Get a demo
Select the type of demo you'd like to experience with Indico Data.
Select the type of demo you'd like to experience with Indico Data.
In episode 24 of the Unstructured Unlocked podcast, hosts Tom Wilde and Michelle Gouveia sit down with Matthew Grant, founder of InsTech and a seasoned expert in catastrophe risk modeling. Together, they explore how third-party data and emerging technologies are reshaping how insurers assess and price risk. Matthew shares insights from his 25-year career in catastrophe modeling and offers a behind-the-scenes look at how insurers are leveraging satellite imagery, LIDAR, and generative AI to modernize legacy systems, streamline underwriting, and build competitive advantages. From the evolving role of reinsurers to the challenges of regulatory approval, this episode is a must-listen for anyone navigating the intersection of insurance and innovation.
Tom Wilde: Welcome to another episode of Unstructured Unlocked. I’m your co-host, Tom Wilde,
Michelle Gouveia: And I’m your co-host Michelle Govea.
Tom Wilde: And today, Michelle, we’ve got a great guest, Matthew Grant from Instech Matthew’s got a really fascinating background in risk modeling. I’ll let Matthew introduce himself, but we have a great topic today talking about how to leverage third party data to help in quantifying risk. So Matthew, welcome to the show.
Matthew Grant: Thanks, Tom. Thanks Hel, it’s great to be on the other side of the microphone and I need to make sure I don’t have bad habits that I ask my guests not to have. Very briefly. As I said, I spent actually 25 years in catastrophe modeling. Really sought that from the beginning and really understood the power that insurers can get when you get access to new data, but also some of the challenges. And then just coming out from 10 years ago, we started insec and the really idea of insec is to bring together insurance companies around the world with technology companies and really help both sides understand what are the most relevant areas of analytics to be looking at what’s working and try and cut through some of the noise and hype to the solutions that can really help people make a difference in their day jobs.
Tom Wilde: Perfect. I mean, I think we’re very much sort of in the golden age of insurance technology. I think we have alignment with both the carriers and brokers embracing technology fully and the insureds really benefiting from that. One of the things that we hear from our customers certainly is that while they get a lot of data from the brokers and from the insureds, it’s never a hundred percent of the data that they need to try to figure out risk. And in a weird way, insurance is about trying to take the uncertain and make it certain, and that certainty is in the form of the policy premium and the terms and conditions of those policies. Let’s talk a bit today about how to use third party data like catastrophe modeling to help quantify that risk and the growth in that from things like artificial intelligence, access to data, access to compute power. So maybe set the stage for us a little bit broadly and then we can kind of drill into some of these topics.
Matthew Grant: Yeah, I mean there’s a couple of things, Tom. So first of all, if you were somebody outside of insurance looking at insurance, particularly property insurance, you think, how hard can it be to understand where a building is, what’s it made of? What’s the proximity to hazards like flood? In reality, it’s still really difficult, as you’ve said, for insurers to get that information. That’s one thing. And then when we look at the world of catastrophes and catastrophe is generally a defined, I forget the exact number, but it’s something in the region of a few billion dollars of loss. So it’s a aggregation of loss, typically hurricanes, earthquakes. Now of course, wildfires, flood, severe convective storm, those are things you can’t price, analyze, manage actuarily. So this is kind of where models are come in and means 30 years of building these models to help insurers price better, but really to manage their overall portfolios and being incredibly successful really about allowing insurers to understand what drives the risk, educate their clients, how to manage that risk exposure, and then ultimately be able to write more insurance. They’ve got more confidence of what their total potential loss could be and ultimately everyone should benefit from that.
Michelle Gouveia: There’s been a lot of, I’ll say entrepreneurial activity or startup activity in trying to get either better data, so better quality data that’s digestible by brokers and carriers or new data. So something that has been difficult to obtain in the past. A lot of satellite imagery companies coming about a lot of companies using lidar and visualization to help underwriters or data scientists better identify what you said those materials things about location. But we hear a lot about there’s still that baseline of data that has existed that carriers still want to leverage. What’s the best way to think about how carriers approach incorporating these additional data sets into something that they’ve already built internally that is working but could be improved?
Matthew Grant: Yeah, Michel, if you break it down into two parts, that helps us understand it. So first of all, carriers, whether they’re running homeowners, commercial, will have somewhere information they need about their policy holders, and that’s not just in the new business. Also the challenge is actually how do you get hold of that information because a lot of these systems are running on legacy, so you’re seeing quite a lot of work going on to be able to actually help people manage across different silos. And then the second challenge is once the data comes in from the underwriter, from the broker, how do you efficiently extract that data so you’re not re-keying it? So there’s just a basic, I’d almost call it hygiene. The data’s there, I just can’t get my hands on it. If I’m an underwriter, I can’t get my hands on it easily. We’re starting to see a lot happening.
A lot of that’s been driven by generative AI ability to ingest data. The second one though is when you look at this third party data, you’re right there. There’s a lot more data coming in now from things like satellite imagery. As you said, there’s some quite intriguing things going on with cameras on the ground sensors. The big challenge though, it’s not really a technology challenge. I trained an engineer and I remember my first boss saying to me the difference between an engineer and a scientist, there’s a scientist, can almost get anything done with enough money and time. An engineer gets it done at the right price on time. And it’s a bit like that when you are trying to extract data, there’s only a certain amount of money that’s available really to assess a risk. You’ve only got a certain amount of a premium as an underwriter and you’re going to decline risk.
So the reality is, yes, there could be companies that come up with really clever ways to get information, but the real barrier to adoption is was two really. One is can you do that? Can you provide it at a point that is cost acceptable to an insurance company that’s probably already using some analytics? And the second one is because they’re making decisions on this data, how confident can they be that your new error imagery, location level data about your property is better than what they’re seeing either from their broker or client or they’ve got on their system? Those are two really important thresholds to think about alongside the ability to do it. It’s actually not the ability, but actually how confident and how cost effective is it?
Tom Wilde: Is AI changing carrier mindset around whether they should be building proprietary predictive models or just being good at integrating them? If you follow my distinction here, are they more likely to think, Hey, we can make this a competitive advantage if we have our own proprietary models, or is it really just let’s take best in class and be really good at that synthesis?
Matthew Grant: No, it’s a great question and the answer is there are definitely companies out there seeing this as a competitive advantage. I mean, you’ll know this, well, your listeners are probably familiar with the technology adoption curve. Essentially companies individuals move a different speed. Some of this have queued up to get an iPhone. Some of it’s waited two years until it have been properly sort of robustly delivered and insurance companies are the same and people in insurance companies are like that. We did an event in London a few weeks ago and we had the head of modeling for Zurich Insurance, one of the biggest insurance companies he was really excited about. And when the Swiss got excited to pay to pay attention, he was really excited about the use of AI for them actually to be able to build their own models. And they built a terrorism model using ai.
This is an individual, I know he’s an engineer, he’s very thoughtful. They were definitely seeing a competitive advantage because the generative AI tools now and another more traditional AI tools do allow you to much more quickly perform the analytics and deal with that uncertainty in there. So yes, there are some doing that. I think that I’d say most of them, just because where we are in the technology option curve are probably looking for uses of, and I’m just going to keep talking about Genta. I think we can all intuitively understand it. That’s more on what Michelle’s talking about, which is the efficiency play. How do you more efficiently get data in? How do you more efficiently write emails to your brokers and your clients? And I can talk a bit more about, there’s some parallels to what happened in the past about how some insurers started to use analytics when we were still using spreadsheets and didn’t have ai. But I think we can see playing out here as well.
Michelle Gouveia: We’ve made this distinction a few times about data coming in and using it, and I could be wrong in how I’m thinking about it, so please do correct me if the piece of this question is wrong. But using AI to do the actuarial modeling potentially using the data that you’re getting, and then there’s using AI from an underwriting standpoint as that information comes in to add some efficiency. But those two things do need to be in concert with each other because you need to be assuming similar things about your whole portfolio or book of business. So I’m curious if you think that from a cat risk perspective, it’s more imperative that AI be used correctly on the data scientist side or on the underwriting side?
Matthew Grant: Well, somebody to say both meaning after
Michelle Gouveia: Oh yes, yes, you should do it accurately on both, but what’s answered maybe of adopt, I
Matthew Grant: Suppose I might disagree a little bit. I think you can do this incrementally in steps and I mean the example I just used was Zurich and they’re doing this for one book of business. So most companies in life we don’t change overnight. So what tends to be happening, people might be looking at one area. So for example, severe convective storms, tornadoes, hail, really hard to model. Traditionally, we probably all know these are happening more frequently. AI can actually provide much more sophisticated ways to actually do some of the analytics around those and understand them. But essentially it’s picking out where traditional models have failed. So you could have an insurance carrier goes, right, we want to focus on our severe convective storm risk exposure and understand more about our portfolio, understand more about our aggregation. Now that’s partly that’s underwriting because the underwriter part of good underwriting is you don’t write too much risk in one place.
You’ve got to manage your aggregation, your data scientists and your mostly people as well are also recruiting now PhDs who could have built models themselves. So to your point, they’ve got the skills internally, they know the limitations, they will basically look for areas that the traditional modelers haven’t been able to solve or haven’t chosen to solve. Test it out. And often what happens is you see this, you get your leadership advance about building a model, and then two years later there’s a commercial model available and you start to see these companies books saying, well actually now the commercial world has kept up. We don’t need to keep building our own model. But they get an early advantage of understanding more about the risk. And one thing, I mean it’s worth mentioning as well, is it’s always tempting when you think about insurance to think about insurers saying no, but actually underwriters like to say yes.
And one of the interesting things that Sarah Russell who launched bellwether out of Alphabet or Google X was saying when we spoke recently was that what they’re looking at is where the industry may have perception that the risk of wildfire is too high. If you’ve got confidence in the data and you use the analytics, well, you can identify where in fact the actual risk is not as high as might be perceived and therefore you can offer insurance and you can take advantage. So I think it’s really important to understand that that’s an early mover advantage is if you trust your model and your data, you can actually take, and again, everyone benefits, people who couldn’t get insurance can now get it. And we’re seeing quite a lot of examples of that in different ways of people being innovative about better pricing of risk because for all the reasons we talked about, you can be more confident and more active.
Tom Wilde: The reinsurance folks who have been super active in this for some time, right? Swiss three, Munich Re, those kind of folks with catastrophe modeling and other models that they’re commercializing talk about is this access to data is the commercial incentive to really help sort their collection of customers. Where does the reinsurance segment play in terms of this category?
Matthew Grant: Well, I mean it is a little bit like I just said actually, Tom, without giving away any secrets, I mean mini cre, Swiss Street as you said, were really there amongst the earliest of 25 years ago along with million companies about building their own models. I mean the benefit they had of course, because a lot of their business is not intermediary by a broker, they get direct access to the insurers that get access to claims data, access to claims data is really important when you’re building a model. So they were able to build models, but the reality was at some point they too said actually this, we talking gone as far as we can. We can either pretty magnitude of 10 investment in modeling or we go and work with some commercial modelers. So they’ve been quite sophisticated there about looking at, again, IC build versus buy model and reinsurance traditionally wanted to be quite agile and London lawyers is probably when it works well.
One of the best examples, you have a very lean team, you work out where you buy effectively your analytics and you don’t really want to build up a strong modeling team even if you could. So I think that’s good. I’d say yes, they’ll build it when they need to build it, but if you’ve got a commercial solution that works, and particularly you can actually tweak the dials and levers so it can represent your view of risk mean, then you’d be better off buying commercial the same way. Would you build your own computer these days or would you go and buy one off the shelf from Mac or whoever else you’re buying one from?
Michelle Gouveia: In thinking about how fast AI moves and how quickly decisions one get made via ai, but that carriers need to make about using ai, what do you foresee or have you seen the concerns about leveraging AI more so in the cap modeling and what does that, I’ll call it review period or how long is that decision making tale to say this really is additive to our modeling or it does improve our underwriting approach? How does that change now with AI being in the mix?
Matthew Grant: Yeah, last year I talked to a few people in the roles who when we talked about building models, assessing models, and they were basically not allowed to use chat GPT at work because the insurance company are still trying to figure out what is this new tool, how dangerous is it, you can’t use it to work. I mean that was a massive barrier to what they were doing that’s now changing. And the second area about AI and the use of AI is the regulators. And we all have heard about how the regulators are concerned about the use of AI for consumer applications. Insurers are really careful using it there. But of course when it comes to catastrophes and insurance pricing and every US carrier will be very familiar with is yet in the US the ratings have to be filed to be approved by the insurance commissioner.
The insurance commissioner is a political post. And so there’s been quite a few situations that we’ve seen and most recently in California with wildfire where either regulator can make it really hard for an insurer to use a model that looks at the more recent history and is pushing back on longer term views. And so even though AI might be better to actually give you a invo better answer because there’s still a constraint on using it. And so therefore that business, fortunately London is there as a market that will pick up what is being declined in the us. But that’s a really important one to remember because the regulators have got such an important role to play and we’ve just seen how slowly that they come back from that they have now changed their mind on wildfire in California. But it does, Florida still goes through a really extensive expensive process for reviewing the hurricane models every couple of years before they’re approved for rate filing.
Tom Wilde: Do you think that, you mentioned Google X, which I think is a pretty fascinating idea, creating a digital twin of the earth. Is there a point in time where catastrophe model becomes much more certain and much, much less perspective? Are we on a path to that? If you look at the availability of data, availability of compute, the availability of ai, are we on a path where there’s much more certainty around things that today seem very uncertain?
Matthew Grant: It’s a curve and you get closer to the knowledge, but you never get to the full knowledge. And the challenge is nature itself is in the short term, is very unpredictable in the long term you can predict it. Although of course climate change starts to add an odd dimension to that, it’s hard to totally predict. So I would say the answer Tom, is if you look at, let’s take at hurricane model, make it very specific. Hurricane models has been around for 30 years, they’ve been through three or four generations. You get hurricanes every two or three years. So there’s good loss data to calibrate them with. I would, I mean I’m sure somebody will that take me up on this, but the potential to get a better hurricane modernized really potentially for the US anyway is really marginal. They’re not really going to get a better model, but we’ve still got all the massive uncertainty. You saw it last year about what is the hurricane activity going to be? Even looking out four weeks into the season ahead, last year was a really strange year thought
Tom Wilde: Hurricanes within three weeks on the west coast of Florida.
Matthew Grant: Yeah, exactly. But nothing for most the season and one at the very beginning. So short term uncertainty is huge. You’ve got the physical characteristics of the hurricanes are still really quite surprising. So I think to answer your question, and this is kind of where the underwriters have advice, this is what their skill is, is you’re working in a world of uncertainty. How do you understand and be aware of uncertainty but still made decisions and not get paralyzed by the uncertainty in there. So I’d say the other thing is although CATA modeling has been around for 30 years, mostly it’s been for portfolio management and it hasn’t really been fit for purpose for individual location level underwriting, I mean flooded. If you try to use a model to price watch out 30 variables in there between when the rain falls and how much goes in the ground, you’re kind of kidding yourself really. If you’re going to try and use that accurate now, give you an indication, but you shouldn’t be trying to use that silly to price.
Tom Wilde: There’s always this tension between should an insurance company be in the business of loss prevention or loss protection in high risk areas. It still seems that it’s very slow for insurance companies to play a more active role with the insured in helping them figure out how to avoid losses in the first place. We’ve seen examples from the Palisades where some houses survived very well and the next door neighbor would completely flatten. The same is true of the Florida to hurricanes. You see those aerial photos where five houses are look pristine amidst a debris field. Is that a changing attitude do you think? Or will it continue to be more passive that hey, if it happens, we’ll pay for it, but we’re not really in the business of helping you prevent it in the first place?
Matthew Grant: Yeah, I think the carriers credits insurance credit, they want to help prevent it. If you like Chubb have got a big mandate to prevent, and of course you’ve got to distinguish between homeowners and commercial, but even in commercial it is really difficult and mean insurance transactions, people, it’s a grudge purchase.
Most people, either they’re buying it individually or they’re with a broker, it’s just like you want to get it done, you kind of shop on price. If someone sells you well, you’ve got to prove, show me photographs because you removed all the vegetation or things like that. It just gets in the way of it. And so I think that’s part of the problem. I mean in Florida, and I think this is still happening a few years ago, but there were insurance companies that were looking providing credits, and I think even the government was looking at this where if you had your roof tied down or nailed down in a certain way, used tie downs,
You would get credit on it. But I just want to add example. It’s a while ago, but there was Andy Thompson who’s now got his own business safeguarding sensors for earthquakes, but for a number of years when he was an engineer, he was lucky to see could he help his clients pay for the cost of retrofitting for earthquake with a reduced insurance cost And it was virtually impossible. And part of the reason for that is what happens back to the point about reinsurance is particularly in this excess and surplus lines where you’ve got high risk businesses coming out of the US and being insured by people that are focused on this high risk areas, they might be able to give a favorable price because the transaction level with the client, they see the benefit. Unfortunately, when it all gets rolled out for reinsurance and you get out to the market, the reinsurers probably don’t allow for that because it’s just out of the normal. And so nest to a company like Factory Mutual where your whole business is based around ensuring people that have got better engineer structures. Unfortunately at the moment it’s a lot of friction in the system to actually make what should be in everyone’s interest, ultimately work properly.
Tom Wilde: What’s always struck me as peculiar as a consumer, I’ve never gotten a sort of risk report card from any of my insurance companies about my house, about my health as it relates to life insurance, about driving patterns. And it just seems like that would be very useful. But I’ve never seen any such thing from any insurance company I’ve ever worked with as a consumer. Obviously I find it a little bit peculiar because they certainly have this data, they’re using it to write the risk. Why don’t they share it and say, here’s some things that will long-term protect your property, your life, whatever it might be. So it’s interesting.
Matthew Grant: I’ll tell you why Tom though. I mean I’ve got a theory. I mean you and I, I’m sure Michelle, we’re responsible to citizens. As we maintain our property, we look after it. And you’ve seen this happening with telematics and driving as
Tom Wilde: Well. Yeah,
Matthew Grant: Essentially, if people are going to start offering us discounts, then at the moment we’re subsidizing all the people that are not as diligent as we are. It doesn’t work. You can’t go and charge people more because somebody else will undercut them and price it. So again, it’s a little bit of friction in the system about why that doesn’t work. I think that’s a reason for it. Call me be a cynic, but I just don’t think it’s going to change anytime soon in that model,
Tom Wilde: Right? If you tie it to pricing, I agree, that becomes a challenging one.
Michelle Gouveia: The carriers have tried it or are probably still trying it. I was part of a team that was trying it at one of the carriers I worked at for a bit, and outside of what Matthew said, which is a limiter for sure, there’s also just that constant desire to make sure your data is accurate or how quickly does it change. So that constant question of is our insured even going to care because if I get one data element wrong on this, they’re not going to believe any portion of it. And then there’s what does that really drive from an engagement standpoint? Is it better that I’m giving them this information that I know I have, they may not know I have? Does it harm that relationship? And then what’s the best way to communicate it? I think there there’s been partnerships with Google Alexa before. Are they mailers? Do you send them out as part of the quote or post bind? So there’s just a lot of just execution questions there. I don’t disagree that it makes sense to come from those people that are the ones assessing you based on the data that they have, but I think there’s just a lot of concern for miscommunication as part of that. And so that’s another challenge, more of a qualitative
Matthew Grant: Challenge. Can I give an example actually as well, IT motor again, and it may be anecdotal, but I remember hearing on a podcast from the founder, I think it was by miles, and they’re saying that, I dunno how this works in the us but in the UK when you take out auto motor insurance, it asks, where’s your car parked? Is it parked on the road? Is it parked in your drive? Is it parked in your garage? And when people said it’s parked in their garage, they actually loaded up their price because it was actually turned out. That’s a leading indicator for fraud because unless you live in somewhere like Minnesota, and it’s snowing all year round all the winter, but it certainly in the uk, the weather’s never so bad. You need to put your car in the garage for day or two of the year. So the point being people who said, I put it in the garage are on balance lying because they’re trying to gain the system, so actually penalize them because they’re lying at the point of entry. So even when you ask people for that information, Michelle, it’s bit to your point, are you going to validate what they said or do you have the risks that they’ve actually lied to you and then you’ve got an even worse exposure because A, they’re lying, and B, they don’t have the protection in place. They told you they did.
Tom Wilde: But maybe the obligatory wrap up question, what jumps out at you is sort of the biggest changes to this category, catastrophe, remodeling and third party data kind of in a three to five year horizon.
Matthew Grant: I think it’s where you started Tom, which is what I find just tremendous, is there’s so many more companies coming out now to help and offer solutions, just different parts of the value chain solution. Whereas, and I’ve worked for one of them five years ago, there were two modeling companies that pretty much dominated space. We were worried about hurricanes, earthquakes happened at those stages. Today you’ve got all these new perils coming out. We talked about, we didn’t talk about cyber, but that’s out there liability. There’s more granular data, we’ve got better access through platforms. You’ve got the generative ai, the cost of building a company is much better. There’s more capital around. Thank you, Michelle. So what we’re seeing, and again, we sort of source evidence by our recent conference and some of the people we’re talking to, a whole lot of companies that are coming out and really quickly offering value around the margins of what the core companies are doing. And also what’s really nice about this space is that it’s really collaborative. And because the number of insurers might now use two or three models or data sources, the competition is not as aggressive. It might be somewhere else. There’s room for everybody in that space. And so you get this really great collaborative effort. So I think what do I see coming? I see more companies offering real value, probably more niche rather than dominating the market or mainstream. But they’re all, well, the good ones are performing a really valuable service and there’s money there. And I think as I said earlier on, they’re actually helping people write and offer more insurance as opposed to more reasons to say no. Great.
Tom Wilde: Well, Matthew, thanks so much. We’ve been talking to Matthew Grant from Instech. I am your co-host, Tom Wilde,
Michelle Gouveia: And I’m your co-host, Michelle Gouveia.
Tom Wilde: Thanks, Matthew. Great conversation.
Michelle Gouveia: Thanks for joining us.
Tom Wilde: Thank you.
Check out the full Unstructured Unlocked podcast on your favorite platform, including:
