This article was created in partnership with ICAT.
AI is innovating all arms of insurance – but perhaps none more than underwriting. According to data from Capgemini, 62% of executives believe AI is improving underwriting quality and reducing fraud – with 43% of underwriters adding that they trust and regularly accept automated recommendations from predictive analytics tools. However, the true magic manifests when firms start to combine AI with the power of data science.
Speaking to IB, Robert Klepper (pictured), chief underwriting officer at ICAT, explained that the strategic integration of innovative new tools is helping redefine how underwriters evaluate properties nowadays
“The currency that we use is really location-level attributes for property underwriting,” he said. “That goes across the range of residential into small commercial and then large commercial.”
But with access to more data than ever before, the challenge isn't scarcity — it's structure. As Klepper told IB, ironically one of the biggest issues underwriters face is dealing with too much data, especially if you're a desk underwriter.
“From there it becomes a question of how do you structure the data? How do you organize it in a way that can give you insights to make decisions?” added Klepper. Here, ICAT employs both passive and dynamic data sources. Dynamic sources include physical inspections and sensor data, while passive data often comes in the form of satellite or aerial imagery.
“One of the game changers for us has been the aerial imagery,” Klepper added. “Using that granularity of location-level information… then we couple that with AI – which can interpret risk attributes that even an inspector may not see when they’re at the site. Inspectors may not be able to access certain areas - [however] with aerial imagery now underwriters can identify and interpret various physical conditions and attributes. It’s a way to analyze and structure data in a manner that most underwriters couldn’t have imagined 10 years ago.”
More importantly, this isn’t a one-time snapshot. The technology allows underwriters to view “a timeline with condition,” offering insights into how risks evolve.
“Since properties are fixed locations, you can see location conditions over time, including exposures external to the location, contents or yard storage that have been moved in and out,” added Klepper. “Additionally, proximity to other buildings or hazards, such as chemical storage, tree overhang, or other potential hazards, can be identified with precision. It just gives you a much different perspective - and you can do all of that virtually.”
Aside from its relationship with third party data, AI is also reshaping how underwriters process and prioritize vast quantities of information. “Natural language processing is one [main] area,” said Klepper. “It gives you a way to structure the information so that... you can prioritize what data is most important, what insights are relevant, and which are maybe not as important.”
With AI handling the heavy lifting on data structuring, underwriters can focus on strategic decision-making, at both an account and portfolio level. Coupled with sensor technology, underwriters can also enhance pre-loss insights and begin to approach nearly real-time risk mitigation.
“It can also help mitigate claims, because if it's water sensing or heat sensing, you can hopefully find a problem before it becomes a claim,” Klepper told IB. “With the geospatial technology we've found that to be a huge benefit post-event as well. For many natural catastrophe events, updated aerial imagery can be available almost immediately – with this perspective, you can start to prioritize the locations where there's obvious damage post-event, so you're going to get immediate feedback to inform deployment of claims teams.”
And while technology might be transforming underwriting, it’s essential for customers that human brokers maintain central to the risk transfer decision-making process. For both Klepper and ICAT it’s collaboration, not replacement, that takes center stage here.
“The broker acts as both advisor and advocate for the client, and I think the more the broker can understand the risk for their client, the more they can recommend insurance decisions and give other valuable advice. As underwriters, we get a certain amount of standard information in a submission, but more insight helps in the transaction and overall risk transfer decisions– [meaning] we're all operating on the same set of conditions.”
At the end of the day, it’s a matter of alignment. With data parity between broker and underwriter, decisions can be made with greater accuracy—and pricing can more precisely reflect actual, not assumed, risk. “Underwriters price for uncertainty,” Klepper added. “The price is higher because they may not understand certain elements of the exposure.”
Looking more broadly, Klepper identifies complexity and dynamism as defining characteristics of modern risk in 2025.
“Risks are increasingly complex,” he told IB. “Especially as you move up the scale into the commercial and large commercial, it’s very dynamic. You have commodity prices, the economy, supply chain, inflation, and then you have physical conditions at locations - they're heterogeneous. There are no two locations, operations or production lines the same.”
Natural catastrophes, like the recent California wildfires, however, are compounding the challenge. Research from Munich Re found that in 2024, US insurers paid out approximately $140 billion in claims due to natural disasters – with the total economic cost in 2024 was estimated at $320 billion.
“These incidents expose risks that we didn’t necessarily know were as extreme or as broad as previously expected,” Klepper added. And while traditional models handle perils like wind and earthquake relatively well, other exposures—wildfire, flood—require deeper data enrichment.
“The more we can inform the models, including the probabilistic models, the vendor models... the uncertainty of the modelled result is mitigated. Ultimately, I think the biggest challenge, and the biggest opportunity, is how we can further identify and harness big data. This will further inform AI models to better interpret the data to make more accurate decisions.”
