Design Dynamic Pricing System for Lyft: Key Features & Models
Overview
This question evaluates a candidate's ability in dynamic pricing and real-time machine learning system design, focusing on demand/supply forecasting, price elasticity modeling, multi-objective optimization, fairness and safety constraints, feature engineering from streaming and historical data, and online experimentation.
Lyft
Aug 4, 2025, 10:55 AM
Data Scientist
Technical Screen
Machine Learning
15
0
Scenario
You are designing Lyft's real-time dynamic-pricing system to jointly optimize rider experience and marketplace health. The system should adjust prices at fine spatial and temporal resolution while accounting for demand spikes, driver availability, and regulatory/fairness constraints.
Key outcomes to balance:
- Rider ETA (wait time)
- Rider conversion (request and completion probability)
- Driver earnings and utilization
- Company revenue/GMV
Task
Design a dynamic pricing algorithm. Describe:
- Features and data sources you would use in real time and historically.
- The modeling components and overall architecture (e.g., forecasting, elasticity, policy/optimization; tree models vs. RL/bandits).
- How you would balance trade-offs across objectives, including surge caps, fairness constraints, and safety guardrails.
- How you would evaluate and monitor the system online.
Notes
- Consider real-time demand/supply, location grid, time-of-day, historical price elasticity, driver proximity/ETA, weather/traffic, and special events.
- Discuss model choices such as gradient-boosted trees vs. RL/bandits, and when to use each.
- Include guardrails (e.g., surge caps, rate limits, regional fairness) and an experimentation/safety plan.
Solution
Login required
Show