CLIENT DETAILS
Mid-sized Mortgage Lender Focused on Affordable Housing
The client is a Texas-based, full-service mortgage lender licensed in over 30 U.S. states. With a strong emphasis on affordable housing, the company partners with community-based lenders and credit unions to offer government-backed loans such as FHA, VA, and USDA, along with several proprietary assistance programs customized for first-time homebuyers and borrowers.
The company manages everything from loan origination to correspondent loan purchase, including document management, borrower verification, and compliance workflows. Known for its flexible loan programs and nationwide partnerships, the client plays a crucial role in providing access to sustainable homeownership for low- to moderate-income segments.
As a certified GNMA-approved issuer and seller/servicer for multiple federal loan programs, the lender is subject to stringent documentation and regulatory requirements. This creates operational complexity that demands accuracy, quick turnaround times, and strong workflow visibility.
BUSINESS NEED
Automate Loan Document Indexing, Versioning, and Exception Handling
The client needed a solution to handle borrower documents that often arrive as bundled PDF files — sometimes uploaded at different stages of the loan process. These packages had to be split and indexed by document type and matched to a predefined stacking order.
To meet tight service-level requirements, Flatworld.ai was tasked with building an automated document indexing workflow backed by a Human-in-the-Loop (HIL) system.
Deliverables Requested
- Split and categorize documents by type (indexing)
- Identify and track missing documents
- Ensure accurate uploads to the LOS (Loan Originating System)
- Manually verify edge cases and corrections through HIL review
- Maintain a processing SLA of 2 hours
SOLUTION DELIVERY
Structured 5-phase Rollout of the Document Indexing Solution
The Flatworld.ai team followed a 5-phased delivery model to deploy indexing and versioning automation using MSuite.
1Requirements Gathering and Planning
The project began with close collaboration between Flatworld and the client's operations team to define key indexing rules, expected document types, and LOS integration needs. A delivery plan was created to cover milestones outlining deliverables, configuring MSuite, user acceptance testing (UAT) timelines, production launch, and post-deployment checkpoints.
2Document Collection and Model Training
To customize the indexing system to the client's workflows, a diverse set of borrower document samples was collected. These were used to train the classification model, with adjustments made to accommodate variations in formatting, layout, and stacking order conventions.
3UAT Execution
The indexing and versioning engine was deployed in a controlled UAT environment. This allowed the Flatworld team to identify and resolve edge cases — such as mislabeled pages or non-standard formats — before production rollout. Feedback from this phase was used to fine-tune system behavior and reduce false positives.
4Production Deployment
The indexing and versioning modules were deployed to the production environment, allowing MSuite to begin processing live loan files directly from the client's Loan Origination System (LOS). Throughout this phase, the processing pipeline was closely monitored for system stability, output accuracy, and consistent performance under real-world conditions.
5Ongoing Monitoring and Optimization
Post-deployment, the system was continuously monitored and refined. Flatworld implemented planned retraining cycles to support new document types, adapt to changes in stacking requirements, and maintain classification accuracy over time.
While the project was delivered on schedule, several technical challenges added layers of complexity.
- Handling document volume at scale: To meet the 2-hour SLA, the system needed to process large volumes of loan documents simultaneously. Flatworld introduced queue management to handle peak traffic.
- LOS integration issues: Loan files were occasionally locked due to simultaneous access attempts from multiple systems. This created temporary access issues that required coordination with the client's LOS workflow.
- Turnaround pressure: Meeting the client's real-time processing requirement meant building a pipeline with both automation and HIL support that wouldn't bottleneck during peak times.
- 60% Faster Turnaround: Document processing became much faster, driven by automation across the indexing pipeline.
- 99% Classification Accuracy: Optimum accuracy was ensured by reducing upload errors and ensuring correct document placement in the LOS.
- 80% Time Savings: Thanks to the streamlined document download, classification, and re-upload processes, the client saved 80% of the time in operations.
- Flexible and Scalable Architecture: Capable of handling increasing document volumes and adapting to new document types with minimal retraining.
- MSuite: Flatworld's internal document classification and processing platform
- Loan Origination System (LOS): Client's platform for managing loan files and document uploads
Takeaway
Accuracy with Speed, Even at Scale
Flatworld.ai’s MSuite helped the lender cut processing times, reduce manual effort, and maintain high levels of accuracy. The addition of a human review layer ensured no critical documents were missed, helping the client stay compliant and consistent, even during busy cycles.
Contact Us 
Email 
Chat Case Studies
Mortgage firm cut response time by 80% and tripled conversions using Telenok AI for program queries.
Learn More
Mortgage lender automated document indexing with MSuite, achieving 80% time savings and 99% classification accuracy.
Learn More
