# Rocket Close's 15x Speedup in Mortgage Processing via AWS Generative AI > Coverage of aws-ml-blog **Published:** April 02, 2026 **Author:** PSEEDR Editorial **Category:** enterprise **Tags:** Generative AI, Intelligent Document Processing, Amazon Bedrock, Amazon Textract, Machine Learning, Mortgage Tech **Canonical URL:** https://pseedr.com/enterprise/rocket-closes-15x-speedup-in-mortgage-processing-via-aws-generative-ai --- A recent case study from aws-ml-blog details how Rocket Close leveraged Amazon Bedrock and Amazon Textract to automate mortgage document processing, achieving a 15x speedup and 90% accuracy. In a recent post, aws-ml-blog discusses how Rocket Close successfully transformed its highly manual mortgage document processing workflow into an automated, highly scalable solution. Developed in partnership with the AWS Generative AI Innovation Center (GenAIIC), the new architecture leverages Amazon Bedrock and Amazon Textract to drastically reduce processing times and improve operational efficiency. The mortgage industry is notoriously document-intensive and heavily reliant on legacy processes. Processing abstract packages-which often contain dozens of pages of unstructured data, varying formats, and complex legal terminology-traditionally requires immense manual effort. This manual review cycle creates significant operational bottlenecks, limits scalability, and delays customer service. As financial enterprises look for ways to modernize these operations, intelligent document processing powered by machine learning and generative AI has emerged as a critical capability. The ability to accurately parse, classify, and extract data from unstructured documents is no longer just an operational upgrade; it is a competitive necessity. The aws-ml-blog publication details how Rocket Close addressed these industry-wide challenges head-on. Prior to this implementation, processing a single abstract package could take up to 10 hours of manual labor. By integrating Amazon Textract for robust optical character recognition (OCR) and Amazon Bedrock for access to advanced foundation models, Rocket Close built a pipeline capable of handling complex document segmentation, classification, and field extraction automatically. The scale of the operation is notable: the system currently processes approximately 2,000 abstract package files daily, with each package averaging 75 pages. The results of this implementation highlight a massive return on investment. The new automated workflow is reported to be 15 times faster than the previous manual process. Furthermore, despite the complexity and variability of mortgage documents, the system achieves a 90% overall accuracy rate across its extraction and classification tasks. Designed with future growth in mind, the architecture is built to scale and is projected to handle over 500,000 documents annually, ensuring sustainable business expansion without a linear increase in manual processing costs. For engineering leaders, data scientists, and enterprise architects evaluating generative AI for heavy document workflows, this case study provides a strong signal of viability. It moves beyond theoretical AI applications to showcase a practical, production-ready workflow using managed AWS services to solve a critical business problem. While the technical brief notes that specific details regarding the exact foundation models used and the granular algorithms for segmentation are not fully detailed, the operational outcomes speak volumes about the maturity of these tools. We highly recommend reviewing the complete case study to understand how these AWS services were orchestrated to achieve such significant efficiency gains. [Read the full post on aws-ml-blog](https://aws.amazon.com/blogs/machine-learning/rocket-close-transforms-mortgage-document-processing-with-amazon-bedrock-and-amazon-textract). ### Key Takeaways * Rocket Close automated its manual mortgage document workflow, achieving a 15x speedup over the previous 10-hour-per-package process. * The solution processes around 2,000 abstract packages daily, averaging 75 pages each, using Amazon Textract for OCR and Amazon Bedrock for foundation models. * The system delivers 90% accuracy across document segmentation, classification, and field extraction. * Designed for high scalability, the architecture is expected to handle over 500,000 documents annually. [Read the original post at aws-ml-blog](https://aws.amazon.com/blogs/machine-learning/rocket-close-transforms-mortgage-document-processing-with-amazon-bedrock-and-amazon-textract) --- ## Sources - https://aws.amazon.com/blogs/machine-learning/rocket-close-transforms-mortgage-document-processing-with-amazon-bedrock-and-amazon-textract