With the introduction of OpenAI’s ChatGPT to the public in November 2022, the topic of Artificial Intelligence (AI) has been on the minds of many. Organizations are beginning to search for ways to leverage this technology in their businesses, and the insurance industry is no exception.
Terms such as Generative AI and Large Language Models (LLMs) are thrown around without much insight into what they mean or how they integrate into existing business processes.
Before delving into how the insurance industry can utilize Generative AI and LLMs, it’s essential we define terms.
What is Generative AI?
The concept of artificial intelligence is nothing new. Any technology that allows machines to perform human intelligence-associated tasks are AI - popular examples include Apple’s Siri and Amazon’s Alexa. Both systems use natural language processing to read and understand human language. Generative AI takes this concept a step further by giving machines the ability to create new content that resembles human-generated language. Large language models (LLMs) are a specific type of generative AI that focuses on human-like text. They are typically trained on billions upon billions of publicly available data to learn/recognize patterns in human language so that these LLMs can respond in a conversational/human-like way.
LLMs are attractive to organizations because they produce similar outputs to humans without the additional overhead. While insurance companies can benefit from publicly available LLMs (e.g., ChatGPT or Google’s Bard), a chasm exists between the availability and the effective use of these models when juxtaposed against the needs of the insurance industry - specifically, limitations with data privacy, data security, and response accuracy. This is why traditional generative AI is insufficient for insurance, and a tool built specifically for insurance is needed.
Introducing InsurGPTTM
InsurGPTTM is the first fine-tuned generative AI model designed specifically for the Insurance industry.
InsurGPTTM is a proprietary Large Language Model (LLM) able to read and analyze data from structured and unstructured documents such as ACORD forms, insurance submissions, first notices of loss, and other correspondence.
This technology is embedded into (and exclusively available through) Roots Automation’s Digital Coworkers, which enable organizations to work more efficiently, while eliminating human participation in routine tasks and accelerating the adoption of AI without the heavy burden of tech investment.
Most importantly, InsurGPTTM mitigates the inherent limitations of public generative AI models by focusing on data privacy, response reliability, and bringing forward insurance expertise.
Privacy and Security
Insurance companies routinely work with sensitive, personal or non-public information and cannot risk information leaks. Publicly available Generative AI isn’t designed with a focus on protecting this type of information.
These models will learn from and absorb this information into the knowledge base. At a minimum, insurance regulators expect insurance companies to be great stewards of their customer information. Any information fed into a public model is susceptible to surfacing through these models.
InsurGPTTM was built for private use by insurance companies. Private data within the models are tokenized and the underlying infrastructure is enterprise-grade.
Reliability and Accuracy
Insurance companies often work off data sets to make crucial decisions on paying claims or declining risks. It’s imperative that the information used to base their decisions are accurate and reliable.
That is not always possible with public generative AI.
The outputs are only as good as the information going in, and reinforced by people with limited or no understanding of the industry or with no concern for industry-specific use. Therefore the reliability and accuracy of the outputs are typically no better than 40%. You expect your humans to perform at better rates, so it’s reasonable to assume you need your AI to do the same.
InsurGPTTM is trained on a deep corpus of millions of diverse insurance documents and, as a result, provides an accuracy rate that is typically better than 95% and continues to improve thanks to regular input from insurance experts who work at insurance companies every single day.
Insurance Expertise
Public generative AI models lack industry-specific expertise and don’t understand the nuances of insurance language. They also aren’t able to interpret or analyze unstructured data unless the information is first organized in a manner that the model can consume.
InsurGPTTM is trained on structured and unstructured insurance data including correspondence between brokers, insureds, claimants, medical providers, attorneys, and insurance carriers.
The benefits of InsurGPTTM are already being realized.
For example, a national Third Party Claims Administrator (TPA) is leveraging InsurGPTTM to streamline and scale its first notice of loss (FNOL) process.
InsurGPTTM reviews non-standard FNOL documents and uses the data to set up claims with little human input.
Learn more about InsurGPTTM
