AI cost management has become one of the core concerns of corporate executives. Some companies, such as Travelers Insurance, are alleviating cost pressures by increasing flexibility in model selection and building their own models.

In June this year, the insurer, which is projected to generate $49 billion in revenue in 2025 and employs 30,000 people, released a proprietary large language model called TravelersLLM. According to Mojgan Lefebvre, Executive Vice President and Chief Technology and Operations Officer at Travelers, this internal model outperforms commercial AI models in answering insurance-related questions and operates at a lower cost than frontier models.

Although the company has invested resources in building and integrating the low-cost internal model, Lefebvre emphasized that it works alongside frontier models and does not replace the innovation and progress brought by frontier developers. When Travelers' applications handle queries, the system routes requests to either TravelersLLM or frontier models based on the type of task.

"Whether it's AI or other technology investments, we are very focused on return on investment," Lefebvre said. "The idea behind TravelersLLM is: if you don't need to use the most expensive frontier model to answer the same question, you shouldn't. It directly reflects this ROI-driven approach, namely 'choosing the right model for the right question.'" Beyond reducing operating costs, using multiple models in the ecosystem also means Travelers does not rely on any single external model or vendor. She added, "This has absolutely improved the economics of using AI at scale across our enterprise and given us strategic flexibility."

Building TravelersLLM

The insurance industry is a specialized field where general-purpose models cannot meet its specific needs, and TravelersLLM fills that gap. Lefebvre said the model was trained on millions of company documents and evaluated against tens of thousands of insurance-domain questions. It also incorporates knowledge from subject matter experts in areas such as underwriting, claims, operations, and service management.

The model empowers employees across the company with the institutional knowledge of domain experts. Applications access various models, including TravelersLLM, through APIs. "For most employees, the underlying model should be largely invisible—they don't need to care which specific model is being used," Lefebvre said. "They just need the best intelligent support at decision points."

Mapping tasks to the appropriate AI model is a trend across industries in response to the high costs of running queries through the most advanced AI models at all times. In response to user concerns about cost, Snowflake this week introduced an AI cost management feature that dynamically routes tasks to the appropriate AI model based on cost and quality. AWS and Oracle have also implemented AI cost management features and tools this year.

The future of AI depends on data

Lefebvre said Travelers began modernizing its technology foundation more than a decade ago, with data being a core part of the company's strategy. Nevertheless, the company still operates some legacy systems because its goal is not full modernization but selective advancement, focusing on areas that deliver the greatest value to customers and employees.

Data accessibility and usability remain core concerns, while semantic layers, ontologies, and knowledge graphs are crucial for enterprises to connect data to AI systems. Lefebvre added that Travelers currently runs 70% of its computing in the cloud. "If we hadn't invested in modernization, and hadn't focused on data and its connectivity, accessibility, and quality, we would never have been able to carry out AI work the way we do today," she said.

Lefebvre noted that as AI evolves from a tool that enhances individual productivity to an element embedded in company workflows, the technology's potential within enterprises remains far from fully tapped. According to Travelers, TravelersLLM itself will serve as a foundational capability for agentic AI. "The next frontier is true transformation—with AI, you can begin to envision entirely new business models, products, and services," she said.