As the EVP and Global CIO of Liberty Mutual and the Chief Technology and Operations Officer of Travelers respectively, Monica Caldas and Mojgan Lefebvre have navigated the complex intersection of legacy insurance systems and the high-compute demands of artificial intelligence. Their expertise highlights a shift in corporate strategy where modernization is no longer about total replacement, but about surgical precision and aligning technical debt with business outcomes. In this discussion, we explore how leading insurers are prioritizing their infrastructure roadmaps to unlock the full potential of AI-driven customer experiences and operational efficiency.
High-compute AI ambitions often clash with the limitations of aging infrastructure. How do you categorize legacy systems into those requiring immediate upgrades versus those that should simply be monitored, and what specific metrics help determine which processes are “stable” enough to leave alone?
The decision-making process must be rooted in a structured framework, which is why we utilize tools like a technical debt radar to visualize our landscape. This allows us to categorize systems based on their stability and their relevance to current product goals, ensuring we aren’t just modernizing for the sake of novelty. We look closely at variables that measure how a system supports the business perspective; if a process is reliable and does not hinder the delivery of customer value, it is classified as “stable” and left to be monitored. By focusing on these specific outcomes, we can divert our energy and budget toward the 30% of systems that are truly impeding our ability to scale compute-hungry AI ambitions.
Some organizations choose to isolate legacy subsystems behind APIs rather than performing a full overhaul. In what specific scenarios is it more effective to build around an old system, and how does this “critical path” approach influence your overall modernization roadmap and budget?
We often find that the most effective strategy is to modernize what lies directly in the “critical path” of the customer experience or operational productivity. In scenarios where a legacy subsystem still performs its core function reliably, we choose to isolate it behind APIs and build modern, value-added layers around it instead of engaging in a costly full overhaul. This approach allows us to be extremely disciplined with our budget, investing only where the modernization will yield a tangible increase in customer value. It shifts the roadmap from a purely technical checklist to a strategic investment plan that prioritizes high-impact areas while maintaining the integrity of the existing infrastructure.
Transitioning compute power to the cloud is a significant prerequisite for scaling AI in claims processing and customer service. What are the practical steps for migrating data-hungry operations to the cloud, and what specific productivity gains have you observed after reaching high levels of cloud saturation?
The journey to the cloud must be aggressive and forward-looking, as we saw when we began our heavy modernization push nearly eight years ago. To handle data-hungry operations, you have to move beyond experimentation and commit to migrating core platforms, aiming for significant milestones like the 70% cloud compute saturation we have achieved at Travelers. Reaching this level of saturation provides a foundational agility that is absolutely necessary for deploying large-scale AI models across the enterprise. We have observed that this foundation allows us to rethink entire workflows, moving from manual data entry to automated systems that handle complex tasks with much higher speed and accuracy.
Modernization requires total alignment at the executive level to ensure investment decisions trickle down effectively. How do you secure “top of the house” agreement on technical debt trade-offs, and what methods do you use to communicate the long-term business value of these infrastructure shifts?
Securing “top of the house” alignment starts with framing technical debt not as a developer’s grievance, but as a strategic business risk that impacts the bottom line. You have to ensure that every executive understands that investment decisions made at the top will trickle down and dictate the company’s ability to innovate or respond to market changes. We communicate value by linking infrastructure shifts directly to outcomes, such as the ability to deploy AI faster or lower the overall cost of operations. When leadership sees modernization as a prerequisite for competitive AI adoption, the trade-offs involved in tackling technical debt become much clearer and easier to justify.
Automating complex processes, such as the first notice of loss in insurance, requires a foundation of accessible data. How should teams rethink traditional workflows to integrate AI agents for physical damage claims, and what benchmarks indicate that these automated systems are successfully supporting the customer?
Rethinking the workflow for the first notice of loss involves moving away from traditional call-heavy models toward a system where AI agents can guide a customer through the reporting of physical damage. By plugging automation directly into the intake process, we allow personal auto insurance customers to complete their claims documentation efficiently and on their own terms. Success is measured by the percentage of customers who can successfully navigate the process from start to finish with the support of AI, without needing to be diverted to a human representative. This benchmark of “completion rate” serves as a primary indicator that our data is accessible and that our automated agents are providing a seamless, high-value experience.
What is your forecast for the future of legacy system modernization in the age of generative AI?
I believe that the push for generative AI will act as the ultimate catalyst, turning modernization from a back-office IT project into a primary engine for business growth. In the near future, the gap will widen between companies that have achieved high cloud saturation and those still tethered to rigid, on-premise legacy stacks. We will see a shift toward “intelligent modernization,” where AI itself is used to refactor old code and bridge gaps between subsystems, making the transition faster and less manual. Ultimately, the organizations that thrive will be those that viewed modernization not as a one-time destination, but as a continuous effort to keep their data accessible and their compute power ready for the next wave of innovation.
