In governance terms, which statement describes model performance drift?

• A. The sudden, large drop in performance due to a single data shift.
• B. The time to retrain a model after deployment.
• C. The budget variance during maintenance.
• D. The gradual change in model accuracy over time on new data.

Answer: (D)

Performance drift is about the model’s accuracy slowly changing as new data come in. In production, the patterns the model learned during training can shift over time due to changing user behavior, new subpopulations, or evolving data distributions. This gradual drift means the model may become less accurate on current data, prompting monitoring and possible retraining. It’s not a one-off, large drop caused by a single data shift—that would be a different issue—and it’s not about how long it takes to retrain or how the budget varies. So the statement that best describes drift is the gradual change in model accuracy over time on new data.