When Matt O’Malley and Steve Stabilito, underwriting leaders from AXA XL, described a ground-up process for transforming their teams into data-driven businesses recently, Carrier Management had questions about activities in the C-suite.

“Was there any top-down messaging?” asked CM Guest Editor Michael (Fitz) Fitzgerald, after hearing Stabilito describe the growth of a cross-functional data networking group that started at the grassroots level, and after O’Malley described the development of machine learning models to improve underwriting and distribution processes. “Is there anything from leadership that’s either helpful or not helpful?” Fitzgerald asked. (Related article: How Underwriters Win Business with Data and ML at AXA XL)

“From a senior perspective, particularly when we did our machine-learning driven model, that comes with the budgetary ask,” O’Malley said, highlighting the decision of AXA XL leaders to put modeling teams in a corporate center as a key ingredient to success.

“One of the challenges is when you ask a business to pay for the project upfront,” he told Fitzgerald, an insurance industry advisor for SAS Institute, who has had his own experiences with transformation projects at insurers Zurich and Royal Sun Alliance. As a business unit leader, “it’s hard to take the leap that I’m going to spend this money when I’m not quite sure what the benefit is going to be. Intuitively, I believe there will be a benefit, but until I’ve spent a significant amount of money, [I can’t really] determine whether these models are actually going to generate the outcomes we’re hoping for,” O’Malley said.

“With AXA setting up those modeling centers as centers of excellence, we’ve done proofs of concepts around what we think will work. We’ve tested them and then launched them,” he said.

Fitzgerald also asked the two underwriting executives to offer recommendations to other insurance leaders “that are now faced with all of the noise around AI and leading their organizations from this point forward.”

Test and learn, they said.

“We are fortunate to be one of Microsoft’s pilot companies for Copilot, and we have a small group that is testing what can AI do within our organization, and then reporting back on what those outcomes could be,” O’Malley said. “We’re going to learn as we go along. And I think we’ll actually learn quickly just how much has advanced in the last year-and-a-half. But I would say: build a core team, experiment and then figure out where do you want to place your bets.”Stabilito endorsed the idea of a team effort and reiterated the need to understand your data. “To understand AI, you have to understand what’s underlying that. You have to understand the data.”

“That’s what we did with the data network. We started our group as a very targeted initiative in a very low-risk and low-cost environment, …where we tested and we explored some of our curiosities before we tried to scale them across the organization or even across our individual teams.”

He likened the network approach to forming a think tank that tests theories in a very controlled environment to start an AI journey. “You might not necessarily have the quality of the data or the quantity to use AI in an effective manner,” Stabilito said.

“Charging into this saying, ‘we’re going to adopt AI’ may not be the best approach. It might be best to proceed a little bit more slowly or more cautiously in a lower risk environment—but knowing that starting small can really deliver big results in the long run.”