Case Study

Mortgage Credit Advisor Transforms Manual Loan Risk Review to Automated Categorization with Claude on AWS Bedrock

At a glance

A mortgage credit advisory firm managing a multi-billion-dollar portfolio of residential whole loans, servicing rights, and RMBS relied on analysts manually reading loan servicing comments to flag risk issues and correct debt-to-income figures, capping how many new clients the firm could onboard.

NMD built a generative AI pipeline on AWS Bedrock that reads loan comments, matches them to a risk category taxonomy, and returns corrected adjustment fields automatically, funded entirely through AWS Activate.

The firm can now replicate this categorization service for each new servicing client it signs, turning a single internal workflow into a repeatable offering. At the firm’s internally modeled per-customer AWS cost, this supports onboarding 5 to 10 new customers in year one, a pipeline that did not exist before this engagement.

Industry

Use Case

Intelligent Document Processing, Compliance & Governance, Financial Analysis

Solution implemented

The value equation

Company Snapshot

A LegalTech-adjacent capital markets firm — a $2B+ AUM, multi-billion-dollar mortgage credit advisor specializing in residential whole loans, mortgage servicing rights, and RMBS — used generative AI to automate risk categorization of loan servicing comments across 100+ funded transactions.

Location

United States

Customer Situation

A mortgage credit advisory firm prices residential whole loans and servicing rights across a multi-billion-dollar AUM portfolio built on 100+ funded transactions. Each loan file includes free-text servicing comments that analysts read manually to identify risk issues and correct attributes like debt-to-income ratio, directly affecting how a loan is priced.

That manual review created a ceiling on throughput. As the firm looked to extend loan analysis to new servicing clients, it needed a way to categorize comments at scale without adding headcount per customer.

Across the broader mortgage servicing and whole loan market, unstructured servicing comments remain a common bottleneck: any credit advisor or servicer scaling beyond a single portfolio faces the same manual review wall this firm did.

NMD Solution

NMD reviewed the client’s loan comment workflow and found that risk categorization and debt-to-income corrections were fully manual and tied to a single analyst’s read of free-text notes. The solution required an AWS Bedrock large language model to match each comment against a defined risk category and adjustment taxonomy, with Amazon Textract for document parsing, AWS Glue for data preparation, and Amazon S3 and DynamoDB for structured storage and retrieval.

What We Delivered

NMD delivers an API-based pipeline that ingests loan comment data, matches each entry to a defined risk category and adjustment field, and returns corrected loan attributes for underwriting and pricing review. The solution runs on AWS Bedrock with Textract, Glue, S3, and DynamoDB, and the client’s team validated output against their existing risk taxonomy before signing off on the proof of concept.

Ready to Transform Your Financial Services Business?

See how New Math Data can transform your financial services business with AWS-powered innovation.