As AI takes industries by storm, Liberty Mutual Insurance is charting a strategic course through the deluge as it selects use cases for the technology while revamping its legacy systems.
The insurance company in May launched a conversational AI auto insurance quoting app, allowing users of OpenAI’s ChatGPT to get a quote from Liberty Mutual through a chat-based interaction rather than filling out an online form.
The ability to partner with the large language model provider comes after years spent future-proofing the company’s technology foundation including a transition to the cloud and simplification of systems, said Andrew Palmer, EVP and CIO of global retail markets at Liberty Mutual. As AI advances, the insurance company will continue modernizing its core platforms, keeping data AI-ready and carefully selecting use cases and partners for new capabilities, Palmer said.
Although Liberty Mutual partnered with OpenAI on the insurance quoting app, Palmer said the company’s strategy is to remain model agnostic by building an abstraction layer on top of its mainframe data that allows the company to work with AI providers through Model Context Protocol capabilities, an open standard connecting AI to where data resides. Liberty Mutual is working with its mainframe partner IBM on building the abstraction layer.
“That ability to maintain optionality as well as context and memory within our models so we can traverse across a very complex business and not get anchored into one is a strategic bet, but one we’re quite confident in,” Palmer said.
Preparing an AI-ready, agnostic foundation
The 114-year-old insurer has spent more than 10 years migrating workloads to the cloud, transitioning from 13 data centers globally down to one primary data center, Palmer said.
While the company has moved 85% of its workloads to the cloud, Palmer said Liberty Mutual maintains a hybrid cloud strategy, which is similar to the company’s approach to AI model providers.
“We are heavily weighted AWS on our compute and storage, but we have a cloud agnostic strategy that lets us swing workloads to Google and Azure as needed. We use all three,” Palmer said. “That worked out well for us on many different dimensions. With the frontier labs, it’s especially important.”
As AI model providers continue to advance their offerings, Palmer said “you don’t want to get anchored in” to one product, especially as consumption of AI tokens rises at a dramatic rate among enterprises.
Liberty Mutual’s AI FinOps team looks across the company’s AI models and monitors token usage at individual and group levels. Tokens can be capped or added for AI use cases, and Palmer said he is working to place accountability on operations managers for linking use cases to business outcomes.
Data will also play a critical role in improving Liberty Mutual’s AI model use and token consumption going forward, Palmer said.
“You can’t run AI on bad data,” he said, adding that having data ready to pull into data catalogs and context windows for AI models will make them run more efficiently.
“If you’re making it sweat working across all these different stores and applications and trying to figure out what’s going on, you’re burning a ton of unnecessary tokens,” he said. “You probably also won’t get the right outcomes you want from a context standpoint.”
Mainframe modernization
Liberty Mutual’s core platforms remain on the mainframe, which is the big solve ahead for the insurer, Palmer said. The company’s mainframe modernization journey will span five years as it replatforms to Guidewire Software.
In the meantime, Liberty Mutual’s work with IBM on the abstraction layer enables its front end modernized technology to run in the cloud and allows the company to take advantage of the latest AI capabilities.
“We’ve done a ton of investing in modernizing the core platforms to make sure they’re ready, the data is ready,” Palmer said.
Ultimately, Palmer said enterprises cannot scale AI on top of a weak infrastructure foundation, a challenge he believes many companies are struggling with. While companies are adding AI capabilities, he said few have translated those use cases into real business value.
“If you’re trying to just chase the shiny toy and you think you don’t have to do the heavy lifting, that’s going to catch up with you,” Palmer said.
