How Underwriters Win Business With Data and ML at AXA XL

Imagine being able to tell a broker prior to a submission, “I can take this account in my underwriting portfolio,” instead of giving a broad appetite listing for the distributor to puzzle through.

Executive Summary

A machine learning model that flashes stop-and-go signals to underwriters in AXA XL’s environment unit and a distribution tool that fills the U.S. business pipeline with carrier-selected accounts are two of the successes that underwriting leaders Matt O’Malley and Steve Stabilito attribute to a cross-functional focus on insights hidden in decades of insurance data. Here, they provide a high-level view of what it takes to win with data—curiosity, simplification, funding outside of business units, bite-size proofs of concept and communication.

Michael (Fitz) Fitzgerald, Insurance Industry Advisor for SAS Institute Inc., served as guest editor for this article and others featured in CM’s Q1-2024 magazine, “Leading the AI-Powered Insurer.“

A grassroots effort by underwriters to incorporate data in their decision-making has turned the vision into reality at AXA XL, according to Matt O’Malley, AXA XL’s U.S. country manager and East zone manager, who described the benefits of a machine learning tool for pipeline management that was created from predictive relationships buried in the carrier’s data.

“The aha moment has been realizing what the conversion ratio has turned out to be,” O’Malley said. There’s “a difference between telling someone specifically, ‘I can really perform on this account based upon our data and analytics’ versus describing to a broker that this is generally what I want, and then leaving the broker trying to figure out as things come across their desk, ‘Does this fit what that person told me?'”

“Now, we’re able to do that across multiple lines of business,” O’Malley continued, explaining that the machine learning tool is able to identify where AXA may have interest in offering different lines of insurance to the same client. “This really brings an additional client lens to AXA. There are lots of places where we can engage and solve needs for clients that we hadn’t been addressing in the past. And it’s all coming from the data,” he said.

Steve Stabilito, underwriting manager for the construction professional and pollution teams, agreed. “By being proactive—not reacting to a submission that comes in but being proactive in seeking that business—we can have a better sense of ownership of those opportunities within our teams. Then we can come to a faster, more confident decision when it comes time to make the decision.”

In addition to the distribution tool, AXA XL has developed a machine learning model that is literally “greenlighting” the best business with flashing signals built into existing tools used by AXA XL’s environmental underwriters, according to O’Malley.

Data Networks

The starting point for both machine learning tools was a decade-long effort to understand, clean and harvest the insights hidden in the company’s data, the two underwriting leaders stressed during an interview with CM Guest Editor Michael (Fitz] Fitzgerald, Insurance Industry Advisor for SAS Institute.

Stabilito led off, noting that he has taken on a role within his team to be a “steward of data and data analytics,” essentially bringing data into every conversation underwriters have about planning strategies or individual risks. “On our monthly calls, I incorporate elements of data to give different insights that maybe team members have not seen before.”

“Using the data is bringing us to faster, more confident decisions.”

Steve Stabilito, AXA XL

Stabilito has also created a dashboard for his team, which serves up the right information at the right time for his underwriters to make a decision, according to O’Malley, who reported that AXA XL is looking to roll that out more broadly across the organization.

Beyond his team, Stabilito started a data network throughout the company. “That network is essentially a dozen or so people who have similar backgrounds throughout the company who are interested in bringing data insights to their teams,” said Stabilito, whose education in geological engineering confirms a lifelong interest in unearthing information beneath the surface. “We get together once a month and talk about data. We share experiences—the knowledge and the tools that we’re using for analytics,” he said, noting that the group includes members from actuarial, claims, underwriting and finance—”those of us that live in the numbers every day.”

While the current group has been getting together for about six months, the network started two years ago as a two-man operation—Stabilito and one of his peers in a risk engineering group at AXA XL. “He and I have both started using a lot of the tools that are available…to analyze our books and build dashboards that can deliver insights to our internal and external customers,” he said. Recognizing that the effort should move beyond the two of them, they “put feelers out” to see who else was interested.

“The real theme around the data network is empowering the use of data analytics in our business,” Stabilito said. “Just by demonstrating to everyone, showing what some of these tools are capable of doing, it sparked an interest among the group [members]. And then it went from there. We had a curiosity about our business—segments that might be profitable or what class of business are we more successful in pursuing.”

From Data to Model Building: Green Means Go

O’Malley, who previously led the North America environmental practice, recalled how that business started to look for ways to improve the underwriting results using machine learning back in 2014. “We had a lot of stakeholders at the table,” he said, noting that a separate strategic analytics unit, as well as the environmental unit’s pricing team and underwriting managers, were involved at the outset, working with 10 years of existing data. “We needed to make sure that everybody understood the mission and what some of the critical values were,” he said, reporting that the task initially involved identifying some 350 or 400 key variables that were key drivers of desired outcomes and building a multivariate model.

“One of the really important pieces in terms of how we use the data was making sure that we had alignment not only within the underwriting teams but with our actuarial team, with our claims team.

Matt O’Malley, AXA XL

“But at the end of the day, we turned our analytics model into a red, yellow, green [signal]. So, underwriters could understand, without needing to understand the mechanics, ‘Is what I’m doing at the desk level making an improvement to the book or not?'”

Explaining further, O’Malley said, “When it’s time to bring [something like this] to market, when you’re going to take those new insights and build a strategy around it, it needs to be executable. We tried to break it down into bite-size pieces for the underwriters. So, instead of understanding that we worked on 300 variables, which we told them at a high level, we basically said the way the model works is you use your pricing tool the same way you did, but now you have these icons of red, yellow, green [to indicate] how that modeling layers into what we’re trying to accomplish in the market.”

By taking out the complexity for users of the tool, you avoid the prospect that they will try to “reverse engineer” the process, he said. “When it is different than we have done in terms of go-to-market strategy for the past 20 years, you want to make sure that people are focused on using the tool as it’s designed versus trying to figure out why this shouldn’t work,” he said, noting that another desired outcome is driving efficiency.

Engagement Strategy Matters

O’Malley confirmed that while the signals flash account by account, the underwriters also were able to see portfolio results at monthly meetings. “Our actuarial team would come in and actually present the results for the portfolio to the [underwriting] team. So, they could see how we were driving loss ratio improvement over a month-to-month basis, which then became a year-to-year basis…”

“At that team meeting, we had our entire underwriting staff, our assistant underwriting staff, our claims, our risk engineers. Everybody heard the same message. People could see the actions that underwriters were taking every day were turning into the results that were being reported out,” he said.

Stabilito re-emphasized the importance of showing the underwriters the loss ratio impact at a portfolio level. “Underwriters are looking at their individual risks—the individual risk that is right in front of them at that particular time. They may or may not be thinking about how that affects the whole portfolio. They may not even know how it affects the portfolio.”

He also described similar tools that his team uses for pricing. “We have models that will allow us to input our targeted pricing or what we anticipate offering for an account. The tools show us what impact that has on our individual results, team results and the portfolio as a whole,” he said, adding that an unfavorable indication overall could prompt the underwriter to “go back to the well and think about making some adjustments.”

O’Malley added, “I think one of the aspects that people overlook [is] how important change management is as we implement new ways to incorporate data into our everyday [work],” explaining that an engagement strategy needs to involve everybody—”from the senior leaders who were sponsoring the investment in the models all the way down to the underwriters.”

He continued: “There needs to be an entirely different way that you think about that end-to-end execution [and] communication…because it’s easy to just go back to what you’re comfortable with—and that’s usually not using the data in the way we’re looking to use it today.”

Making Data Exciting

Stabilito reported that his team isn’t quite as far along as O’Malley’s in the adoption of machine learning or AI technologies. “We’re in the infancy of incorporating data into our regular conversations, with the goal of taking advantage of those technologies as they are built more and more around our business.”

In Stabilito’s view, “It really starts with making data exciting, getting everybody involved in what’s going on—maybe not in the weeds and the mechanics, but at least incorporating data into conversations” that normally didn’t draw upon data insights. “We get together and we talk about our book profile. We talk about specific accounts or specific risks. My intention on this is to incorporate an aspect of data analysis in every one of those conversations, and ultimately end the conversation with ‘I took this approach’ or ‘I went in this direction with this opportunity because that is what the data is telling me.'”

“We have not had that as part of our culture until very recently,” he said.

The process of becoming a data-focused insurer is not without challenges, Stabilito confirmed. “I don’t think any company’s data is ever 100 percent clean. In fact, probably not even close to 100 percent. And we’re certainly not an exception to that. But what we’re really trying to do within our group—and this is one of the goals of the data network, too—is we’re trying to get away from the apathy that comes from knowing that the data is not always clean.” Now, instead of saying, “It’s somebody else’s job to fix that. I know it’s not right, so I’ll just ignore it for now,” Stabilito is encouraging a proactive approach. “If you see something that’s consistently wrong, let’s fix it so that we can use the data,” he said.

“It’s our responsibility to ensure the accuracy of this. It’s not IT’s responsibility; it’s the business’ responsibility,” he stressed.

As more teams adopt data-driven approaches and use models based on the data, there’s a payoff in the marketplace, O’Malley believes. “It allows us to be much more consistent across a significant number of variables so that when we execute in the market, our distribution channels see us being more consistent than we would have been in the past,” he said.

Fitzgerald observed a sense of satisfaction coming across as the two underwriting leaders spoke about their data-driven approaches to targeting profitable business.

“Absolutely,” that’s another benefit, O’Malley said. “Being an underwriter is kind of like being a baseball player. Nobody bats 1,000, but if all of a sudden you can bat 400, you feel really good about that. [With the data insights], you win—and you win together.”

“You’re a Hall of Famer,” Stabilito added.

***

This article is featured in Carrier Management’s first-quarter 2024 magazine, “Leading the AI-Powered Insurer,” conceived by Guest Editor Mike (Fitz) Fitzgerald.