What challenges did Huntington National Bank face before AI implementation?

Redact sensitive customer data across hundreds of millions of documents stored on-premises. Meet strict PCI DSS and access-control requirements. Avoid a timeline that would have taken years with the original approach.

What was the impact of the AI solution?

Processed documents at a rate of approximately 10 million per day. Reduced the estimated processing time from years to just a few months. Reduced total processing cost to approximately 5% of the original estimate. Redaction accuracy exceeded 95%. The solution met compliance requirements and supported data security objectives.

Source

Scaled productionEvidence: Medium50/100

Huntington Bank: Redacting sensitive data from 400M+ documents with AWS

Use case typeDocument processing automationUpdated Jun 24, 2026

Huntington National Bank used AWS services to redact sensitive customer data across a repository of more than 400 million on-premises documents. The solution moved files into Amazon S3, used Amazon Textract and AWS Step Functions to detect and process sensitive fields at scale, and then replicated redacted outputs back to on-premises storage. The program was designed to meet strict PCI DSS and access-control requirements while handling varied document formats.

Organization: Huntington National Bank
Industry: Finance
Location: United States
Published: June 2026

Reported outcomes

Documents processed per day: Approximately 10,000,000 documents/day

Automation & deflection

Planned next steps

The source says the organization aims to achieve Processing timeline: −83.3%.
The source says the organization aims to achieve Processing cost: Approximately 5 % of original estimate.

Why do we believe this?Outcome claims, sources, and evidence checks

Normalized claim

Documents processed per day: 10,000,000 documents/day increase

AWS Machine Learning BlogJun 24, 2026Blog postExplicit claimMedium evidence strength

Using AWS, Huntington processed documents at a rate of approximately 10 million per day

Normalized claim

Processing timeline: 83.3% decrease

AWS Machine Learning BlogJun 24, 2026Blog postInferred claimMedium evidence strength

reducing estimated processing time from years to just a few months

Normalized claim

Processing cost: 5 % of original estimate decrease

AWS Machine Learning BlogJun 24, 2026Blog postExplicit claimMedium evidence strength

The cost of processing the entire document repository was approximately 5% of the original estimate.

Normalized claim

Redaction accuracy: 95% increase

AWS Machine Learning BlogJun 24, 2026Blog postExplicit claimMedium evidence strength

Redaction accuracy exceeded 95%, meeting compliance requirements and supporting data security objectives.

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks

Customer: Huntington National Bank
Provider: AWS
Maturity: Scaled Production
Linked source: AWS Machine Learning Blog

The solution moved files into Amazon S3, used Amazon Textract and AWS Step Functions to detect and process sensitive fields at scale, and then replicated redacted outputs back to on-premises storage

Customer identity supportedSource describes one deploymentMaturity supported

Primary read

Use case focus

Showing 2 of 2

1Document processing automation
2Workflow automation

Migrated documents from on-premises storage to Amazon S3 using AWS Direct Connect, AWS DataSync, and AWS Key Management Service.
Used Amazon Textract in an orchestrated AWS Step Functions workflow to detect sensitive data and write metadata to Amazon S3.
Scaled processing with Step Functions map state and CloudWatch monitoring to handle very high concurrency and then replicated redacted outputs back on-premises.

Technologies

Amazon Textract AWS Step Functions Amazon S3 AWS Direct Connect AWS Key Management Service AWS DataSync AWS CloudWatch

Processed documents at a rate of approximately 10 million per day.
Reduced the estimated processing time from years to just a few months.
Reduced total processing cost to approximately 5% of the original estimate.
Redaction accuracy exceeded 95%.
The solution met compliance requirements and supported data security objectives.

Architecture

Huntington moved documents from on-premises file shares into Amazon S3 using AWS Direct Connect, AWS DataSync, and AWS Key Management Service. AWS Step Functions orchestrated large-scale Amazon Textract jobs using a map state for high concurrency, with CloudWatch used to monitor throughput and throttling. Detected fields and metadata were written to S3, then redacted files were synced back to on-premises storage with DataSync.

Sources & evidence1

Evidence: Medium50/100Evidence strength