How Liberty Mutual’s CIO is leading the shift from AI experimentation to execution

As global CIO at Liberty Mutual, a 113-year-old insurance provider, Monica Caldas is determined to modernize her company’s technology. That includes fully embracing AI, while walking the line between transformation and governance that a regulated organization like Liberty Mutual demands.

This year, Caldas was named the 2025 MIT Sloan CIO Leadership Award winner for her vision in guiding Liberty Mutual’s IT strategy. She joined in 2017 and previously spent more than a decade in technology leadership roles at GE. Having lived through a number of major shifts from cloud to generative AI, she’s learned how to lead through change.

She recognizes that with generative AI, everyone, including her, is learning as they go. “What is interesting about GenAI, and this inflection point that we're all experiencing is that there is no playbook, and we just have to move through,” Caldas told FastForward. She said that means that there’s a lot of experimentation, hypothesis and validation of assumptions as they iterate and learn.

But she definitely has a plan to deliver AI across her organization, and find value where other large companies have struggled. As ServiceNow head of global innovation, and author of the book Mindshift, Brian Solis told us in a recent interview, “AI cannot work in the box of business as usual,” and that’s something that Caldas seems to inherently understand.

Three-pillar approach

She says she and her team take a three-pronged strategy to AI implementation, and the three pieces are closely aligned. Much like Workday CIO Rani Johnson, the first is to find easy wins, something the two leaders call everyday AI. “We wanted to make sure that we were not only demystifying AI, but also teaching through learning campaigns,” Johnson said in a recent FastForward profile, and Caldas is taking a similar tack.

The idea is to train and familiarize people with the different AI models, while teaching them about the limitations of these tools before they actually use them on the job. “[The training is designed] so that they can develop a deeper intuition about what this can do, what the limitations are, and that will help us then be more thoughtful in our approach," she said. And this is especially true when it comes to model management, risk management and accounting for hallucinations. 

The second pillar involves practical application. How do they put AI to work in ways that make sense for the business? She notes they have identified so many areas that could benefit from an AI makeover, there’s now a growing backlog to work through.

But it’s not enough to simply identify areas to apply AI, which brings us to the third pillar: carefully reexamining and reinventing workflows through this new AI lens. “It's not just about choosing use cases,” she said. “It's really about thinking broadly about processes and figuring out how to rewire sections and rethink parts of them.”

Caldas points to her own IT organization as an example. After updating its core service management system and consolidating the knowledge and data around those processes, her team began developing an agent that could interact directly with employees to answer questions faster. Now they’re taking it a step further by developing AI agents to proactively detect issues like laptop performance degradation and fix them remotely before employees even notice that there’s a problem.

Overcoming the legacy problem

As a large organization, Liberty Mutual has a number of legacy systems that can make implementing technology like AI more challenging. Caldas has worked hard to understand the level of technical debt and begin to address it. Technical debt refers to the accumulated complexity associated with outdated systems and flawed decision making related to software development, often because there isn’t time or budget to do it in an optimal way.

What is interesting about GenAI, and this inflection point that we're all experiencing is that there is no playbook, and we just have to move through.

She understands that if you don’t deal with that legacy debt, it limits the organization’s ability to take advantage of AI because they need access to the data locked inside those older systems.

“So we have a proprietary framework we built that is a technical debt framework that looks at a variety of variables and really maps them to the business strategy, the business aspirations, the risk profile and exposure of the technology stack,” she said. “Based on that, we have made selected investments to modernize in phases, and so we are very aware and have prioritized modernization at the company.”

The governance challenge

At the same time, they can’t do this willy-nilly because they have to ensure they comply with regulatory requirements and protect customer data as they make this shift — and that’s true regardless of whether the data is locked in older systems or more updated ones. 

“Part of what I think about quite a bit with my technology team leaders and executive team is we have to make sure we protect customer data and meet all our data privacy rules and policies that we have in the company,” Caldas said. That includes blocking access to unauthorized models to reduce the risk of data leakage, while acknowledging there is no perfect way to eliminate Shadow AI, where people use unsanctioned AI tools.

Following a similar course to Workday’s Johnson, she established a responsible AI committee pretty early in 2023 as they anticipated the changes that generative AI would bring. “We still have that today, and there's a working team that manages risk according to a variety of different levels that we look at, and that also helps make sure that we're doing things in the right way and keeping all the policies that we have intact,” she said.

While governance helps set up guardrails, especially around data protection, innovation often comes from outside. Startups can help with a lot of these areas, offering new ways to solve some of their more entrenched problems using emerging technologies. “We have a continuous pipeline where we test and iterate and experiment with startups,” but with the understanding that to expand the relationship they have to eventually level up to the needs of an enterprise like hers. That involves introducing things like enterprise agreements, disaster recovery, data protection and so forth.

Caldas must constantly balance the drive to update the company’s systems with the imperative to safeguard data privacy. Her recognition as an exemplary leader is evidence that she’s advancing the organization and doing what’s necessary to keep Liberty Mutual at the forefront of technology.

Featured photo courtesy of Liberty Mutual.