A bank deployed an XGBoost classifier in SageMaker to decide which customers qualify for a low-rate credit promotion. The model’s accuracy is acceptable, but the team must explain why particular customers were denied. What is the most operationally efficient way to produce per-customer explanations?

SageMaker Clarify is designed for exactly this purpose: providing explainability for machine learning models. It can compute SHAP (SHapley Additive exPlanations) values, which quantify the contribution of each feature to a specific prediction. Using the training data as a baseline allows Clarify to understand the typical feature distributions and calculate meaningful SHAP values for individual customer predictions, generating charts for easy interpretation. The other options are less efficient or incorrect. plotimportance() in XGBoost shows global feature importance, not per-prediction explanations. SageMaker Debugger is for debugging training jobs, not for post-deployment explainability. SageMaker Model Monitor tracks model quality and data drift, but it doesn't inherently compute SHAP values for individual predictions.