A recommendation model for an ecommerce site must use all client interaction data as input. Traffic volume varies throughout the day. Which type of SageMaker inference endpoint is the most cost-effective for this usage pattern?

Serverless inference endpoints are the most cost-effective for this scenario because they automatically scale compute capacity up and down based on traffic, meaning you only pay for the actual inference duration and data processed. This is ideal for variable traffic patterns, as it avoids over-provisioning resources during low-traffic periods. Batch transform is for offline processing of large datasets, not real-time or near real-time recommendations. Asynchronous inference is suitable for requests with large payload sizes or long processing times, but still incurs costs for idle provisioned instances. Real-time inference endpoints provision dedicated instances, leading to higher costs during low traffic periods compared to serverless options.