A company has a model deployed to a SageMaker real-time endpoint and needs to deploy a new model, but must compare the new model’s performance to the current model before shifting all traffic. Which approach accomplishes this with the least operational effort?

Deploying the new model as a shadow variant on the same endpoint is the most operationally efficient solution. SageMaker endpoints support A/B testing and shadow testing natively, allowing you to deploy multiple model variants and distribute traffic based on predefined weights or send a copy of the production traffic to a shadow variant for evaluation without impacting live users. This built-in functionality minimizes the need for custom infrastructure or complex traffic routing logic. Deploying to a separate endpoint and manually splitting traffic or using CloudFront requires managing multiple endpoints and external routing, increasing operational overhead. Using AWS Lambda for custom routing adds complexity and maintenance burden compared to SageMaker's native capabilities.