Lyft has invested in improving the customization of mode recommendations to better serve customers. The recommendation system is used to optimize marketplace demand and supply, providing recommendations to riders throughout the ride journey.
- Recent changes to the system include a one-tap option for high-frequency users, ranking and preselection of modes for easier decision-making, and post-request cross-sells to offer an upgraded ride experience.
- The recommendation system uses machine learning models to predict riders' propensity for converting into each mode and customizes rankings accordingly.
- To ensure newer offerings gain awareness and habituation, a post-processor layer was introduced to adjust machine learning model results.
- The system optimizes not just for ride reliability and supply/demand balance, but also for rider propensity, price, and wait time trade-offs.
- Upcoming improvements include expanding the one-tap module to include more relevant use cases, while Lyft's in-house contextual bandit system will enable more dynamic user-system interactions to be considered, such as longer-term user engagement behaviors.
