Huntington Bank redacts 400M+ documents in months using AWS AI pipeline

Huntington National Bank, a top 10 U.S. bank, faced a massive compliance challenge: redacting sensitive customer data from over 400 million documents stored on-premises since 2015. Initial estimates indicated the effort would take years. By designing a scalable redaction workflow using Amazon Textract (ML-based document text extraction), AWS Step Functions (orchestration), AWS Lambda (serverless compute), and Amazon SageMaker (ML model hosting), the bank reduced the timeline to just months. The solution also leveraged AWS DataSync and AWS Direct Connect for secure, encrypted data transfer, and AWS KMS for encryption key management. Documents were processed in a distributed map state in Step Functions to maximize throughput against Amazon Textract service quotas, with monitoring via Amazon CloudWatch dashboards.

Key requirements included encryption at rest and in transit, PCI DSS compliance, on-premises output replication, and redaction accuracy of at least 95%. The architecture uses Amazon Textract to detect sensitive fields like Social Security numbers and account numbers from scanned documents, outputting JSON with bounding box coordinates. By running concurrent jobs within adjustable service quotas and using Step Functions' map state for parallel processing, Huntington achieved processing rates of millions of documents per day. The solution reduced manual review time and improved accuracy, demonstrating how serverless AWS services can handle extreme-scale document processing while maintaining security and compliance.