USAA: AI efforts
USAA, which banks and insures U.S. military members and their families (~14M members), is one of the most AI-mature U.S. insurers — ranked #3 in Evident's 2025 inaugural index and #8 in 2026, and #1 on the Talent pillar. Part of InsureBench's overview of AI in insurance; every figure is sourced.
Where USAA ranks on AI
USAA ranked #3 overall in Evident's inaugural 2025 index (the top US insurer) on the strength of a #1 Talent ranking — nearly 1 in 20 staff works in AI, about 3x the field — and #8 in 2026.6
It runs a deliberately multi-vendor stack and a structured intake-and-governance process, moving about a dozen generative-AI solutions into production in 2024 after a year of pilots.2
global insurers
The AI build-out
From early computer-vision photo claims to a governed, multi-vendor generative-AI programme.
AI photo-estimating for auto claims
USAA, Google Cloud and Mitchell launch AI photo estimating for near-real-time damage estimates.5
Vertex AI claims pipeline
A computer-vision pipeline identifies damaged parts and estimates labour; 120+ models, up to 28% performance gains.4
Gen-AI pilots
USAA runs numerous generative-AI pilots to learn safe implementation patterns.2
~12 gen-AI solutions in production
MSR Co-Pilot, EagleGPT and a gen-AI pair-programmer go live; ~38,000 staff take "AI Aware" training.1
IBM watsonx for documents
USAA builds gen AI at scale with IBM watsonx and Granite to structure unstructured claims/underwriting documents.3
Evident AI Index
USAA ranks #8 overall, among the top US-headquartered insurers.6
Inside the programme
A member-service co-pilot, an employee assistant, and a multi-vendor model strategy.
MSR Co-Pilot
A gen-AI co-pilot that surfaces answers, auto-summarises member interactions and captures follow-ups for service reps.1
EagleGPT
An internal workforce assistant plus a developer "Code Assist" pair-programmer.2
Model stack
Gemini/Vertex AI (used extensively), AWS Bedrock, IBM watsonx and Meta Llama open-weight models.2
Computer vision
A Google Cloud vision pipeline assesses vehicle damage for repair/replace decisions and touchless claims.4
"With a technology like gen AI, it's important to learn our way into the right patterns that give us not just speed to market, but also reliability, accuracy, and preserving trust and privacy."
Ramnik Bajaj, Chief Data, Analytics & AI Officer, USAA — 20256Experiment, then scale — under a three-layer governance model
USAA's posture is "experiment and decide": an AI Council scores each idea on feasibility, value and risk; a model-risk function validates; and performance is monitored in production. It pairs this with a public responsible-AI commitment.
That intake-and-validation discipline — does the model actually perform? — is the same question an insurance AI benchmark answers.
- AI Council scores feasibility, value and risk at intake
- Model risk management validates before scale
- Ongoing monitoring of deployed models
- ~38,000 staff trained via "AI Aware"
Why this matters for benchmarking
USAA puts AI into claims (computer-vision estimates) and member service at scale. The more decisions models drive, the more it matters whether they're right — which is what InsureBench measures. See the methodology.
Sources
- MIT Sloan Management Review — How GenAI helps USAA innovate (Oct 2024). sloanreview.mit.edu
- TechTarget — USAA's experiment-and-see approach with AI (Sep 2024). techtarget.com
- SiliconANGLE — USAA builds gen AI at scale with IBM (May 27 2025). siliconangle.com
- Google Cloud — USAA and Google Cloud transform insurance operations (Mar 2022). cloud.google.com
- Repairer Driven News — USAA + Google Cloud + Mitchell AI photo estimating (Sep 2019). repairerdrivennews.com
- Evident Insights — AI Index for Insurance 2026 (USAA #8; #3 in 2025). evidentinsights.com