A company uses Amazon Textract to extract text from thousands of scanned legal documents daily. Documents that fail business validation are routed back to human reviewers, delaying loan processing. What should the company do to reduce overall loan processing time?

Amazon Augmented AI (A2I) is designed to streamline human review workflows for machine learning predictions, including Textract. By configuring Textract to send low-confidence words to A2I, human reviewers can correct these specific extractions before the full document proceeds to business validation. This proactive correction reduces the number of documents failing validation, thereby decreasing the overall loan processing time. SageMaker Ground Truth is primarily for building and labeling datasets, not for real-time human review of live inference results. Using synchronous operations would not address the accuracy issue and might introduce rate limiting. Amazon Rekognition's text detection is less specialized for document processing than Textract and would not improve accuracy in this scenario.