EP16 – Behind The Shroud – Visualizing and Quantifying Property Exposure Within Buildings Using AI. Cole Winans, CEO of Flyreel

As insurance professionals we do a fairly good job of assessing the value of a building. Why? We can see it. The walls, the size and now we can use third party data and visualizations to scale all of that.

Where we fall flat is what is happening within those walls. Satellite imagery is not going to help. Third-party vendors can’t help us with that.

In this episode, I spoke with Cole Winans, CEO of Flyreel. Flyreel is using AI to analyze imagery and video that any property owner can quickly take and categorizing and ultimately valuing the interior contents of a property owner. We also discussed how this technology can extend to providing property owners with information that can reduce or eliminate losses such as from shoddy electrical fuses or old faulty water heaters. AI has come a long way since hotdog/not hotdog (adult language: https://www.youtube.com/watch?v=ACmydtFDTGs)

Watch here:

Connect:
Cole Winans
Flyreel Homepage

Musical Credits:
Shadows by David Cutter Music:
https://davidcuttermusic.com
https://soundcloud.com/dcuttermusic
Free Download / Stream:
https://bit.ly/shadows-david-cutter
Music promoted by Audio Library:
https://youtu.be/qiBHOiEl9EI

Video Credits: Intro Stock Footage by Videvo

Transcript

Nick
I start this off like um, so I am recording I start this off really easily.

And, you know, why don't you introduce yourself, talk about your company and what it is Does.

Cole
Sure Sounds good? Well, my name is Cole Lyons. I'm the CEO and founder of Flyreel. And what we do is we actually sell an AI assistant to property and casualty insurance carriers that guides their policyholders through their own self inspections and claims filings.

Nick
Yep. So every I hear AI lot. Okay, so let's break this down in two pieces for those in the audience that may not understand what AI is because it's funny. I had a conversation yesterday with someone where we talked about AI and machine learning and I got it wrong. So I don't even I didn't quite understand AI. So before we dig in a little deeper on how your AI might be different than anyone else's. Describe for the audience, how youth how your company defines or basically comes up with the concept of AI.

Cole
Yeah, so a couple of things. There's always semantic debate. And I think technical and engineering focused people really want people to respect the difference between deep learning and machine learning AI versus ML. And it just doesn't matter in a business context. So I'll come at it from from that angle at the simplest level. it's pattern matching at scale, and using methods to determine probability and statistical probability, to try to predict things as accurately as you can. And then in terms of, you know, what, why does this matter? How are we doing it? We specifically use computer vision technology. And we've trained our platform, our models to look at images of homes and videos of homes from the lens of an inspector and the risk engineer so that it can automatically extract and identify the data that matters to carriers. So, if you play this out now, you know you can have thousands of properties scanned with video walkthroughs of all them. And you don't have to look at every single frame, watch every video to extract the data that you would want to pull out of those to know this person has a wood stove in their home, or they have a pool but it doesn't have a fence around it. That's a task best suited for technology today. It's a simple task. It's repeatable tasks. You can train computer vision models to do that for you automatically. So outsource it.

Nick
So training is the key word right. So that's a that's the one of the more complex parts of what it is that you have to do is that In order to get the computer to, try to train it into get the computer to actually learn, you have to feed it information. So can you talk about what it took for you to be able to actually offer a product that actually does this? Give a sense of you know, how much how much data and information you had to feed the computer to get the get the, the wheels turning?

Cole
Yeah, so I think there's to get the wheels turning and then to get particularly good at our focus and the distance between each is pretty considerable. So this is a story I like to share because there are other startups often talking to carriers in pursuit of really neat AI solutions. And there's sometimes pushed off because they don't have the solution yet. You don't get to start there. You have to kind of inch your way into it. And we did get a great partnership with Microsoft out of the gate. We got very fortunate. And in that partnership with Microsoft, we were able to use some data with their help to train on a base knowledge of appliances. So for example, you could pan across the kitchen. And it wasn't super radical, earth shattering technology, but we could automatically identify that you had the refrigerator and we'd label the oven and Is it is it a gas stove? Or is it electrical? And that for us was our moment to prove, okay, while this may not be the most valuable data that you need as an insurance carrier, because you can, for the most part safely assume someone has a refrigerator in their kitchen, we're demonstrating, we've got the core components to do something even greater than this, we can identify things. Now. Let's try to start capturing more and more data...real world data of these homes so we can do more and more nuanced things. So getting from zero to maybe a half to where we could identify things was, you know, just using generic data and images of appliances to demonstrate to our prospective customers. Hey, we've got something here. We're not just making this all up.

Nick
Yeah. Your story reminds me of the HBO show Silicon Valley...familiar?

Cole
Uh huh. Yeah.

Nick
When hotdog and everyone like did identify what a hot dog was, yeah, one got excited and then everything else is not a hot dog.

Cole
Exactly. No, it's you have to start somewhere you really do.

Nick
So so this is this is interesting because your business is focused predominantly on the claim side. Which makes sense because that's where the kind of the rubber hits the road. But I would say, you know, I'm being a property underwriter, I would say since the beginning of time, since the beginning of property underwriting, one of the most understated or under respected aspects of property underwriting is getting the contents value correct. And as a property underwriter I remember I was moving across the country we had three moving companies come in and all three of them were were within a few thousand dollars of one another of the property value after doing the full inspection. But the most amazing part of that was their number was about seven times higher than what I was insuring my contents for.

Cole
Wow.

Nick
So I was I was way off and what what it is, I think kind of understanding like that, not only the difference between what refrige being able to identify a refrigerator and a, you know woodstove is valuable, but also to the extent of well, what make or model is and then what's the replacement value? For something like that is is very much an under underappreciated aspect of property insurance. Have you? It seems like your technology could also potentially solve that problem as well.

Cole
Yeah. So a couple things when we first went to market we thought, you know, claims was was likely where we'd start and we quickly found that we've had most of our focus almost all of our focus right now is in underwriting. And now we're kicking off our claims effort. And because there are glaring issues like that, where the carriers and in the insurers as a whole just have not had the luxury of seeing inside the home., to better understand the risk and to size coverage appropriately and contents is a huge pain point. It's not an area where we're operating right now. However, it's an area where we can see ourselves evolving and a lot of our customers have expressed the pain there and an interest in using our technology for that and we've begun capturing some of the bigger, more high priced contents so we do appliances already but now we're getting into electronics and some sporting equipment just to cover some of that and, but I mean content is one thing but it's also the presence of that I mean, an indoor pool if you're doing an exterior inspection how would you know and so now this we're just seeing eyes really open to gosh, you know, we've done exterior our entire life as a company. And we've never known these things are here.

Nick
The interior of the home has always been the it's been like a shroud around it or walls.

Cole
Yeah, okay.

Nick
Yeah, there was so how how do you you engage? How do you engage? So, you know, most of the business that comes through the insurance world comes through brokers. Right? So how are you engaging? Or what's the what's your business model so that you can get access to the inside of the home?

Cole
Yeah, so usually it's the from a business model perspective. You know, the carrier is purchasing this just like they would pay for an inspector to go out there and look at the home. From a distribution perspective, you have situations where some carriers go direct, and others through agency channels. And what we've found is that capturing this data, and even letting the agent be a part of that is an opportunity for the agent to better serve their customer. And we've actually seen agents get pretty creative with that. So for example, with Flyreel, and one of our integration partners Donan, you can scan an appliance and get back any recall data. It's estimated life expectancy. So as an agent, if you have a book and you see that some of your customers have outdated or expiring appliances that are high risk, can you potentially work locally with some of the small businesses in the area to find discounts for them, and to just serve them at an even greater capacity? So we've learned really how to deliver value not just to the carrier, but how do we better protect the policyholder as well. And then how do we equip the agents with data to better serve their customers to ultimately, you know, that's, that's their core focus. So enabling that is a focus of ours as well.

Nick
I'm assuming is most of this or all of this is done via mobile phone.

Cole
It is Yeah. Usually at you know, new policy or renewal. So a good chunk of our business comes from renewal inspections as well as new policies. And the flow is that typically the policyholder, after that quote, kind of in that underwriting window, will get a text and an email from the carrier, they tap the link, it takes anywhere from 15 to 25 minutes to scan the full home, and you're good to go. Everything's documented and covered.

Nick
Yeah, that was my next question was the simplicity of use? Is it simple enough for a property owner to do it on their own? That does the software almost give them like a guided tour like directions on angles of the phone? Or, you know, or does it? Is it all photography or is it video as well?

Cole
It's both and I think the user experience I would say is one of the biggest differentiators between Flyreel and potentially alternative solutions. Using AI, we're now able to offload the heavy lifting from the user onto the computer. And so they can literally walk around the house with the video camera on and it's just, hey, Nick pan across the kitchen from left to right. Wow, you did a great job, let's go get the other room. Because the AI is doing that heavy lifting, creating that list documenting those materials and everything for you. And so it's a very fluid, very efficient experience. And we've done that by really harnessing the hardware on the smartphone. Just maximizing the capability of the sensors and the video everything there.

Nick
I can't underestimate this enough because it's we're moving from a world where we basically took again, we took the word of the insured on the value of the contents. So take my example, right where I'm a property underwriter. I should know better. I was several magnitudes lower on my evaluation of my own property. So we're taking my word on it and I don't have enough expertise to know this to a world where in sounds like minutes, literally minutes to walk around taking video and having a computer get a much higher degree of accuracy. So the the valuation is off the charts in terms of not only getting the value correct, but also at claims time where someone like me, I would have been hit with all sorts of coinsurance penalties, because I wasn't anywhere close to what the real valuation was. I was basically slapping myself on the wrist.

Cole
Right? Right. Yeah, no, that's exactly right. And, and we're, there's a big shift happening. And and it's going to play out I think over the next few years where technology is affording the market new opportunities to structure products and serve customers in a better way than it's been done over the last hundred or so years. And I think the way that plays out is through much greater accuracy. Far fewer guests and far fewer guesses and assumptions of risk. I mean, it's we kind of joke a little bit that you get these big data analytics firms, and they say they'll put together an AI model. And it could be say, to predict who's in the house next door, you could just go walk over there, knock on the door and ask them their name. And I think insurance has been in that mode for so long. It's why are we crunching 1500 data points, looking at socioeconomics data to try to predict how much these people purchase and having their homes when you could just ask them? Yeah, and they do it and they scan it because they also they have a mutual interest in protecting their things that they care so much. About and, and I think that's what you pointed out is a very important piece is customers want the appropriate coverage, they want to know their stuff is protected. And we haven't been able to do that in a hyper personalized way at scale. Until now. Now we've got the luxury of this technology and we can do those types of things.

Nick
So give us the backstory, how to Flyreel where, what was the genesis of all of this?

Cole
Well, I I wish I could say that. This was our idea on on day one, and we've been working on it ever since.

Nick
What was your idea on day one? Now Now, since you brought this up, I gotta see how far you've pivoted.

Cole
Geez, it's not good. It's not great. No, it's it's it's fair game. It's a good question. So I started the company with some good friends. Technically, we formed the company in 2013, and my background is a software engineer and more specifically a product guy, I really enjoy taking things from design all the way through to development, I'd had the opportunity to do that for a couple companies and got to witness and experience so their success and got the itch to do it myself. So I formed a company and didn't quite know what we wanted to do. And at that time, everyone was building a social network. So so where we were going to build this social network, but using AI, we were going to and we built a solution where you could ask a question, AI would listen to your question, understand the skills necessary to solve that and connect you to someone with the answer live video. So then, that was the initial concept and we had to apply some focus and we applied focus by using this in the context of home repair. So if something broke in your home where you needed something fixed, you could just pick up your phone and say, Hey, you know, my water bills gone up, I'm not sure why or I see this leak, I'm not sure how to trace it. And you'd be face to face with a plumber in your town in about 30 seconds. And it worked. It was very, very cool. But the side, you know, the takeaway was we built a better mousetrap, but we could not distribute it. We couldn't compete with Angie's List and Homeadvisor these great companies with lots of money to spend on ads. And so a better product wasn't a better business. And so we kind of took that and said, What if we expanded the AI and if we didn't just understand what you're saying in your home, but what if we could understand what you're showing in your home? And if you pan across the room, can we document the details? Could we look at damage could we create a list. So we created a little proof of concept and had a couple early mentors and investors that came from Eagleview technologies. In the insurance space, and said, you have no idea what you have for insurance. And I said, Well, I'm not gonna I don't want to repeat, we didn't want to repeat what we've done before creating an awesome product with no demand. So we went out to a conference with this kind of basic prototype, and said, Let's show it some carriers see what they think. And we left there with a lot of interest and a lot of excitement. So that's our full Genesis story. Definitely a pivot, but you can kind of see incrementally how we found our footing.

Nick
Yeah, completely. And the future I'm assuming is a continuation on what you described, in terms of ratcheting up the safety potentially the safety value of, you know, hey, your your, your water heater, for instance, is 15 years old. And you know, when it goes it's going to cause a lot of damage or things with electrical...there's probably a whole, both interior and exterior is probably a whole bunch of things that then play into the risk management of of potential loss like that the curve is now switching. And if you don't take action soon, you're you know, you're going to have a loss. And that's not fun. I'm, I don't want to speak for you. But I'm assuming that seems like a natural stepping stone.

Cole
It is. And we're now you know, now that we're growing as a business, and more and more data is coming into the platform, we're finding we're able to roll out new capabilities that deliver more value to our customers in the realm of risk management. And so some examples of that are, you know, now we'll have the homeowner scan their electrical panel, and we can automatically detect the stab block breaker that was recalled in the 80s. And it causes 40 or so million in losses annually and 13 or so deaths. I mean, this is stuff that now we can prevent these things from happening by flagging them early on in the process and catching those hazards and risks. And more importantly, I think you have so much knowledge bottled up into these carriers and their risk management teams, but their hands have been tied behind their backs because they haven't had the data inside the property and really even at the ground level. And so when you hand them that data, you see their efficacy skyrocket, and now you can actually amplify their knowledge and experience through the use of AI.

Nick
Yeah. What about the commercial realm of that same same problem, you know, because we we write residential and commercial I see the exact same thing where you know, these hundred thousand square foot warehouses come in, and the contents value is like 5% of the total building value. I can gauge the total building value but that contents number does not seem right at all. And it happens all the time. My guess is, it's probably might even be worse. Because I think on the residential side, you know, there's some tools that the carriers can use to kind of fix that the contents, they don't have the same tools and commercial. What can you do there?

Cole
Yeah, well, we've now gotten into commercial and we're taking a step by step approach, making sure that wherever we enter we are delivering hardened value with tangible results. And so we've started in restaurants, and now we're getting into LRO and also multifamily housing units as well. And so we'll catch things you know, restaurants the kitchen is really the hotspot mean literally, and so we'll detect sprinkler heads, presence, absence of fire extinguishers, look at the fryer, all these things. You've got slipping fall. hazards in the main eating areas when you have the soda machine, is there a mat there? Is there a weather map? Do you have exit signs? We can automatically detect those types of things. So, you know, we've focused in those areas to start and we're seeing some good impact out of the gate.

Nick
Yeah. Do you anticipate how quickly is the AI technology advancing? I've seen like a big spike in I think AI productivity, AI efficiency of what it can do. Is there any cause to believe that it will not it won't continue at this accelerated pace?

Cole
No, I don't think so. I think that the amount of research that's happening in AI is just outstanding and the work and the intelligence of the people pouring into it as an overarching practice is just I mean, it's hard to sleep sometimes because it's just so fascinating and people are so passionate about it. AI is is really going to be the epicenter of software going forward. It's just you will use probabilistic modeling and this approach, in most things going forward, it's either you'll look at, does it make more sense to develop a model, or kind of a simple binary rules solution? And that's kind of where things will go. So it's, it's just going to get further and further entrenched in each business. I see.

Nick
Yeah, it sounds like similarly to what I do with Nat Cat. I feel like we as practitioners are going to struggle to keep up with the scientific technology...sounds the same, where the theoretical aspects of AI will advance so much faster than we can potentially consume it. And so I feel as though on my side like I'm I'm always asking scientists and engineers that are building things, to kind of dumb it down for me, so I can make some decisions off of it because it's too complex...sounds like it's going to be the same in AI, potentially?

Cole
I'd say it's the same already, there is a delta, I think between the academic side of AI, which is way out here, and then the operational side of AI, and both need each other. I think sometimes both will roll their eyes at each other. Academia is out here and they figured out a lot of problems but and they look back at the current state of market and they say, Oh, you're so far behind. And then the operational side of AI rolls their eyes and says, you guys don't actually have to put this into market. So I think it's, you have to really tread that line of academia and actually operationalizing this stuff so that you're on the edge and you're continuingly pushing the boundaries and capabilities.

Nick
Yeah. We're living in amazing time we've gone from hot dog/not a hot dog, to do all of this stuff with, you know, being able to tell the kinds of fuses that are being used. I, you know, it probably it's probably not too far off. And you've probably already thought about, you know, there. I've seen cameras with infrared that can look through walls and already detect rot and things like that. And so, the ability for us within minutes to be able to detect everything that's going on in a home. We're probably not too far off. Could you talk about folks that are listening to this, that have fear of AI, you know, they're going to take our jobs is going to be intrusive. We're going to lose privacy. He talked about this those sorts of issues as well. Sure.

Cole
Well, first, I'll start with more of a light hearted entry into this, which is I as someone that's entrenched in this every single day, I personally don't think there's too much need for concern about this wiping away jobs. And I'll give you an example. A personal example with with Flyreel, I think early on, we would see situations where you'd pan across the backyard, and there's a patio table there. And it's got this glass surface reflecting the blue skies. So if you have the circular table with a blue surface, that's a hot tub. Right? And then you'll get a computer to monitor and tell the difference between a computer monitor and a television screen is a tremendous amount of work. Yeah. And so when you're working on these basic problems, You hear debates about robots taking over the world. I mean, you kind of roll your eyes, I think, you know, more realistically, what we're seeing is AI as an amplifier. And so we come back to that example of the risk engineers are now being equipped with more data than they've ever had before, and more visibility across their book of business than they've ever had. So now what happens, they're actually considerably more valuable, because they can actually parse the data with the help of this AI assistant on 10,000 homes rather than 10. And so I see it more in the near term acting as a knowledge amplifier than a replacement for for people.

Nick
Yeah, I generally tend to ask that question for anyone that is associated with AI I partially makes me feel better, you know, but you know, but I think also partially, it's that it's a tool, right? Because, yeah, as you're, as you're talking all, you know, as in, I can't help but think as the world evolves, that we'll have new stuff and that is just going to be a need to constantly train the software and then this and then we always go from like one software platform to another and is going to have someone's gonna have to port that over that. Yeah, I think it is in the realm of technological space, like you need technological sophistication to be able to sort of keep up but I don't know, to me that's the exciting part is like from from a human standpoint, as long as we if we're, if we're eager to keep learning, we can keep up with it. And then it's a tool for us to use no different than a hammer or a saw.

Cole
No, I I totally agree. And I think the more we work on on this technology from from a personal perspective, the more impressed I am with with the design and our capabilities as people. Because you start to learn about the things that we do just naturally in terms of understanding context, which is, and so when that's, that's one thing that we've had to kind of bake into, into our solution is, if you see a rectangular object that is silver in a bathroom, the probability of that being a refrigerator is almost non existent, duh. We don't even have to think about that. But for an AI or computer vision model, to to know that you have to have you have to introduce context and weight things differently based on potentially where you are These are things we just take for granted, it happens naturally we don't think about. And as you try to engineer this, you start to have a greater appreciation of the things that we just come fully baked with.

Nick
Yeah, you know, like that that slot that slight, very slight I roll that my wife gives me when I tell a joke, and she doesn't like it that a computer would be like, what does that mean? Yeah, I know exactly what that means.

Cole
Exactly.

Nick
Well, I think I'm in celebration of this conversation with you. I think I'm gonna have a hot dog.

Cole
Yeah.

Nick
So Cole. Thanks so much for coming on talking about Flyreel. You're always welcome to come back.

Cole
Thank you so much, Nick. I appreciate you.

Nick
Appreciate it.