Drift & monitoring

Data drift rate (per month): 1.2%

Enable monitoring + auto-retrain

Retrain trigger threshold: 8% accuracy drop

Real-world data distributions shift over time (new customer behaviour, seasonal patterns, a competitor's product launch). A model trained once on old data slowly becomes miscalibrated for the world it's now scoring — this is drift.

Status at month 36

Current accuracy—

Retrains triggered—

Months below acceptable accuracy—

Model accuracy over 36 months (▼ = retrain event)