A retailer is collecting purchase records from 20,000 stores into Amazon S3 via Amazon Kinesis Data Firehose. Daily model training requires simple attribute transformations and some field combinations. Which change requires the least development effort to apply these simple transformations before training?

The correct answer is to attach an Amazon Kinesis Data Analytics application downstream of the Firehose stream. Kinesis Data Analytics for Apache Flink allows for real-time processing and transformation of streaming data using SQL or Java/Scala. This approach directly integrates with the existing Kinesis Data Firehose pipeline, enabling transformations before data lands in S3, which is efficient and requires minimal development effort for simple SQL-based transformations. Running an Amazon EMR cluster or a fleet of Amazon EC2 instances would involve batch processing after data is stored in S3, adding latency and requiring more operational overhead and development for scheduling and managing these compute resources. Having stores switch to AWS Storage Gateway is a significant architectural change to the data ingestion method, not a transformation solution, and would involve substantial effort and disruption.