Onity Group Cuts Document Extraction Costs 50% and Improves Accuracy 20% with Amazon Bedrock AI

Title

Onity Group Cuts Document Extraction Costs 50% and Improves Accuracy 20% with Amazon Bedrock AI

Content

Onity Group, the West Palm Beach, Florida-based mortgage servicer operating through PHH Mortgage Corporation and the Liberty Reverse Mortgage brand, deployed an intelligent document processing solution on Amazon Bedrock that reduced document extraction costs by 50% and improved overall accuracy by 20% compared to its previous OCR and AI/ML pipeline. Specific use cases reached approximately 85% accuracy on credit report processing and a 65% improvement on home appraisal checklist review. "We needed a solution that could evolve as quickly as our document processing needs," says Raghavendra (Raghu) Chinhalli, VP of Digital Transformation at Onity Group. "By combining AWS AI/ML and generative AI services, we achieved the perfect balance of cost, performance, accuracy, and speed to market," adds Priyatham Minnamareddy, Director of Digital Transformation & Intelligent Automation. Onity processes millions of pages across hundreds of document types annually — dense legal texts such as deeds of trust, inconsistent handwritten entries, and notarization and legal seal verifications — tasks traditional OCR and ML models struggled with. Verbose documents, inconsistent formats (e.g. "GA" vs "Georgia", "200K" vs "200,000"), and limited contextual understanding drove the search for a more sophisticated solution. Onity began with Amazon Textract, then layered Amazon Bedrock FMs for complex visual and textual reasoning. The dual-model architecture dynamically routes tasks between Textract and Bedrock by content complexity: custom AI models handle high-confidence classifications, while Claude Sonnet or Amazon Nova Pro take on untrained document types. The five-stage pipeline ingests documents to Amazon S3, applies preprocessing (enhancement, noise reduction, layout analysis), runs three-step classification orchestrated by Onity (Textract extract → custom AI model → Claude Sonnet in Bedrock if unrecognized), executes extraction with algorithm-driven routing between Textract and Bedrock FMs, and persists output to operational databases and Amazon S3. Data is encrypted with AWS KMS at rest and TLS in transit, aligned with the AWS Well-Architected Security Pillar and FSI Lens. The system handles hundreds of document types in production at millions of pages annually — deed of trust extraction, notarization info, rider extraction with checkbox detection, home appraisal review (65% improvement), and credit reports (85% accuracy across Equifax Beacon 5.0, Experian Fair Isaac V2, FICO Risk Score Classic 04). Onity continues to expand the dual-model approach to additional document types, leveraging Bedrock's flexibility to select the FM that best balances accuracy, performance, and cost — positioned as a reference implementation for intelligent document processing in mortgage servicing on AWS.