Example use cases
Transaction volume trend monitoring: An e-commerce platform processes ~50,000 orders on weekdays and ~30,000 on weekends. A behavioral metric learns the weekly pattern and alerts when volume drops more than 20% below the expected level for that day of the week — something a static threshold would miss on Sundays when lower volume is normal.
ML feature drift detection: A propensity_score feature historically ranges between 0.3–0.7 with a stable mean. After a code change to the scoring logic, the mean drifts to 0.15. A behavioral metric on the Mean of propensity_score alerts on the trend break before the degraded feature reaches production model inference.
Behavioral metrics are most effective for signals with predictable seasonal or cyclical patterns — volumes that peak on certain days, batch loads that always complete within a window, or revenue that follows a monthly growth trend. Use Independent mode for single-segment monitoring and Correlated mode when dimensions interact and isolated checks would miss localized anomalies.
