VA Uses AI to Speed Claims Processing

The Department of Veterans Affairs is using artificial intelligence to accelerate veterans benefits claims processing, and officials told the House Veterans' Affairs Committee the average processing time has fallen by 42%, from 141 days to 81 days since January 2025. VA leaders emphasized that human reviewers make final decisions and that increased AI use has not correlated with higher error rates. Democrats at the hearing challenged that conclusion, with Rep. Tim Kennedy warning, "Speed does not equal success." The agency's 2025 AI inventory lists 367 use cases, many focused on benefits processing and many still in pre-deployment.

VA highlighted the Automated Decision Support tool, referenced as ADS, which applies machine learning to automate retrieval and pre-population of claim elements so reviewers see which items are flagged as satisfied. ADS automates repetitive development tasks and aggregates service dates, service location checks (for example proximity rules such as within 12 nautical miles), and other structured evidence extraction. VA frames the tool as an assistant, not a decision-maker. Key capabilities called out include:

• •automated document retrieval and metadata extraction
• •identification of service dates and location-based presumptive checks
• •pre-population of reviewer screens that flag satisfied criteria

This is a concrete example of AI moving from pilot to operational use inside a large federal benefits program. The VA rollout illustrates common trade-offs: throughput gains and capacity increases versus auditability, error detection, and human-in-the-loop governance. The hearing underscores political and oversight risk for agencies adopting opaque or lightly documented models. For practitioners, the VA case highlights the importance of rigorous validation, end-to-end logging, versioned model artifacts, and operational monitoring for error rates and fairness metrics when models influence high-stakes outcomes.

Expect follow-up oversight on performance metrics, transparency requirements for ADS and similar tools, and potential mandates for external audits or stricter human-review thresholds. Vendors and teams building government-facing automation should prioritize explainability, reproducible evaluation, and clear escalation rules.