Accelerating Disability Benefits for Veterans with AI-Powered Claim Classification

The Challenge

When Veterans apply for disability benefits through VA.gov, they describe their medical conditions in their own words. VA's system needs to classify each condition to determine what medical exams are required and route claims to the right reviewers. But approximately 25% of conditions submitted couldn't be automatically classified because the existing system only worked when Veterans' descriptions exactly matched entries in a predetermined lookup table.

This gap meant Veteran service representatives (VSRs) had to review and classify thousands of conditions each week manually — a time-consuming process that delayed claim decisions. For Veterans waiting to learn if they qualify for benefits, these delays could mean waiting weeks longer for the support they need.

The challenge was complex: medical terminology varies widely and evolves constantly. Veterans might describe the same condition in dozens of different ways. The solution needed to handle this diversity while integrating seamlessly with VA's existing systems, maintaining strict data security standards, and ensuring only current, valid classification codes were used.

Our Approach

Aquia Nava II LLC, our joint venture with Nava Public Benefit Corporation, developed a hybrid classification system that combines the precision of rule-based lookup tables with the flexibility of machine learning (ML). When a Veteran submits a claim, our system first checks if the condition matches known terminology in VA's taxonomy. If not, a machine learning model analyzes the text to predict the appropriate classification.

We built the solution using Python FastAPI and deployed it on VA.gov's infrastructure and continuous integration/continuous delivery (CI/CD) pipeline, with comprehensive logging and monitoring through DataDog. The machine learning models are stored in Amazon Web Services (AWS) S3 and delivered through containerized deployments using Amazon Elastic Container Registry (AWS ECR) for reliability and scale.

Throughout development, we worked closely with VA's chief artificial intelligence office (CAIO) team to complete the model training and required AI impact assessment and risk mitigation plan, ensuring responsible data science practices and AI governance. Our open-source approach emphasized thorough documentation and extensive testing, making the system maintainable and transparent.

The Impact

Since deploying the machine learning classifier, 100% of conditions are now automatically classified upon submission — up from just 74% before the expanded classifier went live. That means nearly 20,000 additional conditions are automatically categorized each week, with sub-second processing times.

The automation has reduced manual classification work by approximately 80%, saving VSRs an estimated 4,131 hours — or 172 days — of working time annually, freeing them to focus on complex adjudication tasks rather than routine data entry. 

For Veterans, the impact is direct: faster claim processing means quicker decisions about their benefits. The system now handles thousands of claims daily while maintaining accuracy and ensuring expired classification codes are never assigned — improving data quality throughout VA's downstream systems.

By automating what was previously a manual bottleneck, we've helped the VA scale its ability to serve Veterans efficiently. The hybrid approach means the system can handle the full diversity of how Veterans describe their conditions while maintaining the explainability and accuracy VA requires for decision-making.

Aquia Nava II LLC partnered with the Department of Veterans Affairs Disability Benefits Crew to develop and deploy this solution under the VA Secure, Performant, Reliable, and User-Centered Experiences (SPRUCE) contract vehicle.