Watch Grace's Session from ITC Vegas 2025

Carriers spend $85 billion annually adjusting complex commercial claims in the US. Most claims technology avoids this segment entirely.

At ITC Vegas 2025, Elysian Founder and CEO Grace Hanson presented on why automation fails where the money gets spent, and what it takes to build technology that can tackle it.

Why Expertise Drives Exceptional Outcomes

The 80/20 Problem

About 20% of commercial claims follow predictable paths and cost little to adjust. The other 80% are the costliest, most complex claims—involving coverage disputes, litigation, fraud indicators, subrogation, and bad faith exposure.

Vendors tend to demo speed on intake and triage, the simple 20%. Carriers keep paying for manual handling on the complex 80%, where the money actually goes.

What AI Is Missing

Grace identified two limitations that only embedded adjuster expertise can address:

1. Training Data Without Outcomes

LLMs lack access to complete claim files with adjuster notes, reserve development, and outcomes. Even with that data, interpretation requires specialized human guidance. A claim that closed for $50,000 means nothing without understanding whether that outcome was favorable, avoidable, or a miss.

Models identify patterns but lack any basis for knowing which patterns led to good results. That judgment comes from professionals who have lived the work.

2. Judgment Built Through Experience

Adjusters develop tacit knowledge that no training dataset can replicate: which questions to ask, when to push back, how to read a claimant's credibility, what a coverage position is actually worth in litigation.

A ten-year casualty adjuster has typically reviewed over 3,000 fact pattern variations, examined a million attachments, and had 12,000+ direct interactions with claimants, attorneys, and policyholders. That volume is how tacit knowledge develops. It cannot be documented, transferred to a training dataset, or reverse-engineered from outcomes alone.

What to Look For

Three criteria separate solutions that work from demos that impress.

• Does it solve expensive problems or automate cheap ones? First notice of loss demos well, but intake was never costly. The value lies downstream: coverage analysis, reserve accuracy, litigation management, settlement strategy.
  
  

• Can it explain its reasoning? Regulatory scrutiny is rising. Solutions that cannot produce auditable logic for coverage and settlement decisions create liability rather than reduce it.
  
  

• Was it built by people recognize good claims outcomes? Technology designed by professionals who have managed claims through trial, settlement, and appeal reflects the knowledge that models lack.