A recommendation model uses ML and calls an Amazon SageMaker AI endpoint to get recommendations.

Option A is correct because Amazon SageMaker AI supports automatic scaling for hosted models, and AWS documentation states that auto scaling dynamically adjusts the number of instances provisioned for a model in response to workload changes. When traffic increases, auto scaling brings more instances online; when traffic decreases, it removes unneeded capacity. This directly matches the requirement to keep the recommendation endpoint available during an expected increase in user traffic.

AWS also documents that SageMaker integrates with Application Auto Scaling, and you can scale SageMaker endpoint variants by using target tracking, step scaling, and scheduled scaling policies. That means the service is purpose-built for this scenario: maintaining availability and capacity as inference demand changes over time. Since the model is already being served from a SageMaker endpoint, the most direct and AWS-recommended solution is to configure endpoint auto scaling rather than redesigning the deployment.

The other options do not address the requirement as effectively. Creating a new endpoint does not automatically handle fluctuating traffic and adds unnecessary operational complexity. SageMaker Neo optimizes models for inference performance on supported hardware, but it does not provide elastic capacity management. Attaching an Auto Scaling group to a SageMaker endpoint is incorrect because SageMaker endpoints do not use EC2 Auto Scaling groups directly; scaling is handled through SageMaker’s integration with Application Auto Scaling. Therefore, to maintain availability during traffic spikes, the best AWS-documented answer is A: configure auto scaling on the SageMaker AI endpoint.