What is data drift and how do you detect it?

Data drift is when production input data diverges from the training distribution over time (e.g. user behaviour shifts, a new product launches), degrading model performance. Concept drift is when the input-output relationship itself changes. Detect it by monitoring feature distributions and comparing them to training (statistical tests like PSI or KS), tracking prediction distributions, and watching live accuracy where labels are available. Alerts on drift trigger investigation and retraining.