How ‘data augmentation’ might speed up AI adoption in P&C insurance

While advances in artificial intelligence have been slow to reach commercial P&C insurance, new trends in data augmentation could help pick up the pace, according to experts on a recent Insurtech Insights panel.

“We do see that machine learning models are coming to the CI space. It’s just at a much lower pace. However, where we do see quite a significant advancement is that we are no longer relying just on the available historical data. Instead, we are trying to augment the data as much as possible,” Ilya Kolmogorov, group chief pricing actuary, Zurich, said.

The key for American P&C insurers, panelists said, lies in balancing AI implementation with risk considerations and keeping an eye on compliance and regulation.

Christy Kaufman, VP, P&C risk, and chief compliance officer, P&C, USAA, added that “a lot of compliance professionals like myself tend to live in the negative, in the downside and all the things that can go wrong.”

“But I truly believe the biggest risk that all of us face is not taking enough risk and not moving fast enough and getting outpaced by the competition if we’re too reluctant to embrace these new solutions. We have to go forth and go forward, but we also have to know where are the pitfalls and then how are we going to guard against them,” she said.

CI’s unique AI considerations

Kolmogorov acknowledged that it takes time for machine learning and AI models to penetrate the commercial insurance space, due to unique considerations that may not exist in the same way in other lines of insurance.

“We have a much more heterogeneous exposure type on the CI side than on the personal line side, and that makes modeling much more complex,” he noted.

Additionally, he said the large amounts of data run the risk of “overfeeding” an AI model, making it less effective. Further, the “sheer amount of data” can sometimes not be sufficient to justify using more complex models.

However, Kolmogorov suggested a solution could be data augmentation, which is less about adding more data but rather better data structuring and adding variation to help models learn.

“So, we are not only looking at the historical records in terms of the policy exposure and claims records, but we are really pulling all available information that is out there, and AI and Gen AI could be a great help for that type of augmentation task,” he said.

That would only leave the final decision of whether or not commercial P&C insurers should use more sophisticated models or not. Kolmogorov emphasized that this should be carefully thought through because it will need to be justified to regulators and explained to brokers and clients — which becomes more difficult the more complex the model is.

Balancing implementation

According to Kaufman, implementing AI as a P&C business is a balancing act with risks and rewards. She recommended companies:

1. Invest in training
2. Tailor their approach
3. Involve independent teams

As part of the training aspect, Kaufman urged insurers to make clear to employees what their roles and responsibilities are at a role-based level. For instance, the different responsibilities and expectations of a model developer versus a business leader.

“Make that explicitly clear, and then have some governance around that as well,” she said.

Additionally, she said involving an independent team when tailoring an approach can “be really, really helpful” in ensuring the model accounts for crucial factors like fairness, data quality, transparency, explainability and privacy.

“This is where a team like mine comes into play,” Kaufman said. “We conduct regular audits and risk assessments. We employ a risk-based approach. We don’t apply the same approach to our high-risk models that we do to our low-risk models, but we tailor an approach that looks at our models across many domains.”

Insurers were urged to follow through with these recommendations not only at the onset but throughout the entire useful life of the model.

Regulatory change

A major consideration P&C insurers must keep in mind, however, is the changing regulatory environment in the United States. Kaufman suggested insurers have a “programmatic, systematic way to ingest new regulation as it comes along.”

She acknowledged this may prove challenging as AI advancements and regulations consistently change.

“I think the challenge becomes how do we get in front of every AI model use case going on within the company, because they’re just proliferating so quickly,” Kaufman said. “I don’t think it’s realistic that we’re going to be seated next to every model developer at every turn, nor do I think we want that. So, it’s got to be more of a train-the-trainer kind of approach of working with a model developer, so they understand what are the compliance requirements and they build for that at the onset.”

Insurtech Insights is a global organization providing resources, education, and support for insurtechs, insurers, and investors. It was founded in 2018 and is based out of London, England.

Zurich Insurance Group is a global insurer specializing in life and P&C insurance. Founded in 1872 and based out of Zurich, Switzerland, it is one of the largest Swiss insurance companies.

USAA is a financial services firm that primarily offers products and services to U.S. military personnel, veterans and their families. Founded in 1922, it is based in San Antonio, TX.

© Entire contents copyright 2025 by InsuranceNewsNet.com Inc. All rights reserved. No part of this article may be reprinted without the expressed written consent from InsuranceNewsNet.com.

Rayne Morgan is a journalist, copywriter, and editor with over 10 years' combined experience in digital content and print media. You can reach her at [email protected].