Whenever you're ready, walk me through how you'd approach this.
How do you monitor for data drift and concept drift in a live model at Intuit?
Explain how you would instrument the production pipeline, distinguish input distribution changes from changes in the feature-label relationship, and respond when labels are delayed. Cover detection methods, thresholds, alerting, retraining or rollback decisions, and how you would prevent training-serving skew.