Learn how the latest underwriting technology is driving new views of risk.
Key Takeaways:
- Climate change is affecting all of us, how does it impact underwriting?
- What type of data is required for effective underwriting and analysis for right-pricing the risk?
- How can AI be effectively utilized in underwriting climate Risk?
Kaitlyn Mattson (00:11):
All right. I think we're going to get started. Sounds good. Hello, welcome to NextGen Underwriting for Property Climate Risk. I'm Kaitlyn. I'm the managing editor of Digital Insurance. Today we're going to explore how climate change impacts underwriting, the data required for effective analysis. We're going to try to leave some time at the end for questions if y'all have any, but we're going to get started and you all are going to introduce yourself.
Cory Isaacson (00:44):
You want to go Valerie?
Valerie Turpin (00:44):
I would start. So good afternoon. I'm Valerie Terpene. I'm the Chief Underwriting Officer for Art Insurance, which is a carrier of 27, 28 billion of market cap. We joined the SFP 500 last year and in property we are very active in what we call the excess and surplus market. So a lot of catastrophe. That's one of our speciality.
Cory Isaacson (01:09):
Great. And I'm Cory Isaacson. I'm the CEO and Co-Founder of Rethought Insurance. We specialize in flood insurance and flood risk. So we have a lot of experience in modeling catastrophe risk and how we do it. And it's a fascinating subject. And so one of the things I was going to say is with a small group like this, it's a really great advantage because you're all going to be very well informed by the end of the session.
Kaitlyn Mattson (01:33):
Awesome. So we're going to just jump in with our questions and the first one is, what are the limitations of traditional catastrophe models in capturing the complexities and kind of uncertainties associated with climate change risk?
Cory Isaacson (01:47):
Yeah, well, I can pick that one up I think. Cool. So we have worked a lot with catastrophe models. I worked for a large catastrophe modeling company before on the technology side, so I have a lot of experience there. They all have very strong teams in climate research, so they're all incorporating that as best they can, but models only come out typically every two years for a given model. So how current is it? And so what we know from underwriting when it comes to climate risk and climate change is that we, there's a question mark and we don't really know, so we have to allow for that. The other thing that's very interesting, especially when it comes to flood, is flood used to be storm search, that was the main concern was storm surge by the coast. Now it changed completely to what we call fluvial risk, which is rainfall.
(02:36)
Even if you looked at a big storm like Ian, which was massive, the damage was from the rainfall after the storm surge, not from the storm surge itself. So it's completely shifted considerably. And so we've had to shift a lot of our underwriting expertise and technology to take rainfall risk into account, which is a much more challenging thing, especially like I said, when you consider risk, how many of you have had a flood at your homes or anything? No one lucky. Okay, I have so I know what it's like. But anyway, the point is it the rainfall, what we used to think was a freak rainfall storm like IDA is now the norm. We've seen many, many of these in a row in the last two, three years. And so that's what we really have to be prepared for. And that's definitely because of climate change, because as the storms get, as the water warms up in the ocean, the storms get more water in them. They're much heavier. So they have typically less wind damage and much more rainfall that comes from that. But they also are causing rainfall in the middle of the country. The west coast, you saw what happened here in California, if any of you live here, it's been phenomenal with the atmospheric rivers and other things like that that have happened. So all of this is definitely showing that we're definitely going through a cycle of climate change and it's very precarious for insurers. We have to watch very carefully.
Valerie Turpin (04:02):
And I would ask one, I would add 20, 30 years ago when the model were developed, that was earthquake and name wind storms, pretty much it. And then models were developed for severe converse storms for flood and some others. But when you look at a large event, an event of magnitude, we're not speaking about attrition losses, but magnitude events, the catalog of the events would reach a certain magnitude. I'm speaking about more than 10 billion in total damage and above. It has extended drastically. The number of model have not extended. So there are stochastic models that are available on the market. So it's much more complicated for the carriers because we have used the models to quantify our probable maximum loss, our aggregates, and from there be able to determine the price, but also the appetite, how much more aggregates we want to get. And there are plenty of events RCCC, we've seen what happened after Charles frauds.
(05:09)
I mean nobody can quantify what an RCRCC costs will be in the market today, but we have to consider those events. So that's the first piece. The second piece is that what we learned, as Corey said, after events, we compare what a model told us about the expected loss with the actual losses. And we have seen over the past years with the increase of the frequency and the increase in the magnitude of those losses, that there are more and more discrepancy between what we expect from the model and our actuals due to multiple factors, more constructions, less quality in the building construction, less application in some states of the regulations, social inflations, whatever. We've seen more and more complexity. So what happened, a lot of carriers do not rely anymore on one single models. I mean that the combination of multiple models become the norm plus a little bit of twist of your personal view of these.
(06:11)
And then to go a little bit technical, so we use models to calculate our aggregates maximum amount that the portfolio will generate as a loss for 250, 500, 1000 year return period. But we also use that for pricing. So in the past when I started started using model for underwriting, it was very simple. My insurance department were telling me the cost of three T insurance is X. So we are taking the average annual loss, which is a result of the model would tell you how much you are going to put every year to pay for a loss of 250 year return period. We are multiplying by the cost of insurance, and we call it today the multipliers loss. Cost multipliers or LCM are now are much more complicated. I mean that I need a team of factories to tell me what the LCM is because depending if it's a excess of loss, primary regions of PERS on that, now I've got a wall combination of factors to tell me what my lost cost multipliers are.
(07:20)
And more importantly, I changed my lost cost multipliers several times a year in the same time because climate change is not that integrated into models. I changed my guidelines several times a year when I started insurance, the guidelines existed from 20 years and they haven't changed for 10 more years. I mean right now I changed my guidelines twice a year because we need to adjust to what's going on. They are trend, they are fast paced about changes and we know that the model have not integrated all of that. So we need to add a lot of analytics from our side and make sure that and some margin, make sure that we do have the book that align with the appetite that we have. So definitely model are great, but they are very limited today compared to the type of granularity, the type of consistency and certainty that our board members ask us to get.
Kaitlyn Mattson (08:19):
Great. Yeah, I think there's a lot of business challenges that come with climate risk and I'm kind of curious about how you stay flexible in underwriting and the importance of data in that process as well.
Cory Isaacson (08:37):
Yeah. Well, data is incredibly important. When you're talking about flood risk. We would take a building like this, you can't just underwrite this building. You have to underwrite different parts of the building because the water's going to come in typically from one place. So you have to be able to determine that. So it takes incredibly data, but there's all kinds of things that we take into account. It could be past weather, it can be what the models are telling us. It can be satellite imagery. We have historical satellite imagery that will actually tell us if someplace likely flooded or not in the past. So it's incredible as so there are hundreds of attributes. And the problem about this is that an underwriter can't possibly evaluate hundreds of attributes on every single building. Sometimes we get policies that have a hundred buildings in them, a thousand buildings.
(09:25)
So what are you going to do to actually evaluate that? So what we do is we use technology to capture all of those attributes. We use AIas well. I'll talk about that a little bit later. But the point is what we want to do is prepare the data in such a way that an underwriter can make a valid decision based and have it be an informed decision. Really, really important because otherwise you're just going to miss, and the thing I would say about cap models, just because it's a big part of what we do and what we use is what we learned is there's an old saying in the cap model industry, which is are your models accurate? And the answer is no, but they're sometimes useful. That's how it goes. But the point is they are stochastic simulations. So they are randomized simulations of what losses might be.
(10:12)
That's really what a cap model does. But so you can't trust one versus another. It takes more than one model. Why is that? Because if you look again at flood, which is an incredibly granular peril, there are 3 million square miles in the US so you can't possibly come up with enough event footprints to predict floods everywhere in the US, just not possible. So some models will specialize in, for instance, west, west coast, atmospheric rivers, other models are going to specialize more in the middle of the country by rivers and things like that. So you have to be able to have multiple opinions of loss and be able to put those together. And that's a very, very big challenge. We actually built our own AI-Engine to tell us, if you kind of look at the way it works, cap models look at an event-based footprint that might be several miles, square miles, even more in that footprint.
(11:07)
Well now we're trying to get down to a building level or a part of a building even. So what we did is we built an AI engine to take each building up with an AI capability to give us an idea of which model was more accurate and by how much. And so that's been incredibly valuable and we've proven how good it is. But even with that it's around 90%, a little bit better than 90%. So that doesn't mean it's a hundred percent. So even then we still have to use underwriting technology and people to make sure that we're making the right decisions as much of the time as possible because we know nothing we do is perfect. And I think that will always be the case. So anyway, it's a fascinating problem and I think that all types of property risk is tough. If you look at wildfire, crazy bad, right? It's really, really difficult to predict that. But you're also talking about flood is I think the most challenging one that I know of because it's so granular. You could have one building fine and the one right next door completely gone if you do it wrong. So you really have to know a lot to be able to do these things and assimilate that data and know how to put it into, like I said, a formula or a format that underwriters can actually use. That's a really critical part of it.
Valerie Turpin (12:24):
We do integrate more and more data into the assessment of the risk. And I totally agree with Cory. I mean that we thought algorithm without AI, it's impossible for an underwriter to assess how much if piece of the data will be critical to the assessment of the risk, how to weight them and how to translate all of that into an actionable decision to know, okay, what should be my risk? And then when we can quantify the risk, then we can quantify how much exposure we want and ultimately how much to price it. But they are much more components. And I want to talk and touch a little bit on the regulation because those components are not capturing the models and this is where underwriting needs to keep an open mindset and knowledge and keep going. So on the state of California, so I'm going to speak about the state of California and wildfire.
(13:26)
So you've seen all the big wildfires that we have a few years ago, the glass wildfire. So when the glass wildfires and some other wildfires were around the regulator, the state of California sent a letter to every insurance company and the road to us and they said, I need to remind you something. We are going to use the eminent proximate close in managing our litigation in claims, in case of claims. So what does it mean? It mean that you have a wildfire. So property are destroyed or not property, property are not destroyed, but an area is affected by a wildfire. So the vegetation is gone. And then what happened in California after that, you have a heavy rainfall and the soil is unstable because of the wildfire. And what happened is that you have mudslides and you have landslides and mudslides. Landslides are usually excluded from property policies and they happened six months, one year, two year, three years after the wildfire.
(14:36)
But it's because the soil is unstable because of the wildfire and those landslide mudslides happened and some of them might be more destructive than the wildfire before. And landslide, mudslides are excluded. But because of the application of this rule, this is not a landslide claim. This is a wildfire claim. And the policy enforce at the time of the wildfire will cover the destruction due to the landslides. This is what the state of California reminded us, all of us, this is what happened. This is the claims we paid by the way. So I'm very aware of that. But that is also something that there is no model today when you try to model wildfires was going to tell me that three years later and there's no time limitation in the rules that three years later I'm going to pay millions of dollars of a destruction of property that has zero losses due the wildfire because there was a landslide afterwards.
(15:39)
So are when we speak about the flexible underwriting piece that we need also to evolve with the environment, we need to evolve with the regulations and we need also to consider that there are consequences on each of these cat events to other cat events that are part of the underwriting. So that are great, but we definitely need to get further and the further and the consequences are going to be increasing in the future. So when you do climate risk, you really have to think about the legal environment. You have to think about the facts, the events themself, and how much more frequent they will be and how much more damaged they will be associated to them. You have to think of a lot of different covenants.
Kaitlyn Mattson (16:31):
I think that brings me to parametric insurance and I'm just kind of curious about how that could be used in cases like this. And then we can also talk about how AI can be effective in underwriting.
Valerie Turpin (16:46):
So I'll do that very quickly. So 15, 20 years ago people started to discuss about parametric insurance. So parametrics basically you cover not based on a physical damage or how much the indemnity related to these physical damages, but you based on tangible measurement of a cat. So for example, it started in the industry by ski resorts. We wanted to cover themselves against lack of snow and then measure that if I don't have X inches of snow, I'm going to have insurance. Or also for wind is in the about that. Well, if the wind blows more than X minutes in this area over 39 miles per hour, then I trigger my covers. So you have an indemnity which is based on triggering physical measurable data, but unrelated to the year amount of damage that your property might have found. So in this market where capacity for cat is reduced, price are going up, it's very temp tempting to try to find a compliment of this insurance through parametric deals.
(17:55)
I mean, 20 years ago, only the very large companies were considering about that because it was quite difficult to explain to a board that you are going to maybe not have an indemnity if you have damage. But now a lot of even middle market companies are thinking about these to say I need to compliment. So we do see much more, it's much more interested, they're much more demand of that They are specialist companies that offer these. We don't see a lot of development of those because they are still very expensive. And the reason is that the people who provide capacity for parameters or the same people who provide me capacity, so the pension funds the large renters on that. So at the end of the day they put more or less the same price in Russia was behind that. But are we going with that is that they as the traditional insurance market has much more difficulties to handle the magnitude and the frequency of climate risk. We can anticipate to have much more alternative insurance or border at the edge of insurance solution because you still have a customer that has some needs.
(19:10)
We will see that more and more customers will be willing to put some money to have more comprehensive coverage that they can't get from the traditional insurance world. So paramedic is a solution, but we will see more and more because today in the traditional insurance we are asking clients to take more, to put more skin in the game. We ask them to detect more retention. Not everybody can do that. The lender insurance are pushing for this as well. So we will see much more alternative solution, self-funding parameter deals just to offer a more comprehensive solution for the customers.
Cory Isaacson (19:46):
And I would add that the other advantage of parametric is it makes adjusting the claim very easy. Yeah, because it's a trigger you pay this much money so that that's very, very clear when that happens. And so that does make that part simpler, but I don't think it's going to be a replacement for all insurance. I think maybe it's good for layers but not necessarily for the whole problem. And also I will tell you from recent personal experience of brokers I've talked with, they have a very hard time explaining a parametric product still to customers. And so people are scratch their head and how does this really work? And so I think it will have a place like Valerie suggests, but I think it's going to take time to evolve and when people understand what it is. But I think parametric in combination with indemnity is going to be a great way to go. So I think that'll be a big strength there.
Kaitlyn Mattson (20:40):
Great. Yeah, I'm just kind of curious about what other technologies may be leveraged in underwriting to address different challenges.
Cory Isaacson (20:50):
Yeah. Well I know you asked about AI.
Kaitlyn Mattson (20:51):
AI, yeah.
Cory Isaacson (20:53):
Well one thing that's really important to our company, but I really think everybody in the insurance industry, it was very easy to start an InsureTech being mainly a technologist. Most of my career it was very easy to come into insurance and say A policy is a piece of paper. What could be so hard about that? I can tell you it's very hard. So I don't think that anymore at all. But the interesting thing is how do you apply technology to it in particular AI And what kind of uses do you have? AI means lots of things. Everybody you talk to here today is every vendor is going to say they're doing AI in one form or another. It can be a lot of definitions. Some of them are just marketing, some of them really mean something, you know, have to really look for yourself. But what I know is it's not necessarily the chat GP kind of AI where you ask a question and get an answer back, which is generative ai where the machine is actually looking at data and creating essentially more data and more scenarios to come up with an answer.
(21:54)
What we're finding is that we use it in the property area in a very critical way. I think, like I said, we didn't know, which we knew cap models were different, we would see different opinions, how different, well sometimes 150 times different on the exact same geographic point. What do you do with that? So what we did is we said, okay, why can't we use AI to actually predict what we think the likelihood of flood is at a given point in the United States? And so we evaluated every building we could identify. We did trillions of calculations. It was a pretty massive effort, took us a long time to do. But by doing that we actually came up with a tool. It took us about two and a half years to evolve it. We came up with a tool that was able to actually tell us which model was better than the other and by how much.
(22:45)
And by doing that it's been phenomenal. And then we use that to influence our price. As Valerie said, a common use of cap models is to price on the average annual loss, which is what the model says will happen over a long period of time. But what do you do if one average annual loss is a million dollars and the other one is $10,000? How are you going to make that work? And so that's what we really focused on. So it's paid off very, very well because the combination of the tools has been really, really good. But I think taking one technology and trying to replace everything with it is a big mistake still because these things are all new and maturing and they're going to be a lot different in five years than they are today and it'll be much, much better. So again, you have to weigh everything because none of us know everything we need to know to underwrite anything.
(23:38)
And I'll leave with just one other comment on that subject, which is that the core of insurance is underwriting, that is what insurance is, and being good at underwriting is the most important skill. And you saw in InsureTechs that started out in the industry that didn't know how to underwrite that raised a lot of capital and didn't do so well a few years later. So that was a very common story and we really saw that was a possibility. And the whole way that insurance works is you have, if you're an MGA like us, you work with capacity partners like Valerie's Company Arch is a big partner of ours. But the point is, if we don't keep the capacity partners healthy and profitable, we're going to be out of business really simple. So we had to be very, very good at underwriting first so that we could actually develop that.
(24:27)
So all of our technology and everything else was designed to help make underwriting more effective, not just through technology but through people as well. And that combination has worked very, very well for us and it took us a long time to get it the way we really wanted it, but now it's pretty amazing what we can do. And like I said, our results have been proving it for several years now. So we're really, really happy with that. So I we're just caution everybody here is that when you're talking to any in InsureTech or a company that's going to help you with technology, does it help your underwriting? If it does, it's going to help your business. If it doesn't, well or you don't know how it's going to work, then I don't think it's nearly as relevant as it should be because that is the foundation of insurance is great underwriters, it really is. Our underwriters by the way, not only evaluate policies, they help us formulate our technology. So they actually give us advice and input on, you should look at it more like this instead of like that. So that's worked really well too.
Valerie Turpin (25:32):
Yeah, I couldn't agree more with the interaction of technology. I have to say a lot of traditional underwriters are concerned the fear of not to be irrelevant, the fear also to lose the job, but I don't think that we can replace an underwriter. What we do see with technology is that there's a lot of much more data to digest, there's a lot of augmentation of the data that we need through vendors, through integration of technology, visual technology and modeling and everything. So the augmentation and the preparation of the data definitely can use technology because there's a lot of, we need computing force. We need to integrate all of that to get something into a usable, actionable information because right now there's a lot of data dumped on the desk of the underwriters and we wish for the best. So we need to integrate, we need to help the decision by offering options to the underwriters.
(26:38)
But at the end of the day, there is definition decision that is made using a lot of, much more factor that today the technology can't renew integrated. Like when I said about the legislation and we can discuss about social inflation, I love public justice because the really drive cost of the last app because of some techniques. So that couldn't be replaced today. So I was having this conversation with Cory a little bit earlier and I was thinking that actually the technology doesn't really, and I'm going to be a little bit controversial here, doesn't really threaten the job of the underwriters. But if you remember 20 years ago, a lot of traditional companies decided that there are some non-core low added value task in the underwriting process, data enrichment and all that. So what this company did is that they created offshore platforms. So everybody, were very happy to go to countries where there are good education systems, but the workforce is much less expensive and transfer this task offshores, the volume of tasks transfer offshores has really decreased quite a lot.
(27:54)
And when you look at InsureTech, they don't transfer offshore any of those tasks. They use technology. So to me, the use of technology doesn't really, will not really affect underwriters, people who are work on the decision, but it will definitely affect the platform that has been built. Offshores where it's heavily transactional, the level of decision is pretty limited. And I can see that actually this platform are also evolving by offering us much more added value, much more pre underwriting task because they know that they can be very soon not relevant anymore with the heavy volume of transaction that we are doing because that can be totally replaced by technology and by AI in the future. But the underwriting, the ultimate decision, which has also a component of relationship with your distributions that they are multidimensional. I don't think it's going to be threaten at all. I don't think that or I will be gathering for a long time when that will happen. So I don't see that as a threat. But everything that has the ability to be heavily, which is heavily transactional and can heavily be automated, definitely those jobs will probably be slightly and certainly in the future be replaced by the technology and artificial intelligence.
Cory Isaacson (29:22):
Yeah, I totally agree. And I would say also that the smaller the risk, the more you can automate a bigger percentage of it. So in other words, if we're doing residential homes, which we do, a lot of that can be automated, maybe 70%. But as it gets into bigger and bigger risks and we're doing something that we might do something that's a hundred million dollars or a billion dollars in value and larger limits, no way. I mean it's going to take people for quite some time if ever. I think people are worried, could you ask Chet GBT, should I underwrite this should and what should my price be? I think we're a long way from that. So I don't think that's really going to happen.
Valerie Turpin (30:02):
I would not put my capital on that, definitely.
Kaitlyn Mattson (30:08):
Well, I think we're coming to the end of time. Does anyone have any questions?
Audience 1 (30:17):
Great presentation by the way. Thank you to Corey. You said you're an MGA? Yes. So with the amount of data that you have, especially as it relates to flood and you have it for, you said basically every property. Yes. Right?
Cory Isaacson (30:35):
Yes.
Audience 1 (30:37):
If you're presented with a risk or let's say a program that doesn't write like standard ISO territories and maybe it's a community and it's got non-contiguous areas, how do you underwrite that? How do you price that with all of the information that you have?
Cory Isaacson (31:04):
Yeah, well that's a very good question. And firstly, what we have on every building is essentially we ran AI in every building for the likelihood of flood. So we're not predicting how much damage there might be all about the likelihood of flood in that spot, what we did. And so we used that. The way we would go about it is because we've done that for every single building, we can factor that into all the data we get on each building in addition to that. So is it concrete and steel? Is it wood frame? How tall is it? How close is it to the water? There's a tremendous amount of facts that we actually use to make a hydrological assessment. Our initial goal was can we have our technology do an engineering level assessment of every single building that we underwrite?
Audience 1 (31:51):
And so then would you be pricing it by the property? Each property?
Cory Isaacson (31:56):
We price it by each individual building,
Audience 1 (31:59):
Building, not just property.
Cory Isaacson (32:00):
Okay. Exactly. So if there's a condo complex, for example, that has 25 buildings on it, we'll actually underwrite each individual building on that and come up with a combined price. Okay. We're the only ones I think who can do that, by the way. Yeah. It's a very, very challenging thing to develop. So I can tell you took quite a bit, but it really paid off because again, it's so granular.
Audience 1 (32:22):
That one building could be fine and the other could be not fine. It's also interesting too, because now when someone comes to us and wants insurance, we can usually show them if they have 10 buildings or 20, we can say most of the risk is coming from this one or these two. And sometimes what we'll do is just make the terms more attractive to the insurer on those two buildings and we can make the whole thing affordable. Our goal is to get as much flood out there as possible because flood happens has happened in 99% of US counties since 1996, by the way. So you can't be without it.
Audience 1 (32:56):
So yeah, your discussion about parametric was interesting because I'm now involved in a project where we want to utilize parametric insurance but at a very low limit and trying to define the various triggers for the coverage. And then I think the hardest part for us is getting that data. That's right. Especially when our territory is going to be defined by a community and not necessarily by ISO.
Cory Isaacson (33:25):
There you go. It's an underwriting problem.
Audience 1 (33:28):
Absolutely. And I couldn't agree with you more with the importance of underwriting, so thank you. I really appreciate your presentation.
Cory Isaacson (33:36):
Thank you. You're welcome.
Kaitlyn Mattson (33:41):
Anyone else? Alright, well thank you all. Thank you very much. Appreciate it.
Cory Isaacson (33:48):
Thanks very much.