The Travelers Companies

Commercial insurance carriers process millions of customer touchpoints annually, and for Travelers — one of the largest U.S. property-casualty insurers — claims email volume alone represented a significant operational bottleneck. Manual triaging of inbound inquiries consumed substantial staff capacity that experienced adjusters could better deploy on complex settlements and high-value claims. Beyond routing, the company faced a deeper challenge: interpreting unstructured data at scale, including aerial imagery for property damage assessments and itemized medical bills with inconsistent formatting. Compounding this was a fragmented U.S. regulatory landscape spanning 38 state-level AI laws, making any enterprise-wide AI rollout a compliance challenge as much as a technical one. Isolated pilots were insufficient — the business needed production-grade AI across its entire 30,000-person global workforce.

Travelers deployed the Claude 4 model suite (Opus and Sonnet) via Amazon Bedrock, structuring the rollout across two distinct workforce tiers. Approximately 10,000 technical employees — engineers, data scientists, and analysts — received access to Claude Code for autonomous engineering tasks including legacy code refactoring and ML model lifecycle management. The broader workforce gained access through TravAI, a secure internal AI ecosystem purpose-built for general business use. At the claims layer, Travelers implemented an automated email classification system on Amazon Bedrock to categorize and route millions of inbound customer inquiries. Both tiers are grounded in Travelers' proprietary dataset of 65 billion data points, enabling context-aware underwriting, risk assessment, and damage interpretation across unstructured inputs like satellite imagery and medical records.

The automated claims email classification system achieved 91% accuracy across millions of inbound customer inquiries, directly translating to tens of thousands of manual hours saved — time that claims staff can now redirect toward complex case resolution. Technical teams using Claude Code have seen measurable productivity gains in engineering workflows, particularly in legacy system modernization. Looking forward, Travelers projects the AI-enabled development environment could compress its software development lifecycle by up to 50% by 2027, enabling faster product iteration and more hyper-targeted insurance offerings.

• 91% email classification accuracy at production scale
• Tens of thousands of manual processing hours recovered
• 50% projected SDLC reduction by 2027
• Deployment covers 30,000+ employees globally across technical and general-workforce tiers

• Tier your AI deployment by role: separating specialized technical tooling (Claude Code for engineers) from general workforce tools (TravAI) lets enterprises capture distinct productivity gains without forcing a one-size-fits-all rollout.
• Proprietary data is a force multiplier: grounding generative AI in domain-specific datasets — Travelers uses 65 billion internal data points — is what elevates outputs from generic to underwriter-grade precision.
• Regulatory compliance as competitive advantage: building a documented Responsible AI Framework before scaling reduces friction across multi-jurisdictional deployments and positions the company favorably as AI oversight laws proliferate.
• Classify before you automate: high-accuracy email triage (91%) is a high-ROI entry point — it frees skilled staff without requiring AI to make final decisions on complex claims.