Rocket Close and Amazon Web Services (AWS) launched a cloud-native mortgage document automation system on April 2, 2026, reducing abstract package processing time from about ten hours to under two minutes. The platform uses AWS Textract for optical character recognition (OCR) and Amazon Bedrock for large language model (LLM) analysis. The solution achieves approximately 90% classification accuracy and is designed to scale to more than 500,000 documents annually.1Rocket Close, AWS partner to automate mortgage document workflows
Background
Financial services firms-including mortgage lenders and fintechs-have traditionally relied on manual, fragmented processes to manage complex, multi-page document packages such as deeds, tax filings, and lien records. These abstract document packages, averaging 75 pages, require human experts to interpret inconsistent formatting, handwritten notes, and varied document types. This manual approach results in high labor costs, slow origination cycles, and increased risk of human error.1Rocket Close, AWS partner to automate mortgage document workflows
The mortgage industry is accelerating adoption of cloud computing, AI, and digital workflows to modernize operations. Automation platforms are replacing legacy point solutions, facilitating regulatory compliance through consistent and auditable processes, and enabling scalability for regional and national lending operations.2Google Cloud, UWM partner as mortgage battle revolves around automation, data, AI | Constellation Research
Details
The automation solution integrates Amazon Textract, which converts scanned mortgage documents into structured, machine-readable text, with Amazon Bedrock, which segments, classifies, and extracts data using custom LLM prompts. According to AWS and Rocket Close, the system processes about 2,000 abstract document packages daily, each averaging 75 pages.1Rocket Close, AWS partner to automate mortgage document workflows
Performance testing demonstrated strong results: the initial phase achieved 90.53% overall accuracy across 655 data fields; a second phase with 2,249 data fields recorded 91.28% accuracy; a third, large-scale evaluation encompassing more than 44,000 data fields showed 89.71% accuracy.3Rocket Close transforms mortgage document processing with Amazon Bedrock and Amazon Textract | Artificial Intelligence
The system improved processing speed by up to 15 times, reducing per-package processing from ten hours to less than two minutes. Rocket Close estimates the platform will eliminate around 1,000 hours of manual labor per day, enabling redeployment of staff to address complex exceptions.1Rocket Close, AWS partner to automate mortgage document workflows
Nathan Schrauben, Chief Information Officer at Rocket Close, stated that human experts remain integral to the workflow, verifying data and addressing exceptions to ensure the accuracy of home-ownership rights, while routine tasks are automated.1Rocket Close, AWS partner to automate mortgage document workflows Sri Elaprolu, Director at the AWS Generative AI Innovation Center, noted that the collaboration featured a discovery process with domain experts and iterative model refinement to embed mortgage-specific knowledge into the platform.1Rocket Close, AWS partner to automate mortgage document workflows
Outlook
Rocket Close intends to evolve the system from proof-of-concept to full production, extending its application to additional workflows, including loan processing, purchase agreements, and title clearance documentation. Ongoing updates and model enhancements are planned to support the continuous expansion of automation across business operations.1Rocket Close, AWS partner to automate mortgage document workflows
