Swiss Re P&C exec Anil Vasagiri on climate risk technology

How Swiss Re's Rapid Damage Assessment technology works

November 1, 2023 6:34 AM

Swiss Re Next headquarters in Zurich — Swiss Re Next headquarters building, the newest part of the Swiss Re Mythenquai Campus in Zurich, as seen from Lake Zurich.

The increasing number of climate disasters causing property losses and the increased risk of climate disasters affecting property insurance create greater demand for tech solutions to evaluate climate risks and respond. Swiss Re, as a reinsurer for property and casualty insurance carriers that directly deal with climate losses, has CatNet, a proprietary location intelligence tool for assessing climate risks, and Rapid Damage Assessment (RDA), a tool that uses aerial imagery and artificial intelligence and machine learning to generate actionable information for insurers.

Digital Insurance spoke with Anil Vasagiri, head of property and deputy head of P&C solutions at Swiss Re, about the RDA, the challenges in modeling climate disaster risks and how technology addresses these issues.

What is the Rapid Damage Assessment (RDA) tool and what are its benefits?

Vasagiri_Anil_Swiss Re.jpg — Anil Vasagiri, head of property and deputy head of P&C solutions at Swiss Re.

Rapid Damage Assessment is a new solution that we launched last year, just a few months before Hurricane Ian and it already has established its value proposition in the industry. We see a lot of traction for the platform. The components of what RDA brings exist in the market today – aerial imagery exists, AI/ML techniques exist, loss estimation exists.

What is unique is how we've managed to pack all of these advances in a way to deliver truly unique, actionable insights across the claim settlement process, from just before the event is about to make landfall to when it makes landfall and as it goes through, helping insurers effectively understand, triage and manage their claims and thereby serve their insured better.

What are the problems you see with modeling climate risks? Are you seeing tech advances that address these?

The models and all the tools that exist today rely on certain information. If you're seeing increases in your losses -- actual losses versus your model losses -- then the question is, are your input parameters appropriate and accurate? Is your understanding of the existing units of exposure accurate? Is your lack of understanding or gaps in your understanding causing this increased delta between model losses and actual losses?

Secondly, is your modeling approach appropriate? Historically the industry used prior year losses to calibrate models and re-run them each year. Then they see the loss experience and get an outcome. The problem with that is that it's always backward looking. You're always looking at the prior year or previous events. It doesn't account for the inherent volatility in some of the newer perils. The industry used to call these secondary perils, but they're not secondary anymore, because the losses from these perils are as much as what used to be your primary perils. In the first half of this year, there's been almost $75 billion of annual losses from water caused by thunderstorms and severe convective events, which are not even considered as primary perils.

What does this mean for technological advances? We invest in two areas. First, solutions to address the gaps in understanding and gaps between model and actual losses. These are to present the best view of the underlying built environment, underlying units of exposure, and a comprehensive, accurate and consistent view on risk. The other aspect is bringing forward looking aspects into modeling, for a more dynamic assessment of future events, allowing our clients to have a better view of the true risk.

How do you set exposure parameters using CatNet and RDA?

It comes back to looking at what the built environment is. When you're modeling these events, it is in the context of a given geography and for a given peril. It's a question of what is the existing knowledge base about structures – how many, what they are made out of, code information and industry terms? Is that knowledge base really accurate and where are the gaps? Are the gaps that exist material, especially when modeling the severity of these perils.

Where there are gaps, we address them to present a comprehensive view of the risk by looking at certain factors. We're not questioning the factors that go into modeling. It's more about questioning the quality and comprehensiveness of the detail that the industry is currently using. We look to augment that so when you are modeling your portfolio, you have a full and a true appreciation for the underlying built environment both in the values and the structural characteristics.

What are some examples of how Swiss Re has applied AI or machine learning for this modeling?

One good example is using AI and ML approaches to do feature extraction on a building. If you're thinking about severe convective hail, understanding the quality of roof would be quite relevant, if you're underwriting a portfolio in a hail-prone region. That's about how to get the most accurate, most current view of that vulnerability in a given structure.

Also, aerial imagery at various resolutions is available today. It is more a question of how to apply AI/ML techniques to extract and introduce the underlying quality of the roof. All of this is visible and evident. That's another example of where it can materially impact the quality of the risk data.

This interview has been edited for clarity.