Next in Tech | Ep. 258: CEO Series: Wildfire Risk With Bryan Spear

View Full Transcript

Eric Hanselman: Welcome to Next in Tech, an S&P Global Podcast where the world of emerging tech lives. I'm your host, Eric Hanselman, Chief Analyst for Industry Research at S&P Global. And today, we have the next in our CEO series, where we get some insights into how businesses are putting AI to work. With me today is Bryan Spear, the CEO of Technosylva, where they work to mitigate the growing impact of wildfires and extreme weather by delivering real-time insights and predictive modeling. They work with electric utilities, fire agencies, and insurers worldwide to enhance their ability to make critical decisions to protect lives, infrastructure, investments, and ecosystems in both daily operations and long-term planning. Bryan, welcome to the podcast. 
 

Bryan Spear: Thank you, Eric. It's an honor to be here. 
 

Eric Hanselman: Well, it's great to have you. I guess before we go too far afield and start getting into the details of the AI side of things, I wanted to find out sort of what are the problems that Technosylva is addressing? And what are the kind of decisions that you're working with your customers to enhance? 
 

Bryan Spear: Yes, sure. I guess that will give me a bit of an opportunity to just kind of give some context on why we do what we do at Technosylva. And look, I'm sure it's no surprise to your listeners that extreme weather-driven catastrophic events are becoming more frequent. But what's most concerning is the increasing intensity and the destructive nature of these events. And for example, 1/3 of all of the $1 billion weather events have occurred over the past 5 years. And just last year alone, the global economic losses that are due to the extreme weather events reached $300 billion. And that's a big number, but that's been increasing by 20% per year going back to 2015. And that's just not sustainable for the insurance industry and other industries as well. And wildfires have been a big part of that trend. And you, your listeners, you've seen the headlines just last year, Southern Europe, so Spain, France, Portugal, Greece, they saw some of the worst fire seasons either in history or at least in decades. Canada ranked second only to 2023, which was double the previous record. And then, of course, here in the U.S., the L.A. fires dominated the headlines, most economically destructive fires in the history of the world, sadly destroyed over 16,000 structures and claimed the lives of about 30 people. So what used to be a once-in-a-century event is now happening every 25 years, and most experts would agree that, unfortunately, this trajectory is only going to continue. And to go a step further, electric utilities are often finding themselves at the center of this conversation. Utilities are -- they're responsible for a small portion of the overall total -- overall ignitions, but they're responsible for about half of the most destructive fires. And it's because these fires occur when the conditions -- the winds are high, the fuels are dry and then by definition, their infrastructure is close to homes and communities. And that's why the outcomes are so destructive. And you've seen these over -- just over the past several years, you saw Lahaina, the most deadly fire in the U.S. in over a century. The 2021 Boulder fires, the Smokehouse Creek fire, was the largest in the history of Texas. And utilities are now being held liable for those damages that are associated with their ignitions. And this has really created an existential threat to the industry. So any way you look at it, the data shows beyond the shadow of a doubt that these major wildfire and extreme weather-driven events are increasing in frequency and intensity. So that's the problem that we're solving. And then what we're doing about it is, first off, we're a very -- Technosylva is a very mission-driven company. Our mission is to reduce the impact of wildfires and extreme weather, and that drives everything that we do. And you mentioned it earlier, what we do is we're helping our utility customers, insurance companies, fire agencies predict and prevent catastrophic fires and weather impacts, and we're doing that by using cutting-edge science and AI-driven modeling. And I'll break it down into the 3 different customer types on the way that we're helping them do that. First off, with utilities, we're help -- what we do is we help them prevent catastrophic ignitions, but we help them do it in the most cost-effective and the least disruptive manner. We're working today with about 36 utilities across 32 states or so from the largest IOUs to the smallest co-ops. And what we are able to give them is that situational awareness from an operational perspective to look out 5 to 7 days in advance and see these extreme events coming and see the events that could result in catastrophic fires. So not just ignitions, but consequential ignitions. And that's really the nut that we crack for the industry is to help them look at this in terms of liability and risk, so probability times consequence. We do it at the asset level. And then when you have an understanding and they can make decisions off of that. They can decide to change settings on their network so that things trip faster and they don't create sparks. -- or worst-case scenario, they can do what the industry calls public safety power shutoff, and that's shutting off power and those types of events. But our job is to help them not do that unless it's absolutely necessary. And if it is necessary to limit the scope and the scale and the duration of that because it's dangerous to shut power off to people. It's got such far-reaching implications. And the challenge, of course, though, is that they -- without other tooling, they're going to err on the side of caution. And if they don't have a detailed understanding about fuel situations, about what the weather conditions are, about what weather trends are, of course, out of an abundance of caution, they're going to maybe head towards shutoff as an option, which, again, as you're pointing out, has a whole set of other consequences that you don't necessarily want to pick up. That's exactly right. And what we do is we give them the ability to look -- so we do a historical reanalysis going back 15 or 20 years, and we break that weather data down very granularly. We tie it to their assets. And we're able to draw a correlation between outages like vegetation contact or poles breaking or whatever it may be and weather. So then that's what gives us the context to be able to predict things going forward. So you can see -- you're not only saying, hey, do we have a 50-mile an hour wind event coming up in this circuit, but what does that mean in the past? Maybe 50 miles an hour on this circuit doesn't mean anything because they constantly see that type of wind. But on another circuit, it could be something serious. So we're giving them that historical context. We're putting it into percentages, and we're doing it in a very granular way. 
 

Eric Hanselman: So you get them to see what's anomalous, what could potentially be a challenge versus what's been going on day in and day out. 
 

Bryan Spear: That's exactly right. And then when you have that level of asset level intelligence, then you can make longer-term planning decisions. So you understand which of your assets are at risk of starting these consequential ignitions and then you can make informed hardening decisions. It could be like we're going to target vegetation management in certain areas and then vegetation management goes from not being just time-based to then being like condition-based to actually being risk-based. And that's a significant cost that it's typically the #1 O&M cost that utilities have. So we help them be smarter about that. But also when you're making hardening decisions, you need to be able to understand what the risk spend efficiency is. So how much risk you're buying down based off of your various mitigation options. And as affordability is becoming a theme across all utilities today, as you're probably seeing that, this is becoming a requirement, right? You have to be able to articulate the investments that you're making in hardening and what that's doing to buy down your risks. That's exactly right and managing longer-term capital expenses and ensuring that you're making those investments in the right ways where you're getting that payback. And then it becomes like the next phase that we're doing now is focusing on not just risks associated with wildfire, but with all weather hazards. And there are a couple of different ways to look at this. But one, if you go on the same thread that you just brought up, the way that utilities operate today, they're in silos basically. Their wildfire team is separate from their team that manages emergency response and storms and planning. So you're basically -- there's no coordination in many cases of these investments. So you could be solving problems in isolation, and you could be double spending or triple spending, right? So now you've got the ability to look at all weather-related risks together. So it's going to be much more optimal. But more importantly, what this has done is it's given utilities the ability to predict outages, right? So not just wildfire conditions, but outages associated with wind, with ice and snow, extreme heat, extreme cold or flooding and predict it in advance. And what that means is that you can now preposition crews, you can preposition equipment in the areas where they're going to be needed and dramatically reduce the restoration time. And frankly, that saves lives, and that's in alignment with our mission. 
 

Eric Hanselman: Well, in the Northeast, restoration is certainly a concern for us. We've been having a particularly snowy winter, but it's so much of that, that really is understanding where do you get crews in place? Where are you likely to have the most significant impact. And that prepositioning can help you get back -- get things back online with that much more rapidly. But yet, there are costs associated with how many crews do you dispatch, how many do you get out and that's got just as much of an impact as well. 
 

Bryan Spear: That's exactly right. I was just actually right before this podcast, I was on the phone with the President of a pretty small utility, and he didn't know that we had expanded into extreme weather. And he said -- and he said, "Bryan, what would be really great is if you guys -- because we just had bad storms over the past 4 days. We had a lot of trees down and he said, if you were able to predict the outages and we were having storms come through, I said, interesting. Tell me more about that. And they said, when we've got a snow that's coming in and it's above 28 degrees, then I know it's going to be wet snow, it's going to be heavy. And I'm going to have trees down; I'm going to have lines down. And I just know that from experience. But if you could use your modeling to say, "Hey, what happened in the past when we've had these types of conditions, this type of snow, these temperatures, et cetera, and then be able to predict going forward how many of these outages and where and when, that's extremely valuable. And he said, and then knowing it in advance, I can know when I need to call crews in to come in before the roads get shut down, et cetera. I said, well, it's funny that we're talking about this because we're actually doing this today. So that's our -- that's the value -- the problems that we're solving on the utility side, but we've got 2 other primary customer bases that are just as important to us. And one is fire agencies. This is where we got our start. when our founder, Joaquin Ramera started the company 30 years ago, it was to support fire agencies and give the fire agencies the tools that they needed to protect communities and to keep firefighters safe as well. It's about modeling active fires, where is that fire going to go at what rate, what intensity, assign resources, like where do you do air drops, where to put the dozers and tracking all those resources, et cetera. That is still today a core part of our mission. We work with some of the largest fire agencies across the world with those tools. And then lastly is the insurance market. So our platform combines the scientific catastrophe models with financial simulations. And that evaluates the catastrophic risk of loss, both at the location level, and that's used for underwriting and pricing purposes. So you're getting accurate pricing based off of real risk. And then also portfolio level evaluation, which supports capital planning and reinsurance needs, and we support over 100 insurance, reinsurance, and financial services companies with those tools. 
 

Eric Hanselman: Well, are there things that you're looking at beyond fire and flick because you also look at flood. I'm curious about when you start integrating a set of these different hazards. You touched on this a little bit, but I wanted to dig into this idea that when you've got these perspectives across the whole set of different event criteria that could be heading out there, now you start to get into some interesting space in terms of the kind of things that you can identify and where it can be put to work. 
 

Bryan Spear: Yes. It's really interesting, Eric, and it's really exciting. Let me tell you how this came about. It's an interesting story. I was at lunch with Jason Wells, who is the CEO of CenterPoint Energy. He was previously the CFO of PG&E when they went through bankruptcy. So he knows our tools really well. We're in the process of deploying our wildfire solutions. This was probably, I don't know, roughly 1.5 years ago. And he's very familiar with the fundamental nature of how our solutions work. And he had just gone through Hurricane Barrel. If you remember that, it hit the Houston area really, really hard. I took a lot of unfair criticism from that. And he challenged me, and he said, “Why can't you do the same types of predictions for other outages related to other perils that you do for wildfire ignitions. And my initial response, I'll never forget it, I said, Jason, we're a wildfire company. We can't do that. I went back and started talking with our team and realized that, in fact, some of our customers were already using our data in that way. They were taking it, and they were taking the non-fire weather elements, and they were downloading, and they were using it to predict these things on their own. And the team said, we can absolutely do this. We're in an incredible better position than anybody to do this. So then I said, well, surely, these tools exist. Surely, people are doing this. And what we found continuing to talk to utilities, you had a lot of utilities that have developed them on their own, but you didn't have like commercialized tools using modern machine learning and the latest and greatest technologies. They didn't exist. So we partnered with CenterPoint Energy, and we codeveloped it together. It's operational today. And it was really put to the test during the winter storm fern that came through a month or 2 ago. And -- and what they saw is they were able to communicate to the customers ahead of time, which they weren't able to do when Barrel came through. They're able to position their crews. And they said to them, it was a game changer to be able to see it coming ahead of time and be prepared for. Another CEO that I've talked to you about this says it's the holy grail, being able to do all these things in one platform. So this is something that it's incredibly valuable. And you've got a lot of places across the world where wildfire is not a #1 concern. But one of these other perils is. And if we can bring them a platform that's going to address all the perils and spread the cost across all of them, allow them to plan from one specific set of data, then it's going to be incredibly valuable and hopefully game-changing for many others as well. And when I look at like -- I think about Eric, like the evolution of Technosylva, it was -- in a way, it's kind of 3 chapters. It was the first 15 years were developing the product and serving the agencies. And again, the is always going to be core to our mission, but that was the first 15, almost 20 years of the company. And then about a decade ago, San Diego Gas & Electric, after they reached a settlement for the which fire that they had been found responsible for $2.4 billion settlement, they reached out to Joaquin and team and said, "Hey, can you take those fire models that you're using? And can you work with us to modify those around the use cases of a utility to help us avoid creating those ignitions. And then we did that. We kind of created this market that has really, really grown over the last 10 years. But that was the second chapter, I would say, of Technosylva. And now this expansion into these extreme multi-hazard or other perils associated with extreme weather across utilities and insurance is really the next chapter. It's the future of Technosylva. I think that this is going to be even bigger than wildfire, and it's directly supportive of our mission. 
 

Eric Hanselman: But it seems like there's probably a reasonable overlap in some of the data sets you're working on. Clearly, weather modeling and the weather historical data you've got. Absolutely, ought to be easily extensible. But you probably have also got GIS data, terrain data, satellite imagery, I'm guessing you're integrating to take a look at fuel load and then having to make that pivot to shift into looking at structure placement, structure elevation, a whole set of different things. It seems like there -- it might be easy to look at that instead of say, "Oh, sure, no problem, simple pivot. I suspect it's a little more complicated than that. And I guess to dig into some of the nuts and bolts on the technology and infrastructure side, how does -- is this something that we're living in the age of now frontier models and great multibillion parameter capabilities that are out there. Are they changing this as an evolution beyond traditional machine learning? What are the impacts? And how are you actually leveraging those capabilities? 
 

Bryan Spear: Yes. We've been using -- and actually, the data that's used for the multi-peril or the all-hazards modeling is actually extremely similar to what we do for wildfire. We just weren't using a lot of the data that we were generating and collecting, right, which surprised me. It probably didn't surprise the people in the company that are a lot smarter than I am. But it was -- there was a ton of overlap. So we didn't really have to go generate a lot more new data. The flood solution was a bit different, but we acquired the company that has the pioneered flood modeling. So that wasn't necessarily the data that Technosylva had been generating, but it was from Cat risk that we acquired and then we're able to pull that together into the portfolio. But -- so we've been using different types of machine learning models pretty effectively for many years. We have -- you can imagine hundreds of inputs that go into our models, right? And some of those are publicly available information. Some are publicly available information that we have to modify to make it better and some are just completely proprietary data points that we generate. I'll give you one example as we -- we have a partnership with NASA, where we get satellite data on the moisture levels of fuels and soil. And you can imagine that when you're looking at ignition risk and then spread risk and what the consequences could be, that really matters. And that needs to be dynamic. So we put machine learning models on top of that satellite data to make it dynamic as part of our real-time modeling. That's just one example of what we've been doing for years. But the generative model era changes our road map in really 2 big ways. Number one, it's speed at scale and its better decision support. And first, on the modeling side, the physics-based models are -- they're still the gold standard for Fidelity, but they're expensive and they're slow, right? So generative model approaches, they can generate really high-quality forecast in minutes instead of hours. And for us, that means that we can update risk more frequently. We can run more configurations and especially when conditions are fast -- are changing fast. And we can do it more cost effectively as well, which allows us to reach more and more customers. And second, the real differentiator is the data and the workflow. So a lot of the AI weather models that you're hearing about quite frequently in the news, they were trained on relatively course historical data. So typically, 30-kilometer resolution data. And for many applications, that's good enough, right? But we've built and we've curated these historical data sets at a much higher resolution. In fact, about 15x greater resolution. And that's what we use to train our models, and that gives us the ability to push towards longer horizons and probabilistic forecast without losing that asset level detail that our customers need when they're making these $1 billion operational decisions. So I think over the next 12 to 24 months, the biggest impact of leveraging these generative models is going to be that extending that actionable lead time, so moving towards like a 10- to 15-day risk window in a way that's still operationally meaningful. So we're not losing that fidelity. We'll be able to run much richer ensemble, so more scenarios more frequently with clear confidence signals. And then also, which is really exciting, just AI interpretation at the edge. So imagine agents that can translate millions of simulations and be able to answer what changed, why does it matter? How confident are we in that? And then what's next to do for each specific operator? So we're not replacing physics with AI. We're building a hybrid system where AI is accelerating the forecast engine, and it dramatically will improve how risk intelligence gets operationalized. And then that shift from determinalistic modeling to AI augmented intelligence systems is really what defines the next stage of our evolution. And it's quite exciting what it's going to open up and going to allow us to do. 
 

Eric Hanselman: Well, it's interesting to see some of the evolutions towards physics modeling within AI. And there were a lot of doubters early on. But to your point, this is something that it's being proven out. It's being actually -- it's generating real results. I'm curious, though, you're talking about much larger data sets when you go to higher resolution, bigger models, the computational work and just simply the infrastructure management pieces of this start to become an issue. How do you scale? And what are your biggest concerns in terms of how you manage that growth? Clearly, you want to be able to have more timely, more frequent assessments. You're talking about denser ensembles, more perspectives. Hopefully, you wait into better answers. But how do you tackle that? 
 

Bryan Spear: Well, I'll tell you, one of the ways that we're tackling it is last year, we invested in what we believe is the world's largest supercomputer dedicated to wildfire and extreme weather modeling. And this is what allows us to -- we have massive cost-efficient capacity so that we can run these hyperlocal forecasts at a very high frequency and deliver the risk guidance that is timely enough to act on, not just analyze it after the fact. So we -- today we're running over 1 billion simulations every single day. and that's going to continue to grow exponentially. So that is -- that's an investment that we made because we saw the opportunities that were coming with this. And what we used to do is we would model -- when we signed up a customer and we were deploying; we modeled their territory. And it would take months because it's very detailed analysis. And as we continue to grow, and we were moving east with utilities quite quickly. And like I said, we're across 36 states now. So we said, "Hey, why don't we make the investment? We can do all this modeling upfront. So we're doing the historical analysis for the entire U.S. We've already done it. So 15 years back, and then we can run the operational models in real time going forward, and we're doing it for the whole country. So what that does, it allows us to bring like we can bring new customers on faster. So it's less expensive. right? And it also opens up -- now we've got this data that's available for the entire country. So it opens us up to offer that data, use that data for valuable applications that are outside of our current core use cases. 
 

Eric Hanselman: And so it's meant that now that you've got nationwide data, picking up a new client winds up being that much simpler because you're not running client-specific approaches and being able to handle this. 
 

Bryan Spear: That's right. And it wasn't as big of a deal for the large utilities because they've got a lot of risk and they've got big budgets. But when you get into the smaller utilities, some of these guys are 10,000 customers. And -- but look, our mission is to reduce the impact of wildfires and extreme weather. We want to prevent any utility from ever starting a catastrophic ignition. That includes these small co-ops and munis that might have 10,000 customers. So how do we get to them in a way that's cost effective. And this was one of the primary reasons that we made this investment in this approach. But what we also saw was that it gave us the opportunity then to leverage this data for many other applications into the future. And we're just starting to kind of look at what some of those could be. If you think about, Eric, like there are some that are more obvious, like near term -- near adjacencies like energy planning and demand forecasting, for example, like having high-quality, high-resolution data that's tied to your assets, you can obviously be better at predicting demand than what's out there today. Rail is a good one. Most people don't realize that locomotives are responsible for many destructive fires. In fact, the most. You need is one hot box, right? The most intense fire ever recorded on earth was the Chisholm fire in Alberta, and it was ignited from a spark from a passing train. -- that fire released so much energy, released as much energy as multiple nuclear bombs. And they don't happen, obviously, not that intense, very often, but they happen quite frequently. So if we can give rail companies the ability to use that information and make changes on these hot and dry windy days and avoid another situation like that. If you think about like the insurance and reinsurance markets that we serve today, but that more granular forward-looking hazard layers and event narratives will better support pricing, exposure management, claims readiness. And if you put operational modeling on top of that, it just takes that to a whole another level. And then if you go a little bit further away from what we do today, but the data is still particularly relevant for, think about like site selection, corporate asset risk of where to locate or where to harden facilities, you think about warehouses, hotels, commercial developments. Data centers is one of our favorites, but exactly the same sort of things. There are hundreds or maybe even thousands of those that are currently in the planning process. And they have very limited data of which to make those billion-dollar decisions today. So there are just many, many different uses. We're talking to some large banks now about how they can utilize this data. As investors, they want insights into the full risk portfolio. That's exactly right. And look, we've done it. We have it available already. So we're just -- it's been -- you know how it is, the company our size as you're growing, if you try to do too many things and you don't do anything really well. So we've tried to be really disciplined about where we focus. So we said no to a lot of things, but we're to a point now where I think what we've generated is going to be directly applicable pretty much as is. right, versus having to do something that is customizing that data or that solution. I think it's going to be leveraging that data to help solve some pretty big problems with minimal effort on our end. 
 

Eric Hanselman: Wow, impressive. I'm curious, I want to get your take. We spend a lot of time talking to our clients about what has been digital transformation and really integrating data into critical decision-making within their organizations. And I'm curious about your experiences. I mean, what level of advice would you identify for enterprises in terms of how they should expect to use data in their operations? 
 

Bryan Spear: Well, that's -- that could be a pretty broad question. I think for us, it's -- what we try to focus on above everything is all this data that we generate, so what, right? How do we give it to our customers in a way so that they can make decisions. And these are meaningful consequential decisions -- so we need to give that data to them in a way that is -- we're taking hundreds and hundreds of inputs. We're running literally over 1 billion simulations every day. So you have all this data, all these scenarios. And if we don't put that in a way that gives them information on which they can make a simple decision. And then as we go into areas where these are newer problems for some of the -- or I guess, newly recognized problems for some of these customers, and they don't have a lot of expertise and experience with these types of tools. So we have to make it really simple for them. And that, I think, has been the challenge for us over the past couple of years is how do we take all this very, very complex data and turn it into simple decisions. And that's what I think is the key. If you don't understand it, you can't make decisions off of it, then it's not worth much. 
 

Eric Hanselman: Well, integrating it into a decision-making process that for many organizations, when you were calling out some of the challenges in reaching across silos for some of these decisions. That's a complicated process that organizations have to be able to manage and to inform that better with data. In many cases, can be a heavy lift for them. But again, something, especially when we start looking at some of the areas that you're identifying, the risks are something that really merit and hopefully incentivize people to make those decisions better informed by data. 
 

Bryan Spear: That's exactly right. And you see the implications of some of these risks. I mean, you -- credit ratings has been one that you -- there have been -- over the last 5 years, I think there have been over 100 credit rating downgrades on utilities for wildfire risk, right? And if you think about that, if you take a typical medium-sized IOU that has investor-owned utility and say they have $25 billion of debt and a one notch downgrade, that could be $60 million -- $50 million to $60 million a year. So that -- and that's happening, right? And it's happening if they're not addressing the risk. And then if you do have an ignition or on the insurance side, it can be a whole another set of challenges that you give an ignition, then it's just -- it's almost -- you just almost cannot calculate all the damages that you incur from that because they're not all monetary. 
 

Eric Hanselman: Well, hopefully, it's one of the situations where an asset prevention is worth a pound of readings improvement. So we'll see what happens. Well, this has been great, Bryan. I appreciate all of the insights into what is a very complex environment. 
 

Bryan Spear: Well, Eric, I really appreciate you having me on and just bringing attention to these types of problems that are happening in the industry and also to the solutions that are available to helping solve these problems. So I've enjoyed the conversation. Really appreciate you having me on. 
 

Eric Hanselman: Nothing like sprinkling a little AI on all of the world's problems, but we'll see where we go from here. Well, there are so many more things we could discuss, but we are at time. That is it for this episode of Next in Tech. Thanks to our audience for staying with us, and thanks to our production team, including Sophie Carr, Feranmi Adeoshun, and Kyra Smith on the marketing and events teams. Please keep in mind that the statements made by persons who are not S&P Global employees represent their own views and are not necessarily the views of S&P Global. I hope you'll join us for our next episode because there is always something, Next in Tech!