---
path: /blog/aig-ai-underwriting
title: "AIG and AI Underwriting"
description: "What AIG's recent AI underwriting moves tell us about the future of the industry."
section: Blog
priority: 0.5
changefreq: monthly
source_file: pages/BlogAIGUnderwritingPage.tsx
---
# AIG and AI Underwriting

**Tesora**

# How AIG is Turning One Underwriter Into Five with AI

A look at AIG’s agentic AI ecosystem and what it means for the future of actuarial work in insurance underwriting.

At AIG’s Investor Day in March 2025, CEO Peter Zaffino unveiled something remarkable: an agentic AI system that has fundamentally transformed how their underwriters work. The system, called AIG Underwriter Assistance, doesn’t replace underwriters—it amplifies them. As Palantir CEO Alex Karp put it during the event:

## The Architecture: RAG + Human-in-the-Loop

AIG’s system is built on a Retrieval Augmented Generation (RAG) framework that addresses the fundamental challenge of LLM hallucinations. Here’s how their CDO Claude Wade described the three components:

The critical design principle? Human in the loop. Every extracted data point links back to its source document. Underwriters can click any number to see exactly where it came from—the specific page, cell, or section. This isn’t just about accuracy; it’s about maintaining the professional judgment that makes underwriting valuable.

## The Results: From Weeks to Hours

For North America private and non-profit financial lines, AIG reports:

## What This Means for Actuaries

AIG’s transformation reveals a pattern we’re seeing across the industry: AI isn’t replacing expert judgment—it’s removing the toil that prevents experts from exercising that judgment more often. Consider what underwriters at AIG now experience:

- Every submission is reviewed. Previously, capacity constraints meant leaving “profitable business on the table.” Now, 100% of submissions are ingested, classified, and prioritized.

- Judgment starts immediately. Instead of spending weeks gathering data, underwriters arrive to find submissions already processed with 125 key data elements extracted and cross-referenced.

- Prioritization is intelligent. ML models estimate propensity to bind, so underwriters focus on submissions with the best risk-adjusted returns.

The same pattern applies to actuarial work. Loss runs, prior analyses, applications, SOVs—the document deluge that precedes any serious actuarial analysis. What if that intake happened automatically? What if every data point linked back to its source? What if your time went to analysis and judgment instead of data wrangling?

## The Technology Stack

AIG’s system is powered by partnerships with Anthropic (Claude 3.5) and Palantir (Foundry + AIP). The architecture is deliberately modular:

- Document ingestion: LLM-powered classification handles PDFs, Excel, CSV, and Word formats. OCR for scanned documents. 97% classification accuracy.

- Data extraction: Context-specific models extract structured data points with full source traceability.

- Augmentation: Integration with 30+ third-party data sources plus internal AIG data (policy history, claims data).

- Prioritization: ML models rank submissions by propensity to bind.

The modularity matters. As Anthropic CEO Dario Amodei noted, AIG started with Claude 2.1 and has evolved to 3.5—the architecture allows swapping in better models as they emerge. This is infrastructure for the long term, not a point solution.

## The Tesora Perspective

At Tesora, we’re building the same capabilities for actuarial work. Our Actuarial Workbench ingests entire submission folders—loss runs, prior analyses, narratives, applications, SOVs—and extracts every data point with full source traceability.

The parallels to AIG’s approach are intentional:

- Multi-format batch intake. Process entire submission folders in one operation. Every document classified, every data point extracted.

- Clickable source traceability. Every extracted number links back to the exact page, row, and cell it came from.

- Human in the loop. AI accelerates the work; actuaries retain full control over methodology and judgment.

The insurance industry is transforming. The question isn’t whether AI will change actuarial work—it’s whether you’ll be amplified by it or replaced by those who are.

### Ready to Transform Your Actuarial Workflow?

See how Tesora can turn hours of data wrangling into minutes of analysis.