USAA

USAA, one of the largest financial services providers for U.S. military members and their families, faced mounting operational pressure across both its call center and back-office functions. In auto insurance, where claims accuracy and member responsiveness are competitive differentiators, call center agents were forced to manually search through a corpus of 20,000 internal documents to answer member questions — a process that introduced latency, inconsistency, and agent frustration. Simultaneously, back-office data management relied on manual routing and approval chains across fragmented systems. As a regulated financial institution, USAA had to balance modernization against strict risk and compliance requirements, making the status quo costly but any transition consequential.

USAA took a deliberate, experiment-first approach — piloting solutions in 2023 before committing to production rollouts in 2024. The strategy resulted in over 200 AI tools and products deployed enterprise-wide. Central to the rollout was EagleGPT, an internal generative AI assistant made available to all employees, alongside a dedicated code assist tool for software engineers. For call center operations, a retrieval-augmented generation (RAG) system was layered over the 20,000-document knowledge base, allowing agents to query information conversationally rather than browse manually. On the back-office side, IBM Watsonx Orchestrate was deployed to automate data request workflows, handling low-risk auto-approvals and intelligently routing complex requests. USAA also evaluates and deploys across a multi-vendor model stack including Google Gemini, AWS Bedrock, and Meta Llama, supplemented by proprietary computer vision models trained for insurance image analysis.

By end of 2024, USAA had approximately 12 generative AI solutions in production, moving from pilot to live deployment within roughly 12 months. Key outcomes include:

• Call center agents can now retrieve answers from 20,000 documents via natural-language AI query, substantially reducing average handle time and improving answer consistency.
• Back-office data workflows are partially automated through agentic AI, with Watsonx Orchestrate handling approvals and routing that previously required manual intervention.
• 200+ AI tools deployed across front and back office signals broad organizational adoption, not isolated experimentation.

The foundation built — a unified enterprise data environment developed over three years — is now enabling more complex, end-to-end automation.

• Start internally: Deploying agentic AI on back-office, employee-facing workflows first reduces risk exposure while building the organizational muscle needed for customer-facing automation.
• Multi-vendor flexibility matters: Evaluating Google Gemini, AWS Bedrock, IBM Watsonx, Meta Llama, and proprietary models allows matching model capability to each specific task rather than forcing a single-vendor fit.
• Data infrastructure is the prerequisite: USAA's three-year investment in a unified enterprise data environment was the critical enabler — AI deployment at scale requires clean, accessible data foundations first.
• Pilot-to-production discipline: A structured 2023 pilot → 2024 production cycle kept risk manageable and built executive confidence for continued investment.