Context

RideNow uses a gradient-boosted regression model to predict trip-level dynamic price multipliers and a companion calibration layer to bucket rider risk into 1-5 stars for internal quality review. The model went live 8 weeks ago across 12 U.S. cities. Since launch, finance has reported margin compression in some cities, while operations has seen rider complaints increase in others.

Current Performance

The Problem

Leadership wants a post-launch monitoring plan that can detect whether the model is drifting, miscalibrated, or failing in specific segments such as airport trips, peak commute hours, and bad-weather demand spikes. You need to define what should be monitored daily and weekly, how to interpret the current metrics, and what actions should trigger retraining, threshold changes, or rollback.

Requirements

1. Explain what the current metrics suggest about live model health.
2. Propose a monitoring framework covering prediction quality, calibration, business KPIs, and segment-level performance.
3. Identify likely root causes for the degradation and how you would validate them.
4. Recommend concrete interventions and alert thresholds.
5. Describe how you would balance revenue, rider conversion, and marketplace stability.

Constraints

• Price changes above 12% week-over-week require executive approval.
• Retraining can happen at most once per week.
• Airport and severe-weather segments account for 31% of revenue and cannot be disabled.