password.