American International Group (AIG)

In excess and surplus (E&S) lines—where non-standard risks demand faster triage and sharper risk judgment than admitted markets—AIG's Lexington Insurance unit faced a structural scaling problem. Processing hundreds of thousands of submissions annually required sequential handoffs across intake, prioritization, underwriting review, and portfolio analysis, creating multi-hour workflows that consumed underwriter capacity proportional to volume growth. With a target of 500,000 submissions set for 2030, the status quo demanded headcount expansion that neither solved cycle time nor improved decision quality—leaving AIG exposed to slower broker turnaround and lost binding opportunities in a market where speed directly affects which risks an insurer wins.

AIG deployed AIG Assist, an agentic AI orchestration system built on Anthropic's large language models for reasoning and document understanding, and Palantir's data ontology infrastructure for shared context across agents. Rather than automating isolated tasks, the system coordinates four parallel agent layers—data ingestion, risk prioritization, decision support, and portfolio management—drawing on over 30 integrated third-party data sources. Submissions are automatically classified, scored for propensity to bind, and routed to underwriters in priority order, with agents surfacing relevant historical cases and real-time pricing context during review. The orchestration layer was built before individual tools were deployed, ensuring agents produce compound rather than additive value across the full submission workflow.

AIG's Lexington Insurance unit processed over 370,000 E&S submissions in 2025, surpassing its original 2030 target years ahead of schedule. The system delivered measurable gains across speed, coverage, and productivity:

• 2X–5X faster end-to-end underwriting across commercial lines
• 100% review coverage of financial lines submissions without adding underwriting staff
• 1 underwriter now handles the workload of 5—a structural shift in how operations scale
• The Everest retail commercial portfolio integration, which would have required months of manual data mapping, was completed in weeks

AIG's excess and surplus operations now target $4 billion in new premiums by 2030, with submission processing capacity—not headcount—as the enabling constraint.

• Build orchestration infrastructure before individual tools: Deploying a coordination layer first allows agents to produce compounding value across workflows rather than isolated improvements.
• Data foundation precedes AI capability: Integrating 30+ third-party sources into a shared ontology was a prerequisite for autonomous decision quality—incomplete data degrades agent outputs below the human baseline.
• Governance must be part of initial architecture: Human oversight protocols, audit trails, and bias monitoring built into the system from the start made high-volume automated decisions legally defensible in regulated markets.
• Partnership model externalizes scarce expertise: AIG's use of Anthropic and Palantir allowed them to access frontier LLM and data infrastructure capabilities without building them internally—critical given how rare combined insurance-and-AI engineering talent is.