Agentic AI-Powered Claims Intelligence: A Deep Learning Framework for Automating WorkersCompensation Claim Processing Using GenerativeAI

Abstract

Workers compensation claim processing is extraordinarily inefficient and causes claimants and employers to
suffer. In a typical Australian workers compensation insurer,
the Claim Establishment Process can require up to 28 manual
steps performed by claim adjuster. Claims departments follow
strict business rules that stipulate if a claimant is eligible
for certain claim benefits as well as the appropriate medical
certificates, certificates of capacity, statutory reserves and claim
flags to allocate to the claimant. These are often automated
or semi-automated in complexity, however in practice, require
considerable agentic decision-making to complete the processing.
Claim adjusters perform these decisions, which can be expensive
and introduce wait times that delay claimants and employers
from receiving their benefits. This feasibility study presents a
deep learning framework designed to automate the workers
compensation claim process using generative AI. A proof of
concept application, entitled ClaimGPT, was implemented using
the OpenAI API and Salesforce Einstein GPT, providing agentic
AI capabilities for claim generation and decision-making. These
capabilities were combined with natural language processing
models for claim textual data and image generation for claim
documentation. The framework was implemented in two existing
workers compensation insurers, focusing on the Claim Establishment and Claim Decision Process. Both implementations were
successful in accelerating the claim processing, while reducing the
size of the claim reserves allocated to each claim established. The
image generation models for claim documentation synthesised
novel claim scenes and demonstrated claim simulation capability
on a specific insurance use case.