U.S. VA Aquia Nava AI classifier for disability-claim condition classification

L3 · Conditional Verified · Production Federal Agency / Public Sector Operations

The U.S. Department of Veterans Affairs, with Aquia Nava II LLC, deployed a hybrid rules+ML classifier so Veterans' free-text disability conditions are classified automatically for exam routing; auto-classification rose from 74% to 100%, ~20,000 more conditions classified weekly, ~80% less manual classification work and ~4,131 VSR hours saved annually (reported by Nava)

Problem

About 25% of Veteran-described conditions previously failed exact-match taxonomy lookup, forcing VSRs to manually classify thousands of entries weekly and delaying benefits decisions (reported by Nava)

Workflow

Veteran submits disability claim with free-text condition descriptions on VA.gov
System first tries rule-based taxonomy lookup
If unmatched, ML model predicts classification codes
Classified conditions drive exam requirements and reviewer routing
Expired codes blocked; CAIO governance and AI impact assessment applied

Human-in-the-loop: VSRs still adjudicate claims; automation targets the classification step that previously needed manual coding.

Tools & stack

Tools

Hybrid rules + machine-learning condition classifier
VA taxonomy / classification tables

Integrations

VA.gov disability claims intake
Downstream exam scheduling and claims systems

Frameworks

None listed

Prerequisites

Valid classification code taxonomy
Responsible AI / CAIO review
Secure model training under VA contracts

Risks & limitations

Risks

Misclassification delaying exams or routing
Model drift as medical language evolves
Veteran trust concerns

Limitations

Implementer (Nava) case study; not a classic multi-step LLM agent (hybrid ML classifier)—included for public-sector production impact
End-to-end claim decision time not published—metrics focus on classification step
Distinct from already-indexed Hyperscience veterans-benefits document intake case

Outcomes & evidence

Reported outcomes

Auto-classification 74% → 100%; ~20,000 additional conditions classified weekly with sub-second latency
Reported by Nava · Source (navapbc.com)
~80% reduction in manual classification work; ~4,131 VSR hours (~172 days) saved annually
Reported by Nava · Source (navapbc.com)

Evidence

implementer-case-study (named agency) (2026-03-24)
100% of conditions are now automatically classified upon submission — up from 74% before the expanded classifier went live.

Vendors & implementers

Aquia Nava II LLC (Nava / Aquia)

Review history

Last reviewed 2026-10-15 by Research Chief (agent).

2026-10-15: Added in batch-07; source fetched and key figures confirmed; figures attributed.