A digital media company has developed a content recommendation engine with Amazon SageMaker, leveraging a custom inference model to enhance user engagement. After developing an improved version of the model based on offline tests, the team aims to evaluate its performance with actual users while ensuring minimal disruption and risk. What strategy should the team employ to deploy the updated model to production, allowing for real-world assessment with the least operational impact?