Industry-specific LLMs to become focus of innovation, expert says

Industry-specific large language models are expected to drive innovation as AI adoption continues, according to Arturo Devesa, chief AI architect and head of AI innovation, EXL.

EXL, a global analytics and digital operations firm, recently launched its first proprietary LLM and its first LLM specific to the insurance domain.

“Domain LLMs or corporate-specific LLMs, I think that’s what is and what will continue to be the focus on LLM innovation. Of course, there will continue to be LLM innovation among general-purpose LLMs. Maybe they will be commoditized, but again, it’s still going to be a huge business. But we are a bit separate… We’re building something more specific,” Devesa said.

EXL’s insurance LLMs have been trained on underwriting and premium audits for life insurance, claims for P&C insurance, property surveys for home insurance and some aspects of medical records review for health insurance. The company is also still working on expanding those use cases to other areas of insurance expertise.

It recently released EXLerate.AI, a platform that uses these carefully-trained language models to “help enterprises reimagine workflows, with the ability to seamlessly integrate EXL and third-party AI agents into their business operations.”

Devesa said the demand for industry-specific LLMs has been steadily increasing as more insurance companies look to incorporate AI solutions that can offer better accuracy, speed and cost savings than general AI tools can.

What is an LLM?

A large language model is an AI tool that is designed to process natural language, or human text/speech, and give natural language output in response.

Devesa, who started his career in natural language processing, explained these language models were pre-trained with “massive amounts of data” from the internet — hence the name “large language models.”

“They pre-trained a model to understand the entire words on the internet, and then they would train those models on specific tasks like reasoning task, code conversion, writing code, summarizing, question and answering, classification, extraction of words, etc.,” he said.

The process by which LLMs receive human input and give back a response is known as the retrieval-augmented generation (RAG) pipeline. The AI first searches a document or database to retrieve information relevant to the user’s question, then feeds that data into a language model as context and finally generates a response based off of that.

LLMs became better at doing this and “more intelligent” as they were further developed over the years. But, it was really only when ChatGPT “captivated the mainstream” in 2022 that corporations began looking at LLMs as practical tools and assessing how to use them with enterprise data, Devesa noted.

However, companies encountered three major challenges with general-purpose LLMs:

● Inaccurate responses (also called “AI hallucinations”)
● Slow responses
● High costs

Domain-specific LLMs, which refer to large language models that are for an industry-specific use, can help address these challenges.

Why does insurance need its own?

Devesa explained that an industry like insurance, which can be complex and cover a wide variety of disciplines, could benefit from an LLM that is specifically for their purposes. It not only helps to improve accuracy and remove hallucination, but is also most cost-effective and efficient as it uses specific industry data, he said.

“That’s what’s beneficial for our clients using our domain LLM, or even the clients who want to hire us for us to help them build their own LLM: their own private data,” Devesa said.

He explained that EXL started with a general-purpose LLM then fine-tuned it using industry-specific knowledge from private data, a wealth of private use cases on specific insurance tasks and human training from a large team of operators, data scientists and AI engineers in EXL offices around the world.

“The theory is that fine-tuning will always outperform RAG because you’re taking the knowledge of your use case and you’re embedding that into the LLM. The LLM has never seen the context. The LLM has never seen the task. Therefore, it outperforms the underlying LLM [because] it’s not a pre-trained model anymore,” Devesa said.

Domain-specific LLMs in demand

According to Devesa, a growing number of private clients are looking to AI solutions providers like EXL for industry-specific LLMs. He noted that despite the proliferation of general-purpose LLMs, none of them “are going to start hiring armies of insurance-handling experts and building potentially hundreds of different domain LLMs.”

“In our case, we're already building a handful of them and it’s very challenging and it requires to already have that domain expertise and that domain infrastructure… For us, this is really important. It’s not a commodity, and clients are starting to say that they’re more interested now in our LLMs or domain-specific LLMs, or they call them corporate LLMs, meaning that clients want to leverage an LLM with their private data and not have to do a RAG pipeline with GPT-4,” he said.

EXL Service is a global data analytics and digital operations and solutions company based in New York. Founded in 1999, it now has a team of 1,500 data scientists and operates in a wide range of industries.

© Entire contents copyright 2025 by InsuranceNewsNet.com Inc. All rights reserved. No part of this article may be reprinted without the expressed written consent from InsuranceNewsNet.com.

Rayne Morgan is a journalist, copywriter, and editor with over 10 years' combined experience in digital content and print media. You can reach her at [email protected].