Travelers made a major bet on AI when it committed to developing its own in-house large language model, TravelersLLM.
It seems to be paying off; last week, the company
While many companies are facing ballooning AI costs, TravelersLLM is moving in the opposite direction. Although Travelers uses frontier models for some needs, its in-house tool is "meaningfully faster and less expensive than frontier models" for certain insurance-domain tasks, said Mojgan Lefebvre, executive vice president and chief technology and operations officer at Travelers.
Travelers also measures ROI in terms of speed, consistency and how well TravelersLLM supports its staff.
Lefebvre shared the details of Travelers' approach with Digital Insurance, the development of the LLM and the company's overall AI and technology strategy.
Responses have been lightly edited for clarity.
Why has Travelers built its own LLM?
Our decision to build TravelersLLM reflects a principle that has guided our technology strategy for years: Buy for what is broadly available and build where we can create competitive advantage.
We use leading frontier models, and we benefit from how quickly they continue to advance. They are exceptional generalists. But insurance is a highly specialized domain, and Travelers has proprietary knowledge, data and expertise that is not embedded in general-purpose models.
TravelersLLM was built for that niche work. It was trained on millions of Travelers insurance documents and evaluated against tens of thousands of insurance-domain questions. It has a track record of strong performance on insurance-related tasks while also being more efficient and more cost-effective than relying on frontier models alone.
That matters for two reasons. First, it improves our efficiency as AI usage scales across underwriting, claims, service and operations. Second, it gives us strategic flexibility. We want to benefit from frontier models, but we do not want to be dependent on any one provider for work that is central to Travelers' long-term advantage.
So, this is not an either-or strategy. It's a matter of selecting the right model for the work: frontier models where broad general capability is best, and TravelersLLM where deep insurance-domain expertise creates differentiated value.
How does this fit into the overall AI and tech strategy for the company?
Rather than simply a stand-alone product, TravelersLLM is a foundational capability that is part of a broader AI strategy.
For years, we have invested in the foundations required to scale AI responsibly: Cloud, data, modern architecture, engineering talent, agile ways of working and governance. Those investments were originally made to advance our broader business strategy — extending our advantage in risk expertise, improving experiences, and increasing productivity and efficiency — and they are now enabling us to rapidly harness the power of AI.
TravelersLLM is one piece of a broader ecosystem that includes TravAI, our enterprise generative AI platform, frontier models from leading providers, knowledge systems, data products and the agentic capabilities we are building across areas such as claims, underwriting, operations and software development. Our architecture is intentionally open and componentized so that we can leverage the best external capabilities while preserving the intelligence, data and context that differentiate Travelers from its competitors.
The strategic point is that TravelersLLM helps make our institutional knowledge more accessible at the point of decision — not as a separate AI tool, but as part of how we are embedding intelligence into the flow of work.
What is the return on investment for a project like this?
We evaluate AI investments the same way we evaluate other strategic technology investments. They have to improve business outcomes. That means improving decision quality, helping our people act faster and more effectively, improving experiences for customers and distribution partners and strengthening productivity and efficiency.
TravelersLLM is compelling because it performs well on the insurance-domain tasks that matter to our business while improving the economics of using AI at scale. In our evaluations, TravelersLLM delivered strong performance on insurance-specific questions and was meaningfully faster and less expensive than frontier models for those types of tasks.
But the return is not just model cost. The larger value comes from what it enables across the enterprise: faster access to institutional knowledge, more consistent application of expertise, and the ability to support AI capabilities in underwriting, claims, service and operations using Travelers-specific context.
That is how we think about ROI on this project: not just as a technology benchmark, but as a business capability that can improve quality, speed, consistency and scale.
What's ahead for Travelers in terms of LLMs and AI?
We are still early in what this technology makes possible.
The next phase is moving from using AI as a tool to embedding it directly into workflows. That includes assistants that help employees find and apply knowledge, agentic systems that can execute multistep processes with appropriate guardrails, and domain-specific models like TravelersLLM that bring insurance expertise into those workflows.
We are already seeing this progression across Travelers. TravAI has reached broad enterprise adoption, with tens of thousands of employees using AI monthly and millions of interactions. We are also deploying agentic capabilities in areas such as claims, where coordinated AI agents can support first notice of loss, loss consultation, claim creation and post-claim workflows.
What I am most focused on is making sure we scale this responsibly and in a way that empowers our people. The organizations that lead in AI will not simply be the ones with access to the best models. They will be the ones that know how to combine those models with proprietary data, domain expertise, strong governance and a workforce that understands how to use AI effectively.
Anything else you would like to share?
Every company can access powerful frontier models. The difference is what information those models integrate — proprietary data, institutional expertise, modern architecture, governed workflows and people who know how to use the technology responsibly.
TravelersLLM brings those elements together. Teams at Travelers built it, with close partnership across technology, data, legal, compliance, risk and the business. That cross-functional model matters because in a regulated industry, you do not scale AI by separating innovation from governance; you scale by building governance into the way AI is designed, tested, deployed and monitored.
TravelersLLM combines leading AI capability with our own data, our own expertise and our own standards — and it gives us a foundation to keep building from here.
