A telecom company deployed a model trained on several years of data from one region to identify customers likely to cancel when they call support. The company will roll out a global product and fears the model may wrongly route many calls from regions without training data to a specialist team. What approach most effectively addresses this risk?

The correct approach is to use SageMaker Model Monitor. The core problem is data drift: the new regional data may differ significantly from the original training data, causing the model to perform poorly. SageMaker Model Monitor is designed specifically to detect this by capturing inference data, comparing it to a baseline created from the training data, and alerting when statistical drift occurs. This allows the company to identify when the model's performance is likely degrading due to new data patterns and then retrain with a more diverse dataset. SageMaker Debugger is for debugging training jobs, not monitoring deployed models. Storing calls and monitoring true/false positives/negatives is reactive and requires ground truth, which isn't immediately available for new regions. CloudWatch logging alone doesn't provide the statistical drift detection capabilities of Model Monitor.