We recently had the opportunity to catch up with Attila Toth, CEO, zesty.ai, the Silver Winner of the 2019 Zurich Innovation World Championship, to discuss how Artificial Intelligence is impacting the Property & Casualty Insurance market across personal and commercial lines. Click the link to have a listen. You can also read the full transcript of the conversation below.
Ron Glozman: Hello and welcome to the Chisel AI inaugural podcast. With me today is Attila Toth, CEO of zesty.ai based out of California. Today we have an interesting show planned for you where we're going to talk about the global insurance industry as it undergoes a digital transformation.
As insurance companies find themselves trying to make sense of all these new technologies – artificial intelligence, natural language processing, machine learning, computer vision, – understanding the business case for each can be extremely confusing and daunting. With insurance companies being held to higher customer expectations, the time is now to embrace new technologies to leapfrog the competition. Being status quo is no longer an option. Technology is driving diversity across many industries – insurance included – as it reshapes the value chain. Age-old processes are being disrupted, while new market entrants and changing business models are bringing new threats, as well as opportunities for those who act on them.
Get our viewpoints delivered to you inbox
Some of the questions we'll cover today include: What is the value that AI is delivering to the insurance industry, and how are insurance providers reacting to these seismic changes?
Today, we'll specifically look into how AI, or artificial intelligence, is positively impacting personal and commercial insurance. I'm Ron Glozman, CEO and Founder of Chisel AI, a Toronto based AI technology provider for the global commercial insurance industry. And with me today is special guest Attila Toth, CEO of zesty.ai, a California-based property risk analytics platform for the P&C insurance market. Attila, welcome to our podcast. And congratulations on winning silver at the recent 2019 Zurich Innovation World Championship.
Attila Toth: Thank you, Ron, and yes, indeed, we faced some tough competition in Zurich such as Chisel AI.
Ron: I'm glad we could represent North America and bring home the gold and the silver for these wonderful countries of Canada and the United States.
Attila: It was a great honor to share the podium with you.
Ron: So, let's jump into it. I'd love to hear a little bit about your views from your conversations with partners and customers regarding the current state of property and casualty insurance as well as an overview of what zesty.ai does.
Attila: Thank you, Ron. So, let me start with a quick overview of who we are. zesty.ai is a property risk analytics platform powered by artificial intelligence.
"We at zesty.ai believe that there is a tremendous opportunity to revolutionize a three-hundred-year-old industry, which is property and casualty insurance, and help them underwrite risk and leverage data to delight customers."
At zesty.ai, we've taken a lot of data – some structured data such as, for example, data from weather stations or data from building permits – but mostly unstructured data such as imagery data from satellite providers or low-flying aircraft, and we have developed a cutting-edge machine learning technology in order to extract key risk modifiers from this data.
To date, we have amassed about 115 billion data points on buildings and their surroundings in North America without ever setting foot on the premises of these properties. We not only extract data for understanding risk modifiers, but we also build risk models to understand catastrophic loss events such as hurricanes, wildfires, or floods, for example, with unprecedented accuracy.
So, Ron, you asked what we hear from the property and casualty insurance industry about why they should care about AI. There are two things that I believe represent a great opportunity for disruptive technologies for property and casualty. First, much of the data that is being used for underwriting risk is based on "best guess" data that is self-reported by the insured or by the agent, data that is extrapolated, or data that is simply missing. And, by the way, when I meet with insurance executives, I get a lot of head nodding when we talk about data fidelity and the opportunities therein.
"The second opportunity is that, in the age of Amazon, insurance – and property & casualty insurance, specifically – is one of the few remaining categories that is sold exactly the same way as it was 50 years ago. So, we believe that the customer experience needs to be radically modernized at every single touch point."
Ron, I am very happy and enthusiastic to be at the forefront of leveraging digital technologies, particularly AI, to help modernize property and casualty insurance.
Ron: I love that. And I know in one of our conversations prior to this recording we had talked a little bit about the California wildfires which had just finished. Very unfortunate circumstances of events, but what I thought was really interesting – and can bring this a little bit closer to home for some of our listeners – is the ability to actually assess brush and other combustibles around the house and predict which properties are more susceptible to wildfire. I think that is something that, of course, the underwriters of the risk care about, but property owners would love to know what they can do to reduce the risk of their properties suffering losses. I think some of the work you're doing in that space is just phenomenal.
To jump a little bit further down the road, in some of our work with insurance companies, as you pointed out, a lot of the data they're sitting on is unstructured. And so, you know, where Chisel plays in this space is to structure specifically PDF documents, Word documents, the digital documentation like policy statements, etc. and extract and identify data points. So, similar to how zesty does it for P&C primarily in the property space, but for commercial insurance. You mentioned zesty.ai has the ability to extract, I think you said, a 150 billion data points, which is just unbelievable. I wish we had that many data points. I think it's a slightly different use case for us where we're able to extract hundreds of data points from an individual policy and the underwriters or the brokers can then use that to automate policy checking, quote comparison, submission triage, and submission prioritization.
So, having said that, obviously AI is going to have a significant impact on insurance. Some of the work that we've seen includes better access to data, and reduced inspection costs. Can you speak a little bit more to how you think the data you're providing will help both the policyholders and the underwriters be smarter?
Attila: Absolutely. The data in the models that we are providing can be leveraged within the P&C space across the entire value chain. So, what do I mean by that? You can use the data as inputs into your underwriting models. Whether you're underwriting as a P&C carrier or an MGA, or writing one single policy or a portfolio of policies, you can just ingest the data and use it for better underwriting.
You can also use the data for inspection optimization. We are not trying to completely replace inspections, but we are providing intelligence about which properties to inspect and not to inspect. And also, for the vast majority of properties that are not inspected, we are providing significant data assets that insurance companies don't have today.
So, the first opportunity helps with lowering loss ratios and gaining market share with better underwriting. But this technology can also help you increase customer satisfaction which, as we discussed early on, is a pain point in the industry. How does it help with customer satisfaction? We can pre-fill a lot of data in the coding process and make a very smooth online customer and agent experience instead of asking all the crazy questions.
By the way, one anecdote: The way my co-founder and I got into this industry is going through the home insurance acquisition process.
"We were shopping for home insurance and the questions that we were asked as customers were just mind-boggling: 'Sir, what is the pitch of your roof?' Or, 'Sir, what is the slope of the parcel that your home is built on?' No homeowner should be expected to know this data. That's where zesty.ai comes in. We provide that data."
And on the backend, meaning claims management, this type of technology is going to help insurance carriers to very quickly assess damage after a catastrophic event, and help with claims – claim adjustment, claim verification, and fraud management.
You also asked, Ron, "Why should consumers care?" The insurer and insured should care. They should care because they're going to get a better experience, a more data-driven digital experience. That's number one. Number two: the interests of the insurance company and the insured are aligned, right? Insurance companies don't want losses.
So, if the insurance company has data about potential events like the wildfire and forest fire events you mentioned earlier, they can alert and send that data to their customers, to the end users of the insurance policies, and give them actionable insights about what they can do to decrease their risk profile.
In the case of wildfire, the insurance company can tell them to take down overhanging vegetation, clear out vegetation in the defensible space around the property, or even update their vents to resist these events. Those are a couple of action items that the insurance company, based on this data, can convey to the end users, thereby decreasing their risk.
Ron: Love it! So, without giving away the secret sauce because, of course, that's something that you're going to want to keep secret, can you talk a little bit about some of the different types of AI you use? As we mentioned, there’s computer vision, deep learning, natural language processing, RPA, and dozens of other technologies that are coming to market. Where have you seen success? And for people who are looking to potentially pilot something in-house with their innovation team, what might they want to steer away from?
Attila: As you know, Ron, since you are at the forefront of this, AI is a very broad term. And within AI, there are multiple models, and multiple different technologies to which you can apply AI. First of all, I always say, "Never start with the technology in mind; always start with a business case." So, my recommendation for insurance companies that would like to pilot this or build this in-house is always start with a very strong business case.
"There are so many buzzwords out there about digitalization and innovation, and many people are enamored by the potential of these technologies. At the same time, they're applying the technology for technology’s sake and that’s a moot point. Do not do that. Start with a business case in mind."
Once you have a business case, then think about what data you will need for that business case. Where are you going to get the data from, and how can AI help you gather that data?
So, let's assume you're thinking about the underwriting of properties, and let's assume you are thinking about leveraging imagery from satellites and low-flying aircraft in order to better understand properties, the risk factors inherent in properties, and how those risks change over time. In this case, we would recommend using convolutional neural networks, which is basically teaching a computer to think like the human brain.
I sometimes use the parallel of teaching your 10-year-old kid to play the violin. So, you provide a lot of training materials. The kid is going to make some mistakes and the chords are not going to sound right at first.
That's exactly the same process for neural networks. You will be training your computer vision models to see as the human eye does, or as the human brain, or the human neuron understands vision. But the good news is that you will be able to do this at a tremendous scale – a lot larger scale than what humans can do, and with a lot higher level accuracy. What do I mean by that?
"We have looked at accuracy for physical inspection reports from property insurance companies and we have seen that accuracy – feature level accuracy – ranged between somewhere in the high 60 to the mid 70 percent accuracy. With computer models you can get close to a hundred percent accuracy. Nothing is a hundred percent, but you can get to the high 90 percent accuracy on a feature level understanding."
So, that's one example. The second example is, "now you have all this data, how do you build a model out of it?" The industry historically has resorted to statistical regression modeling. With statistical regression modeling, if you have an equation, you are fitting to a curve, and that has a lot of limitations. I don't want to get into the technical details, but it has a lot of limitations. So you could use things like gradient boosted trees, for example. That's a different type of AI model. In order to build a predictive AI-enabled model that can help you predict certain outcomes, you need to train the model with outcomes based on your loss history.
Ron: I love that. First of all, thank you. That was in-depth and hopefully the listeners will be able to do more research if they'd like to learn more about some of the specific data points. But I think you hit the nail on the head. And one of the things that was even more interesting from the bigger, broader technology conversation is, "What’s the right strategy look like? What is the core element to implementing a successful AI strategy?"
I think you hit the nail on the head when you set a strong business case. I have heard this from our customers, as well as from other start-ups, scale-ups, and full-blown established vendors who are coming into these projects. There has been a mandate set at the executive level to get an AI implemented. They meet with the innovation team – and don't get me wrong, we love meeting with the innovation team – but unless there is a strong business case and an operational person who will take responsibility for deployment, it is very tough to make AI work.
"AI for the sake of AI is not going to be a great success for the vendor, the customer or the actual policyholder. AI really needs to support the operational part of the business."
So, can we dive into that for a minute or two, and talk about what a successful AI strategy should look like?
Attila: Awesome! You hit the nail on the head there. Any innovation initiatives that do not translate into operational metrics and operational success are a failure. By the way, I think the industry should embrace failure, but not on projects that are set up to fail. Failure is okay if you set up the project right.
If you have a strong business case, a strong technology selection, the right resources on this project, and you partner well, failure is okay. And that's what I see in the insurance industry.
"By the very nature of the industry, executives are quite risk averse. That's how they make money and that's how they earn the premiums. At the same time, the pace of technological change is so fast that you have to embrace failure and you need to be okay with failure."
So that's one point I want to stress. The second point is around partnerships.
"Cutting-edge innovation, in my opinion, requires revolutionary thinking, not evolutionary thinking. I believe that this type of thinking rarely happens within very large organizations. And why is that? Because larger organizations have different priorities – they cater to Wall Street. You know, they have organizational structures that don't lead to very agile idea sharing and trying. So large insurance companies will need to partner."
In more cases, they will need to partner rather than build this type of solution in-house. Many of them are capable of building this type of solutions in-house, but the question is speed to market.
You also have to think about your technology adoption cycle when you're partnering with fast-moving startups. These startups are on a very different cycle for technology delivery and, let's be honest, on their fundraising cycle as well. So, if your technology adoption cycle as a large insurance company is two or three years, but your partner is on a short leash of one, one and a half, maybe two years of funding cycles, they will not last before you guys produce results together.
Ron: I love that. I was recently speaking on a panel at OnRamp where we talked about how to set up companies for success, and I love your view that failure is good and healthy as long as the failure is not due to a bad sort of arrangement of the parts, or a weak business case. They often say you learn more from failure than you do from success. You'll know where to be better next time.
I also love your point on funding. We have come across this, and I've spoken with other start-up founders where the cycle for sales can be two or three years and, as you point out, the typical start-up will raise money for 18 to 24 months of runway.
It falls to the insurance carrier, broker, reinsurer, or any of the other players in the ecosystem to have sort of a fast-track or a "startup-friendly" experiment where you can go through procurement, sales, compliance, and the pilot in a short period of time, let's say no longer than six month period, otherwise the company might falter and stall purely due to not being able to meet the milestones for the next raise. That's a loss when it comes to pushing the technology forward. So, it definitely falls on the partners to make sure that they're looking at who they're partnering with and how to work with them.
Attila: Not just from the start-up's perspective, but also from the insurance company's perspective. I think gradual but fast experimentation is paramount. Have a great business case, understand what the business case could do to your entire business, but start small. Celebrate early wins, and build on those early wins, but move fast from the first smaller pilot to a larger expansion in terms of geography or across lines of business.
I mean, that's what I love about this industry: The opportunities for expansion are just unlimited, right? Wherever we look, there are a lot of opportunities, but I think we've got to make sure that we prove that there are results today and tomorrow before we start dreaming about how AI is going to completely revolutionize this industry in the next five to ten years.
Ron: I love that. Where my mind jumps is, I can't wait for the day where we see zesty producing information on Mars. So I know where to buy a nice piece of land for my house.
Attila: Not in our five-year plan Ron. We are sticking to our guns and we are sticking to property and casualty insurance! (Laughs)
Ron: Fair enough! So, a slight change of pace regarding an objection that we often hear. You know, as AI gets smarter and even in its current state, a lot of people are worried that it will replace skilled knowledge workers who may be at risk of being displaced and laid off. I'd love to hear your thoughts on that.
Attila: So, let me start with a small personal anecdote. I came to the U.S. in 1994 as a student and I will never forget the most important burning question on my mind when I landed in Chicago: What is this thing called the World Wide Web? And how is this going to change our lives? I remember sitting in the university computer lab in front of a black screen with some green gibberish on it, and saying, "Ok, this is the World Wide Web. What is this thing gonna do?"
Fast-forward twenty-five years and the Internet has changed my life in many, many aspects, and it has changed a lot of industries like communication, entertainment, financial services, you name it.
"I think we are at that same watershed moment with AI as we were with the Internet twenty-five years ago, but with two differences. It's gonna happen a lot faster and the impact is gonna be a lot bigger."
But let's look back on the past twenty-five years – should we say that the Internet has taken away jobs for humanity, should we say that the internet is for worse or for better for humanity's sake?
"I believe that we are going to be sitting here 10 years from now and we're going to look back and say, yes, AI has cost some reallocation of resources. At the same time, the net impact on our lives, on the insurance customers' lives, I think, is going to be positive."
And let me tell you the zesty story, and how I believe that the net impact is positive. We are not focused on replacing labor at insurance carriers. We are focused on better understanding risk.
If you better understand risk, you the insurance company and your customers are better prepared for those catastrophic events. A few quick stats: The number of catastrophic loss events is increasing exponentially. Out of the five largest catastrophic loss years ever recorded, including hurricanes, earthquakes, and wildfires, four fell into our current decade. So, this thing seems to be getting worse rather than better, right?
If we are using smart AI to better understand how we can protect ourselves and our communities against those type of catastrophic risk, I think that the societal impact on a net basis is positive for AI. And that's our mission at zesty.ai.
Ron: I love that. And I wish I was around in 1994. That is just about a year before I was on this planet.
Attila: You were probably still in diapers. (Laughs)
Ron: You know, my perspective is very similar and I also have sort of a personal anecdote. I remember walking into a client's office when we were going through the implementation stage of one of our projects. I sat down with one of their underwriters and I asked a simple question, “Tell me a bit about your workday. Aside from some specific tasks, what's it like?” And this person said they come in often as early as 8 a.m. and often don't leave before 7 or 8 p.m. simply because it is very time-consuming work, and there's more work than they're able to handle in a year.
For submission intake, some of our customers say they're only handling 40 to 50 percent of the submissions they get, and that's with people working more than 9 to 5. And that means people are missing their son's first baseball game, their daughter's ballet recital, they're missing their anniversary dinner. They're spending less time with loved ones and less time doing the things that they like. And so, what we can do is reduce the amount of time they spend doing routine, manual, non-judgment-based, non-value-added work? I think that's a win because they'll be able to spend more time with their family, be able to go home at 5:00 p.m. and not miss any important events.
So that's our mission at Chisel similar to you guys at zesty. Our company motto is we exist to help people work smart and enrich their lives. If we can do that, we're happy no matter the company valuation. It's all about the people at the end of the day. You can’t have a business without people, and we want to make not just the people in our company happy, but the people on the other side of the table happy as well. And at the end of the day, the policyholder should be extremely happy.
Attila: Well said, Ron.
Ron: So, in terms of wrapping this up, I'd love to hear a little bit about your predictions. We talked a bit about where we see insurance today and some of the ways zesty is helping to shape the industry and enrich the data that's coming out. Where do you see the insurance industry going in the next three to five years?
Attila: As it pertains to AI or in general?
Ron: I'll let you take this question any way you'd like to go.
Attila: Ok, let me start with in general and then drill deeper. This is an AI-focused conversation, so let me focus on the adoption of AI within the industry.
In general, I believe that there is going to be a lot of financial pressure on the industry because of the catastrophic losses that I have already alluded to. At the same time, there will also be a lot of competitive pressure.
"Consumers are used to the Amazon experience in every facet of their life, and they will be demanding it from property and casualty insurance as well."
So, I think there will be competitive pressure. There are a few distribution Insurtech companies like Hippo Insurance, for example, an extremely talented group of people who we have partnered with. They make a great customer experience and they're growing fast. But I believe there will be more of this competitive pressure coming maybe from the Amazons of the world. In light of that, I think the industry has no other option but to embrace technology with open arms.
And insurance historically has been a little bit slow in terms of adopting technology. Why? Because of the risk-averse nature of the industry. We spoke about that earlier, but I believe AI is going to be different. AI is cutting to the core or to what an insurance company is about. Insurance companies are all about data, modeling, understanding and pricing risk. And that's where AI comes in.
AI offers a lot of new types of data and more efficient models. So, in terms of my prediction for adoption, I think in Insurtech, AI with large insurance carriers has moved beyond the "meet and greet" phase. Previously, we were just shaking hands, you know, talking to innovation departments and exchanging business cards, but not much happened. Then we moved into the pilot phase with pilots left and right. Insurers were saying, "I want to pilot this, and I want a pilot that."
"But now we are graduating from that pilot phase and, finally, we are seeing a lot of production scale implementations of AI. That makes me very happy and I believe that, no pun intended, this is going to spread like wildfire."
I believe that once people see the value in terms of a better risk selection, in terms of better customer service, in terms of better-combined ratio, in terms of market share gain opportunities, adoption is going to kick into high gear.
Ron: I love that and I'm on the same line. To reference – I believe this comes from Crossing the Chasm – there are four types of adopters: the early adopter, the early majority, the late majority, and the laggards. If I had to make a prediction, I think we're past the early adopters stage with AI. We're starting to see several production deployments in some big companies.
We're now getting to the early majority stage probably this year or next year. I believe that if, by 2024-25, five years down the line, you aren't using AI, you're a laggard. And there will be companies out there with five years of production experience already in the market. And it's very hard to catch up, especially depending on the types of models that you're looking to implement. It can take quite a bit of time to train the model and train it to sort of read and see or understand as the human mind does. And so, my urging to the audience would be to start on that now. Implementations can take a lot longer than you think.
I was listening to another podcast earlier, and Marc Benioff was on it and he said humans often overestimate what they can do in one year and underestimate what they can do in a decade. I believe it's time to start because, if you don't start now, you will be in the laggard or late adopter or late majority phase. And for any company that believes that they want to become a market player, it's going to be key to make those investments early.
So, having said all that, if you have any final thoughts, please go ahead.
Attila: I agree with your comments. This is prime time for execution. Let's make it happen!
Ron: Perfect. Attila, thank you so much. This has been a fascinating conversation and, hopefully, we were able to provide some valuable information to our listeners. Thank you for your time. If you enjoyed this episode, please connect with us on LinkedIn and Twitter. And, of course, you can visit zesty.ai and Chisel AI to find out more.
Attila: Thank you, Ron. Thanks for the opportunity.