PRODUCTION SCENARIO
A utility company scores 12 million smart-meter records once each night to flag likely billing errors. No application calls the model during the day. The company wants the most cost-effective way to run this workload.
Which Amazon SageMaker AI option will meet these requirements?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Run a batch transform job over the records that are in Amazon S3.
Batch transform gets inferences from large datasets when no persistent endpoint is needed: SageMaker AI starts instances, reads from Amazon S3, writes the output back, and releases the instances. Asynchronous inference takes one invocation per payload staged in Amazon S3, so a 12 million-record run means 12 million invocations against an endpoint the company still operates. A serverless endpoint bills 12 million invocations. Scheduling a real-time endpoint still leaves an endpoint to operate and instance hours to pay for.