AI in the cloud pays dividends for Liberty Mutual

Liberty Mutual is one of the most experienced and advanced cloud adopters in the nation. And that is in no small part thanks to the vision of James McGlennon, who in his role as CIO of Liberty Mutual for past 17 years has led the charge to the cloud, analytics, and AI with a budget north of $2 billion.

Eight years ago, McGlennon hosted an off-site think tank with his staff and came up with a “technology manifesto document” that defined in those early days the importance of exploiting cloud-based services, becoming more agile, and instituting cultural changes to drive the company’s digital transformation.

Today, Liberty Mutual, which has 45,000 employees across 29 countries, has a robust hybrid cloud infrastructure built primarily on Amazon Web Services but with specific uses of Microsoft Azure and, lesser so, Google Cloud Platform. Liberty Mutual’s cloud infrastructure runs an array of business applications and analytics dashboards that yield real-time insights and predictions, as well as machine learning models that streamline claims processing.

As the Boston-based insurance company’s journey to the cloud has unfolded, it has also maintained a select set of datacenters from which to run legacy applications more economically than they would on the cloud, as well as software from vendors that make licensing on the cloud less attractive.

And while McGlennon believes that will change over time, he is far more focused on technologies that will define the next generation of applications.

“We’re really trying to understand the metaverse and what it might mean for us,” says McGlennon, whose mild Irish brogue bares his Galway, Ireland, upbringing. “We’re focused on augmented reality and virtual reality. We’re doing a lot on AI and machine learning and robotics. We’ve already built up blockchain and we’ll continue with all those.”

And that ability to push the envelope, especially around machine learning and AI, finds its foundation in Liberty Mutual’s rich cloud capabilities.

The benefits of a solid cloud foundation

Despite his laser focus on embracing emerging technologies, McGlennon remains highly enthusiastic about Liberty Mutual’s use of and expertise in the cloud. Sixty percent of the insurer’s global workloads run in the cloud, delivering significant savings in hardware and software purchasing, but the big benefit comes in the form of business insights from analytics on the cloud that are immeasurable, he says.

“The cloud has been a huge positive impact on us economically and surely you hear this story all the time, but it didn’t necessarily start out that way,” he says. “It tended to be additive to our legacy platforms when we started building out our cloud initially, but more recently, we’ve become far more mature in our use of the cloud and in our ability to optimize it to make sure that every single cycle of a CPU that we use out in the cloud is adding value.”

Here, McGlennon says governing controls, instrumentation, and observability metrics are key. The CIO would not specify how much the multinational company has saved by deploying workloads to the cloud but estimated it has saved about 5% over the past two and a half years. “It’s a big number,” he says.

Implementing cloud-native architectures for autoscaling and instrumenting Liberty Mutual’s applications to control how they’re performing have been crucial to realizing those savings, McGlennon says. 

Like many other early cloud adopters, Liberty Mutual deploys off-the-shelf tools such as Apptio to monitor costs and automate scaling depending on workloads, he says.

“We’ve worked with our cloud partners to better instrument our applications and better understand how they’re performing,” says McGlennon, who was a finalist for the MIT Sloan CIO Leadership Award for 2022. “That gives us greater insight into where we are potentially wasting resources and where we can optimize — such as moving workloads to smaller cloud platforms.”

McGlennon is proud of his team’s use of Apptio, for example, to best exploit its consumption-oriented model for not just its data on the cloud but for its internal services, software, and SaaS offerings, which, when linked to Liberty Mutual’s business portfolio, essentially provides the insurer’s partners with a bill of materials for all of the resources used.

The payoff of AI

Over the past eight years, the Liberty Mutual IT team, which consists of 5,000 internal IT employees and about 5,000 outside contractors, has used a variety of development platforms and analytical tools as part of its cloud journey, spanning from IBM Rational and .NET in the early days to Java and tools such as New Relic, Datadog, and Splunk.

Liberty Mutual’s data scientists employ Tableau and Python extensively to deploy models into production. To expedite this, the insurer’s technical team built an API pipeline, called Runway, that packages models and deploys them as Python, as opposed to requiring the company’s data scientists to go back and rebuild them in Java or another language, McGlennon says.

“It’s really critical that we can deploy models quickly without having to rebuild them in another platform or language,” he adds. “And to be able to track the effectiveness of those machine learning models such that we can retrain them should the data sets change as they often do.”

The insurer also uses Amazon Sage Maker to build machine learning models, but the core models are based on Python.

Liberty Mutual’s IT team has also created a set of components called Cortex to enable its data scientists to instantiate the workstations they need to build a new model “so the data scientist doesn’t have to worry about how to build out the infrastructure to start the modeling process, “McGlennon says.

With Cortex, Liberty Mutual’s data scientists can simply set their technical and data-set requirements, and a modeling workstation will be created on AWS with the right data and tools in an appropriately sized GPU environment, McGlennon explains.

The insurer also deploys software bots in its claims model to enable customers to initiate a claim, e-mail a digitized photograph of their damaged vehicle, answer a few questions, and arrange a car rental quickly. On the back end, a machine learning model analyzes the photograph of the damaged vehicle to detect whether its airbag has been deployed, for instance, and to determine immediately whether a vehicle is totaled or the damage is limited to a fender bender.

The insurer’s computer vision models may also tap into IoT devices and sensors deployed outside to generate more data for the claim.

Liberty Mutual has come a long way from its technology manifesto to its advanced use of the cloud and AI, and embracing next-generation technologies such as augmented reality and blockchain will yield further advances, McGlennon notes.

But this CIO is happy enough with the cloud and AI platform of today.

“We’ve already seen significant economic payback from being able to use machine learning models to fine-tune quotes and pricing, in fraud detection, and our coding process to make it easier for customers to do business with us,” McGlennon says, pointing to advanced cloud applications’ benefits in its core business of processing claims. “We use it all over the place.”

Although his is a property and casualty company, McGlennon believes CIOs must drive innovation and take risks “to create a culture where people feel there is the latitude to try something.”

“Risk is our business,” McGlennon said during a panel at the MIT Sloan CIO Symposium this week, adding that CIOs need to show that when things go wrong, and sometimes they will, no one is going to be made to feel that the risk wasn’t worth it.

“You have to incubate something, nurture it, give it support,” he said.