What challenges did Onity Group face before AI implementation?

Process millions of pages across hundreds of mortgage document types with dense legal text, inconsistent handwritten entries, and notarization/seal verification needs. Traditional OCR and AI/ML solutions struggled with accuracy and cost.

What was the impact of the AI solution?

Reported 50% reduction in document extraction costs. Reported 20% improvement in overall accuracy versus the previous OCR and AI/ML solution. Credit report processing achieved accuracy up to 85%. Appraisal checklist review improved accuracy by 65% over manual review.

Source

Automating complex document processing: How Onity Group built an intelligent solution using Amazon Bedrock

Use case typeIntelligent document processingUpdated Jun 13, 2026

Onity Group, through PHH Mortgage Corporation and Liberty Reverse Mortgage, processes millions of pages across hundreds of mortgage document types each year. The company built an intelligent document processing workflow that uses Amazon Textract for text extraction and Amazon Bedrock foundation models for complex visual and contextual analysis, including notarization verification, rider extraction, appraisal checklist validation, and credit report parsing. Documents are uploaded to Amazon S3 and routed through custom classification and extraction logic that chooses between Textract and Bedrock based on document complexity, with security controls using AWS KMS and AWS IAM.

Organization: Onity Group
Industry: Finance
Location: United States
Published: May 2025

Reported outcomes

85%

accuracyQuality & accuracy

−50%cost+20%accuracy+65%accuracy

Strategic outcomes

New product / capabilityBuilt intelligent document processing workflowNew product / capabilityEnabled notarization verification and complex parsingScale & capacityProcessed millions of mortgage documentsRisk & complianceAdded security controls for document processing

Primary read

Use case focus

Showing 3 of 4

1Document Processing
2Intelligent Automation
3OCR

Process millions of pages across hundreds of mortgage document types with dense legal text, inconsistent handwritten entries, and notarization/seal verification needs.
Traditional OCR and AI/ML solutions struggled with accuracy and cost.

Built an intelligent document processing workflow that uploads documents to Amazon S3, extracts content with Amazon Textract, classifies documents with a custom AI model, and dynamically routes extraction tasks between Textract and Amazon Bedrock foundation models based on content complexity.
Used Bedrock text and vision models for tasks requiring contextual or visual understanding, such as notarization verification and complex form parsing.
Stored extracted information in structured formats for downstream processing.

Technologies

Amazon Bedrock Amazon Textract Amazon S3 AWS KMS AWS IAM

Reported 50% reduction in document extraction costs.
Reported 20% improvement in overall accuracy versus the previous OCR and AI/ML solution.
Credit report processing achieved accuracy up to 85%.
Appraisal checklist review improved accuracy by 65% over manual review.

Architecture

Documents are uploaded to Amazon S3, preprocessed, extracted with Amazon Textract, classified with a custom AI model, and then dynamically routed to Amazon Textract or Amazon Bedrock text/vision foundation models depending on the document type and extraction complexity. Output is stored in operational databases and Amazon S3. Security controls include AWS KMS and AWS IAM.

Sources & evidence1

Groundedness: 5/5Type: Blog PostPublished: May 20, 2025Publisher: AWSEvidence: VendorConfidence: High

Primary source

AI-generated summary. Verify important details with the linked sources before relying on this case.

Explore related AI use cases

Customer evidence