TLDR: Amazon Web Services (AWS) has introduced a new method for efficiently processing multi-page documents, combining the power of Amazon Bedrock Data Automation and Amazon SageMaker AI with essential human review loops. This innovative approach aims to deliver high accuracy and efficiency in extracting insights from unstructured content, ensuring data quality and compliance while minimizing manual effort.
Organizations across various sectors frequently encounter the challenge of processing vast quantities of multi-page documents to extract accurate information. While automation has significantly advanced these processes, the need for human expertise remains crucial for verifying data accuracy and quality in specific, complex scenarios. Addressing this, Amazon Web Services (AWS) has unveiled a robust solution that integrates Amazon Bedrock Data Automation with Amazon SageMaker AI to create an end-to-end system for intelligent document processing (IDP) that includes vital human review loops.
Launched in March 2025, Amazon Bedrock Data Automation empowers developers to automate the generation of valuable insights from diverse unstructured multimodal content, including documents, images, video, and audio. This service streamlines document processing workflows by automating extraction, transformation, and insight generation. It significantly reduces time-consuming tasks such as data preparation, model management, fine-tuning, prompt engineering, and orchestration through a unified, multimodal inference API. AWS highlights that this delivers industry-leading accuracy at a lower cost compared to alternative solutions.
Amazon Bedrock Data Automation simplifies complex document processing tasks, including splitting, classification, extraction, normalization, and validation. It also incorporates visual grounding with confidence scores for explainability and built-in hallucination mitigation, ensuring trustworthy insights from unstructured data. However, recognizing that human judgment is invaluable for certain scenarios, the integration with Amazon SageMaker AI forms a powerful solution. By embedding human review loops, organizations can validate AI predictions when confidence is low, effectively handle edge cases and exceptions, maintain regulatory compliance, ensure high accuracy while maximizing automation, and create feedback loops to continuously improve model performance.
The solution leverages confidence scores—the percentage of certainty Amazon Bedrock Data Automation has regarding extraction accuracy—to determine when human intervention is necessary. AWS aims to simplify IDP by handling the accuracy calculation within Bedrock Data Automation, allowing customers to focus on business challenges. The service optimizes its models for Expected Calibration Error (ECE) for more reliable confidence scores. Generally, high confidence is 90-100%, medium is 70-89%, and low is below 70%, with the latter typically triggering human verification. AWS recommends testing on specific datasets to determine optimal confidence thresholds.
The serverless architecture for this solution involves several key steps: Documents are uploaded to an Amazon S3 input bucket, triggering an AWS Step Functions workflow via an Amazon EventBridge rule. A Lambda function then invokes Amazon Bedrock Data Automation to extract and process information, storing the output, including confidence scores, in another S3 bucket. If confidence scores fall below predefined thresholds, another Lambda function routes the document to SageMaker AI for human review. Reviewers use a custom UI to validate and correct fields, with the corrected data stored back in S3, providing a final, high-confidence output for downstream systems.
Also Read:
- AWS Unveils Amazon SageMaker Unified Studio for Streamlined Generative AI Application Development
- Cybage and AWS Announce Strategic Collaboration to Propel Enterprise Generative AI Adoption
The solution was detailed by Prashanth Ramanathan, Andy Hall, Joe Morotti, and Vikas Shah. It is designed to be adaptable across various document types and customizable to meet specific business requirements, offering a significant advancement in document intelligence solutions on AWS.
