Are property insurance markets prepared to manage escalating climate risk?
Climate change related disasters are become more frequent and more devastating, and wildfires are now the fastest-growing source of climate related damages in the United States. 2025 brought two catastrophic wildfires to Southern California: the Palisades Fire and the Eaton Fire. Together, these are costliest wildfires in U.S. history.
One likelihood is that homeowners insurance will be increasingly hard to find and afford in areas prone to wildfires in California. State Farm — the largest insurance company in California — has been declining to renew homeowners policies in wildfire-prone areas for a while.
This has led increasing numbers of people to rely on the FAIR Plan, the insurance provider of last resort for California’s homeowners. As a result, the FAIR Plan is going to need an infusion of at least $1 billion from property casualty carriers that operate in the state to pay claims resulting from the Pacific Palisades and Eaton/Altadena fires.
Conventional wildfire risk models used by incumbent carriers rely on historical data and failed to adapt to changing wildfire behavior and environmental conditions. This may change as California’s Insurance Department will allow carriers to use computer modeling of future exposure when setting rates going forward. And it may seem odd, but not all carriers use the same information to price wildfire riders and policies. Some carriers are low-information underwriters, and this is not a good thing for the long-term health of the market.
In the face of escalating climate risks, the availability of accurately priced property insurance provides crucial protection from devastating financial loss. Launched in 2017, Delos Insurance Solutions a specialty MGA (managing general agent), is at the forefront of transforming property insurance in climate-affected regions. Delos is currently active in California, writing homeowners policies in areas that other insurance providers are avoiding. Delos believes that advanced data analysis, machine learning, and in-depth domain expertise will allow it to stay ahead of the state’s rapidly evolving wildfire threats.
Delos is leveraging new types of data and deploying proprietary algorithms to process satellite imagery down to the address level. High-resolution weather forecasts, detailed wind maps, and an understanding of precipitation and vegetation moisture developed by Spatial Informatics Group (an environmental think tank) also inform the company’s modeling. In addition, Delos works directly with homeowners to harden their homes to help them obtain coverage, because prevention is far less costly than paying out after the fact.
Notably, Delos accepts 65% of the business that the primary market is currently declining because of perceived wildfire exposure and says it has experienced zero wildfire losses from the devastating 2025 LA fires on a portfolio of 25,000 insured California properties.
Based in San Francisco, Delos was founded by Kevin Stein and Shanna McIntyre, a physicist and aerospace engineer who serves as Chief Data Officer. Delos has raised a total of $22.6 million in funding. Their latest round was a $9 million Series A announced in October of 2024 which was led by HSBC Asset Management.
In 2024, Delos did $75 million in premiums and had $28 billion in exposure across homeowners, landlord, and vacant home policies. They are adding 2,500 new policies and another $6 million in premiums per month.
Q. Shanna, why has Delos begun with wildfire risk?
A. My co-founder, Kevin Stein, and I wanted to use a lot of these really great datasets and methodologies that have been emerging in the last decade to help people with climate change, and as Californians, we were very aware that wildfires were likely to increase due to climate change and that insurance is a way to help people.
We started there and immediately found that, yes, this is a problem that needed solving and we haven’t stopped since.
Q. What does it mean to say that wildfires are specialist perils?
A. What we mean is not generalist, at least not anymore. Most people get their wildfire coverage through their homeowners insurance and most of the perils that are covered under a homeowners policy — at least from a regular, admitted carrier — are seen as generalist so you can use fairly traditional actuarial methodologies to figure prices. And wildfire used to fit that description. It wasn’t changing that rapidly. You might have had a few bad years here and there, but it wasn’t bad enough to merit a specialist kind of approach by the entire industry.
Obviously, that has changed. In 2017, we had the really awful, surprising wildfires in wine country, and in Southern California as well, and then of course the terrible year following, with the Camp Fire, the Carr fire, etc. It became clear that the traditional approach of the industry was no longer working very well. Our opinion is that this needs to be treated as a specialty kind of peril that requires much more targeted resources to do it right.
Q. Does that mean that there’s a separate rider for wildfire risk?
A. We are not doing that. We are selling whole homeowners insurance policies that include wildfire coverage. But because a lot of homeowners are being dropped by their insurance carriers, they do end up with a split – two policies that cover what should be covered under one. Usually, they get their fire coverage from the California FAIR Plan and then anything the FAIR Plan doesn’t cover they can add back in with a difference in condition policy, or DIC (also known as wrap around).
So already there are a lot of people out there who are getting the fire peril split out and covered by the FAIR plan but what we are doing is bringing it back to being fully covered by their homeowners insurance.
Q. Why was an MGA the right structure for you?
A. We considered offering this as a service to carriers, but when we thought how wildfire underwriting and the science needs to operate in order to be done correctly for markets where things are changing really quickly, we came to the conclusion that doing it ourselves, and having it be really close to the wildfire expertise that’s on our team, was going to produce a much better result — especially for the insureds.
The nature of being a vendor means that it’s a pretty bad experience for end users to have to change quickly. When we looked at how fast we could move compared to some other vendors (not to beat up on them) we saw that they need to move slowly to let all those carriers adjust. We’re the only entity consuming the data we produce from our models, so it allows us to be a lot more nimble, and we think that’s going to end up being a better experience for the insureds. We’ll be a more stable source of coverage for them.
Q. Delos has a unique partnership with Spatial Informatics Group. How does it work?
A. That’s been a fantastic partnership. We’re very lucky. Spatial Informatics Group has been doing this kind of analysis for several governments for decades. They have a very seasoned teams that knows these datasets and has the scientific expertise to analyze these kinds of disasters.
Our partnership is their first foray into insurance. They have an equity stake in us and that is on purpose — we really like having them as part of the team for the long haul and having everyone aligned. It has absolutely worked out that way. They’ve been extraordinary partners. They have a lot of proprietary datasets that are very rich and powerful for this exact problem. Having access to that data and that team and knowing we’ll continue to be partners as we move into the future has been fantastic.
Q. What’s the source of the satellite imagery you’re using? Are the images publicly available?
A. The raw satellite data that goes into the proprietary datasets isn’t totally different from what other wildfire modelling groups use. We’re using LANDFIRE or LandSat and definitely some other interesting datasets alongside that, as well.
But it’s what we do with that data that makes it particularly powerful and proprietary. We can take that data and ask what it says about the vegetation that you can’t get just by looking at the satellite image or the raw data on its own. Having access to these world-class experts who can take that data and extrapolate it and manipulate it into something that is far more predictably powerful and useful for a wildfire hazard model is what makes it special. So, the source of the data itself? We’re not using anything that’s terribly different. But the processing of it is what’s proprietary.
Q. What prevents incumbent carriers from simply adding more datasets and refining their models?
A. The processing that has been done on a lot of these datasets to get them into a format with the precision and resolution that we need to get the accuracy that we’re getting requires a team of PhD-level experts and it takes them weeks if not months to download this hourly data and process it. And then it takes immense resources to do that computation. That right there is a big barrier. Also, knowing that this is the right path to take and knowing which datasets to throw those resources at — these researchers have been doing this type of scientific work their whole careers. They understand the datasets and they understand how to take precautions. They understand how to use it in a way that is proper for the problem we’re trying to solve. That known-how is very important and we don’t see that kind of expertise inhouse at a lot of carriers.
Q. Are you able to hire all the engineering talent you need? ML engineers are in demand and they command very high salaries.
A. We definitely offer competitive salaries — we’re a post Series A venture-backed company. But we have more to offer than just a good salary. We are solving a problem that people find very emotional, very real. A lot of people want the chance to make a difference and care about climate change. Our team has a fantastic culture, and as we grow the team, we want to make sure we preserve that. And, also, we are pretty specific about what we need. Not all machine learning researchers or engineers are the same. There are probably some brilliant natural language processing AI experts out there who could learn how to do this, but what we prioritize are experts who have worked in this field, or in a field adjacent enough that they understand these datasets and the science that goes into it.
Q. How do machine learning methodologies help you ensure that there are no surprises from tail-end risk events?
A. One of the main benefits of using machine learning is that it is able to pick out lots of different combinations of conditions that can lead to increased fire hazard instead of starting with the assumption that there’s one equation to figure out the hazard at any location. I think most people have an intuition that that’s not really true. The factors that will be important in urban Southern California locations that have Santa Ana winds will be very different than the factors for a community in the Sierra foothills. Machine learning is very good at understanding the weight to put on those different variables depending on the location and other conditions. It’s not one size fits all. It’s able to capture all those different combinations with a high degree of accuracy at a large landscape scale.
We can do that in a way that looks at conditions today. If you are locked into the fire event history of a particular location, that’s still probably going to be wrong. In a lot of places the vegetation might have changed, the climate has almost certainly changed. Development and land use have probably changed. Knowing the conditions at the time of a large fire event and asking where we see those conditions today, that’s something machine learning can do a lot better than some of the older methodologies.
Q. Some of the factors you use in underwriting can change substantially from year to year. (Precipitation and vegetation moisture are just two.) What does this mean for policy holders at renewal time? Are people going to see premiums bouncing up and down?
A. We’re aiming for that not to be the experience of our policy holders, and it’s an excellent question. We build that into our model, and this is where using a model that’s specifically designed for insurance underwriting is going to be more appropriate.
We know that whatever we predict at the moment of underwriting had better be accurate for at least a year, but ideally for much longer. We want our insureds to feel some stability and we want our agents to feel that stability as well, so we build some conservatism into the modeling. But that’s the nature of fire hazard, anyway. We get some pretty bad fire years but they’re not every year so we want to smooth it out over time and capture the worst-case scenarios we might see over time and cover for that.
There will be some changes, and it can be in both directions. We might see that there’s an area where a certain type of vegetation has become more predominant and that might mean a higher fire risk, so every year we re-underwrite everyone in our book and if the hazard exposure has changed to a level that we find not acceptable then we’ll nonrenew. But there can also be positive changes. If we think the suppression resources have improved or if a community has done some sort of vegetation treatment that we think is important toward reducing the fire hazard, that can make things more within our appetite for risk.
Q. Is it possible, as we get smarter about accurately pricing risk, that fairly priced policies won’t be affordable for homes that are built in the wrong places?
A. It’s hard to say at this point. We have built conservatism into our model. Even building that conservatism in, we’re finding that about 65% of homeowners looking for coverage with us are within our risk appetite. That means that we think the majority of people looking for coverage with us are okay in terms of wildfire hazard exposure — and that’s with a lot of conservatism. Our models are not right now seeking the answer to that. So from a scientific point of view, I can’t say for sure. But it could be possible.
The FAIR Plan is there for a reason. They are aiming for actuarially sound pricing. There will be always some entity that is willing to price a risk and maybe in some cases the only one doing that will be the FAIR Plan. So we’ll see what happens. But I also have high hopes for progress in better analytical understanding how to harden homes and harden communities, so if today some areas would be considered unaffordable, in the future those homes could perform enough hardening against wildfire that they could obtain affordable coverage. There are a lot of efforts taking place right now, in the aftermath of the LA fires, to figure out how to make these communities safer. The CDI has obviously put a lot of effort into having insurers respond to these kinds of things.
Q. Has Delos considered the potential of parametric policies? Do they make sense for wildfire risk?
A. They can in some cases. In the markets we’re going after right now, they wouldn’t necessarily make sense for us at this moment. We’ve been asked about it and we think it’s interesting, but the big need we’re seeing right now is for HO-3 homeowners policies or similar types of products. The need is very big and our interest right now is in addressing that need.
Q. What is happening with reinsurance for wildfire risk?
A. What I predicted at the time (and I’m very happy to have been correct) is this: Reinsurance is prepared for this. The response to these LA fires has been a lot more day-to-day than to what happened in 2018. 2018 provided a big lesson for reinsurance. It’s clear that they’ve learned from it and adapted. I don’t think anyone is panicking. It’s known that these events can happen.
Q. California has in recent years been investing billions of dollars in improved forest management. Can California’s forest management efforts outpace the effects of climate change?
A. It’s very TBD. There’s a known amount of vegetation that ought to be managed. The state has been making progress in urging more controlled burns. They are easily one of the best tools we have to get the vegetation into better condition for a safer environment. I think it’s too early to say with what’s going on at the federal level to say how much they’ll be impacted, but at the state level there’s been a lot of movement to enable a lot more controlled burns. There’s a lot of work that still needs to be done but we have a chance of getting there.
Q. There are supposed to be wildfires.
A. Yes. To answer your question more thoroughly, wildfires have been part of the ecosystem in California for millennia. We’ve learned how to adapt and we can keep moving in that direction. There are some areas that are going to be impacted by climate change where it’s going to require a lot of effort to make them safer.
Q. Can you explain what the wildland-urban interface (WUI) is?
A. In general, it refers to where developed areas are mixed in with wildland areas. An exact definition has the number of housing units per number of acres. Sometimes it can include how much of that wildland is covered by vegetation. These are generally considered to be at higher risk for wildfire although according to our models that isn’t always the case.
Q. Can you model the failures of PG&E management and equipment? They cause a lot of very big wildfires.
A. We include lots of ignition sources in our modeling. Understanding how things can go wrong is important but it’s not the only thing to model. Understanding the kinds of conditions that can turn ignition into a wildfire that moves quickly and gets out of control before suppression resources can manage it well — that is what our model is good at doing. There can be areas where you have ignition but it doesn’t really go anywhere and other areas where that ignition can get pretty scary pretty quickly. Our model is focused mostly on that.
Q. To underwrite accurately I imagine you have to understand how firefighting strategies are evolving. How do you go about doing that?
A. We can pull in a lot of data from CAL FIRE, and there’s a lot of interaction between CAL FIRE and Spatial Informatics Group because SIG has done wildfire mapping for CAL FIRE in the past. The experts at SIG talk directly to the experts at CAL FIRE to understand strategies — even what the firefighters are thinking when they’re boots-on-the-ground in the middle of an event. So, yes, we put a lot of effort into understanding how suppression strategies impact a fire and how they’re changing.
Q. How important are the structural qualities of buildings in your underwriting?
A. It is important! And, frustratingly, it’s not one size fits all. But we have done a lot of internal R&D and also pulled in a lot of excellent analysis done by other groups to understand the home-hardening actions a homeowner can undertake that do make a difference in reducing risk, and we certainly use those factors in our underwriting.
Q. If, for whatever reasons, the cost of Canadian lumber were to rise for American buyers, or the availability of construction workers in California were to decline, would this translate into higher homeowners premiums?
A. Yeah. That’s unfortunately the nature of the beast. It wouldn’t be because we would think that someone needs to pay more premium for the amount of coverage that they get. It’s that they’ll need more coverage.
Q. It’s going to cost more to rebuild their house.
A. Exactly. The estimate for how much it will cost to rebuild their home — Coverage A — would likely need to increase, even if we don’t change how much we charge in premium per dollar of coverage. If you need more dollars of coverage, then the premium will go up.
Q. The U.S. owns 46% of the land in California and the latest large fire in Southern California, the Hughes Fire, started on federal land. But the federal government has recently taken actions which will degrade our ability to fight future fires (freezing funds allocated by Congress, firing workers, postponing efforts to reduce flammable vegetation in areas vulnerable to wildfire, and more). How can underwriters anticipate and account for this type of self-harm?
A. It’s too early to know how permanent these changes are. Of course, we think that spending some resources on keeping us all safer in how we manage federal lands is a good idea. But there have been some lessons learned in how CAL FIRE and the U.S. Forest Service interact. They’ve definitely done a lot to improve how some of the hand offs work. My hope is that even if there might be a different amount of workers that CAL FIRE might interact with, we can keep improving that.
Q. What’s next for Delos? More states, new perils, both?
A. Right now, we’re working to develop new partnerships, increase capacity, write more policies. A lot of that is focused on California, but yes, we are working to expand to other states that face similar challenges that California has a bit of an unfortunate head start on. The Western U.S. and Rockies are where we’re targeting next. That’ll be in the next few years — we’re working as fast as we can.
Q. Are you hiring?
A. A lot of the roles we’re looking to fill are extremely specific, and so we have found the need to use some very targeted recruitment efforts. But, if people are interested, they can send information to us at work@getdelos.com. Especially people with any sort of wildfire modeling or insurance-specific backgrounds. Those are, of course, always interesting to us.
Q. What else would you like us to know about Delos?
A. We didn’t talk about the way we responded to the LA fires. The point of us going into these markets is that we believe we have more accurate wildfire underwriting models. We were really pleased with how our models performed in the LA fires. Because they continue to be accurate, we have not seen any need to pull back or change our view of risk. We continue to offer coverage to homeowners just as we had before. We’re really proud that we’re able to continue doing that. That’s the main goal — that’s why we’re doing this in the first place.
# # #
