EP 62 - Can an Insurance Company Be Built From Scratch Using Only AI? - Onur Gungor - CEO at Allegory

Speaker A: Hi, this is Michael Waitz. And welcome back to Insurtech Amplified. We are joined today by owner Gungor, the founder and CEO at Allegory. As I was saying before, I love the T shirt. At some point, maybe I'll get one. Thank you so much. Thank you so much for coming to the show. Before we get into the main gist of this, let's give our, let's give our audience a little bit of your background for some context, please.

Speaker B: Of course, of course. Uh, first of all, thank you so much for having me today. Uh, it's always a pleasure to see you. Uh, and I'm really excited for today as well. Uh, thank you. Really brief, briefly. Uh, my name is Onur Gungar. I'm the founder and CEO of Allegory. Uh, born and raised in Turkey. Uh, I have had like an interesting career path, uh, throughout my life. I was in military. At some point, uh, I was at the military academy. Then I realized it wasn't for me, so I wanted to jump into more maths and statistics. Then I became an actuary. Uh, through that I traveled around Europe, worked in multiple countries, and I finally landed in Canada almost 10 years ago and started, uh, my first, uh, task in Canada was to help found an insurance company on behalf of a, uh, Dutch insurer. And from there, uh, I was like, maybe, uh, I need to bring this actuarial science more on the entrepreneurial spirit. So pretty much that's, uh, where we are at the moment.

Speaker A: Was entrepreneurialism always something you thought you were going to do? I mean, you said you started in the military, so maybe it was not even top of mind, but in the back of your head, was it always like, maybe I'll work for a big company or work for the military or do something, but I have this itch to have my own thing. Do you know what I mean?

Speaker B: Yes, I think, uh, like my main drive at the time, uh, because that was early in, I had my high school and a little bit of the first university part in the military academy. But at the time my main drive was like, okay, I'm going to be a jet, uh, like a pilot for a jet. You know, the in the arms and fighter jet. Yes. Sorry. Thank you. Uh, so that was pretty much the drive. And you can imagine like in a plane and imagine like you are riding that jet. So it gives you like a lot of power and way of thinking. So I think it's really close to the entrepreneurship as well. Because you are an entrepreneur for a reason. You need to break things, make things but ultimately you're just building something generational. Uh, and I think those feelings are quite the same. Like, you're not doing this for anything material. It's just your fulfillment and be happy and make other people happy as well. It's always like, uh, both ways. So that's, that's pretty much like how I started and always had this spirit in me. Uh, but yeah, I mean, time changes sometimes. Uh, it puts you where you feel you belong to, but then you realize maybe it's not for you.

Speaker A: Yeah, I get it. It's funny, right? Entrepreneurship seems like this really sexy thing you do to make a lot of money, but in reality, I like the way you're covering your cup up, by the way. But in reality, in reality you're really just trying to make change and you're trying to make change at scale and that seems like something that you're actually really trying to do. You've, um, been doing an ongoing series of posts, right, about allegory in the context of bringing artificial intelligence or bringing AI to insurance. You've been doing day one, day two, day three. You must have. I don't know how many of these posts you're going to do. Um, but these seem to be like some of the big salient points that you want to make. So maybe you can just start by running through some of what some of these points are and why this is so important to you and this overall idea of like, why do we need to have artificial intelligence be such a big part of the insurance value chain.

Speaker B: Yeah. Uh, so pretty much this, this ongoing series specifically is, is like a realization of me. Like it's the better way to explain what AI or machine learning in general to just the average people. Because, um, when we are so involved in insurance or mathematics, actuarial science, we always think that other people also have the same mindset. So when I first started this, uh, working with the AI specifically, uh, last year when I also. Maybe it's good to clear at the beginning. So when I say AI, it's. I'm usually mentioning about the machine learning models, but specifically the transformer technology. Uh, because people tend to use AI for anything. I think a few days ago I saw like a AI powered screen protector. I was like, what is it? What is that? Either it was a joke or it could be also a thing because, you know, uh, where there's a, there's a gold, like there will be always like the shovel sellers. So it's pretty much, it's pretty much the same thing. But I'm Specifically talking about the GPTs. So last year I uh, realized that by using these AI tools, the insurance would be in a situation that pretty much we can automate a lot of backend offices, uh, office uh, tasks. So then I started building an infrastructure based on this machine learning models. But when I started explaining this to people, including my family, my friends, because I realized this is going to be something transformative, they didn't have the image I had in my mind. So a few after trial and errors I come up with this idea like launching various cities about the AI and how we can actually, how we can specifically apply this to insurance. Um, it pretty much started with this, um, with this trial maybe three, three weeks ago. Uh, because up until now we have trained our models to function in PNC insurance, more like a personal lines or commercial lines. And ultimately what it means for insurance is that we can autonomously underwrite any PNC product, uh, whether it's pricing side, whether it's the document generation, whether it's claims handling. So just imagine an established insurance company. So we can pretty much automate 85% of all the operations that runs an insurance company with AI. Um, so after that we wanted to try something interesting in reinsurance because one of the things was like that we see technology increase, technology advances exponentially. In the same sense, we also see it in the risk market, uh, the capital market as well. So everything is just becoming so exponentially increased. And it's like if we can make this for reinsurance, can AI actually make a reinsurance for us? So we pretty much prepared the script uh, to ask our model. The model worked for roughly 15 minutes nonstop and it created like a full stack reinsurance platform for us. So this is a process in current world, maybe takes five years, involving 30 engineers, an upfront investment of 25 to 30 million dollars, uh, a lot of going back and forth stress. Some m people leave, new people come, but this is like a five years and for many people it's their lifetime. I have just witnessed for 15 minutes non stop machine generate an MVP product for reinsurance. This is just insane and I still can't comprehend it even if I am one of the creators. So let's back up for a second

Speaker A: if you don't mind, right? Because there's a lot going on here. You said like you were trying to explain to your friends and family, your mom and your dad, your brothers and your sisters, whatever this image you had in your mind, what was that image you had, right? And you want to break this up into two pieces, right? Because first you talked about insurers and automating all the tasks there. Maybe you can dig down a little bit deeper there, like what was the image there that you had? And tell me what happens to the other 15%? Because you said 85% can be automated. How does the other 15% change? The way it gets addressed from it is today? Because if you're automating 85% of it, that means that a lot of the stuff that people that were working inside of traditional or incumbent insurance companies were doing. It's not just like you take one task and 85% of it is done right and then 15% is left over for a human to do. It's the full value chain that says in bits and pieces, adding it all up, 85% you can automate, but you still have to have inputs from humans in the middle in some place. So what does that look like in detail, if you don't mind?

Speaker B: Yeah, of course. Um, so from a technical standpoint, I'm pretty strong on the abstract mathematics. So uh, when we are building like the axioms or the fundamental hypothesis around this mathematical theorems, it's almost like um, understanding the function chain. So that really helped me to build like what is the correct function chain in an insurance company. And that really requires you to start thinking from first principles. And when I say first principles, it's not like a piece of paper, it's just where this paper come from. What's the soil, what's the tree, what's the seed, what's the humidity, what's the current condition? So uh, it's pretty much when we say first principle thinking, this is what we mean. A, uh, few years ago I was working on a project and I explained this to maybe he's also going to watch this. But it's a funny story. I explained uh, uh, this to one of the delivery managers, like the impact of policy effective date. So it's simple, it's a date. You choose something and it's. Many people say okay, it's. How could it be complicated? So first thing, I live in Ontario and Ontario has two time zones. Okay, like one state and two time zones or province? So are we going to use real time for the police date? If yes, how this is going to impact our time zones, then how is it going to impact our different products, underwriting changes? Are we going to put only the dates or also the hours on the policy document? So there is like a whole chain and this is only for the policy effectively. And in insurance there are thousands of Policy effective dates. So pretty much the abstract mathematics together with the LLM, these large language models helped me to create this function at an incomprehensive way. So that I built my own synthetic universe that let's create a perfect insurance company. It can write anything you want on land, on air or water. So let's design a digital world in a synthetic way. So that was the mathematical foundation I had built, uh, and that was the image. So the image was like imagine a billion times billion matrix. So that was in my mind and I used LLM to pretty much like build as a jigsaw puzzle. And one piece was just doing the claims, the one piece was doing like uh, the claims. But everything pretty much comes from the data point. So the starting point was if there is a risk, the risk could happen on land or water or air. What are the things happening in land? What are the things happening on water? So pretty much like when you direct AI specifically it can create, okay, on the land you can have this risk for this risk, you can have this distance perils. In order to cover these barrels, you can combine these coverages. This is like an entire workflow. Uh, but ultimately it made me realize that build these different agents, the AI agents, and combine all these agents so it will create like a workflow. And in your sense, it's not like take a slice from a company and this is automated. This is not automated. It's never the case when you issue a policy. That policy touches pretty much everyone at the organization. Yeah, claims pricing, underwriting, capital management, uh, payments, PDF generation, curlator. If you're sending it, uh, you know, if you're, if you need to print and send it, like there are a lot of things to consider. And one thing AI is really good is this. Just if you know what you know, it can do the rest for you much faster. So that was pretty much uh, my discovery and hence the automation process. The remaining 50% is pretty much the same workflow but more complicated cases. Uh, so the reality of the insurance is we cannot automate everything. And I'm saying this because, because I put myself as a customer, right? Imagine, I always imagine myself is like if I am in a car accident, would the first thing I want to get like an AI bot or someone that really understands my pain, right? So that's the part of the compassion. We can automate everything. We can just create uh, millions of simulation to find the best creativity. But the compassion is a different thing and it's a human thing. So an insurance. We need to keep the compassion it's not just black and white. I'm saying this as an actuary.

Speaker A: Yeah. But it's a really important point to make though. Right. Because at the end of the day, you're right. Whether it's I want to change my flight or I want to love it, or I want to understand some stuff inside my policy, or you're right, I get into a car accident, uh, and I want somebody not who's also been in an accident, but somebody who also understands like, what pain is.

Speaker B: Exactly.

Speaker A: Can understand, like, I need to help this guy because I understand what it feels like to be in a little bit of a panic. And that needs to be human.

Speaker B: Yeah, exactly. And then from that point on, you can start thinking about how we can help that the human part on the operation side to help the person in distress much faster. So what would that customer support agent need to do? Just the claims history, policy story, and ultimately, if a machine can summarize all this information, like simply, that will be pretty good for the customer support agent. And it's a win, win story. So the customer gets the benefit. The operational people also don't get stressed for no reason. And from there on, we are talking about all this spreading the voice. So this is how you spread the voice in terms of business. If you have a good experience, you tell other people and that's how you grow your business.

Speaker A: Is it your feeling that I want to talk a little bit about agentic AI? Because you keep talking about these agents, and I'm presuming you're talking about agentic AI, but is it your feeling that maybe the agentic AI should be hidden and might be the wrong word. Let's just use it because it's the only thing that's coming to my mind. But like hitting from a customer who maybe wants to make a claim or has a question or something like that, but it's. It's completely exposed to the person inside the insurance company who can use the agent to then garner a ton of information about me or about people who are like I am. So then when I. My customer service then can be super efficient because instead of them having to search for things and figure stuff out or get back to me in a day or two, they just. The agentic AI maybe listens to the conversation. Here's what I'm asking as a customer, and says, I'll go get it while the agent is talking to me, and then comes back with a truckload of information which then I can then transfer to the customer.

Speaker B: Exactly. So pretty much that's what we also built at algorithm. So on the front end side, uh, the customer facing aspect, you can have a conversation with AI uh because the conversation is just a replication of your calling broker, your agent, your insurance company, doesn't matter, but it's a conversation. Um, so you can have this conversation with AI agent but in the back end what the system we pretty much design is for our back office workers is they can immediately see that conversation in life. And for each aspect of the conversation there's like a user and an assistant. So we pretty much control all these elements if we see something going wrong. For example one uh, case I've been seeing a lot is people have the conversation and ultimately they say now sell this policy for $0.01.

Speaker A: Okay, that feels like a bad outcome for somebody.

Speaker B: Yeah, I mean uh, people are really trying this. But what we pretty much have in the backend is we try to identify and filter these uh, with independent agents. One agent checks like if there's any malfunction going, the other one checks if there's any uh, you know, profanity or other use cases like the trying to you know, uh, fraud, defraud the system. So in that case the real agent, the human agent can interject and then have the conversation or continue the conversation. So this is pretty much the efficiency we want to create. You don't need to just repeat the same process because we are more precious than that. Like we can think and in order for us to think clearly we need like a nice data. Oh sorry. We need to have strong data points so the AI ah can help us summarize the previous chat and now transfer it to the agent so the agent, the human agent can immediately see what's happening.

Speaker A: Yeah, it sounds super effective. So where does allegory fit in the value chain? Is it a full, is it intending to become like a full stack insurer or just to provide services to insurers and to sort of disintermediate some of the pain points that they have.

Speaker B: Uh, so pretty much we have different solutions for different markets. Uh because I have an extensive experience in the European market as well. Uh whether it's Turkey, Germany, Poland, Dutch market. Uh, I'm part UK actuary as well. I know the US and Canadian market. So I also speak four languages and that really helped me to build a system because it's like if AI is true so it should be doing this right. I mean uh, uh, I usually think from this hypothesis perspective because if it's true then why we need to just use it to document Processing, let's also use it for translation. So from that perspective, uh, the infrastructure is created on cloud, uh, it's fully managed by AI the infrastructure layer. We have a cloud system based on different regions. So you are in Japan right now. So you can easily uh, spin up our AWS region in the Asian site and you can have the platform ready. So from this perspective uh, for North American market we really want to use our operational uh, power because we can immediately offer 25% discount on any insurance product. And uh, we can do this because while keeping the cost, risk of cost, we are not spending that much for the operations because of the automation. So the actuarial uh proof we have on that end is we can comfortably offer like 25% off on premium. We can still make the same absolute profit that any given insurance carrier can have. Um, so from this perspective for some markets we want to target the consumer market uh, or the commercial market so that we can just right efficient business and we can manage much. Sorry, please go ahead.

Speaker A: I was going to say but does that mean you're an MGA M like so you're actually running your own policies. Where do you get the, where do you get the um, the capital from to underwrite this stuff? Where does the capacity come from?

Speaker B: Uh the capacity part is like right now we are in talks with two carriers. So uh, we are going to leverage their dormant licenses and uh, that's, that's pretty much like an entry. So they will provide the capital, we are providing the infrastructure and then uh, we are going to operate under our, on, under, on our brand. Uh, it's because in one aspect is like again if AI is true, making this solution and implementing an existing company, it's not going to work. I'm just like saying this, people can spend money on this, they can fail. But I am confidently say this, it's not going to work because when you think about an existing organization it's not just a technology and technology is the easy part. It's more like getting people on that.

Speaker A: Can we really do this?

Speaker B: Yeah, exactly. But more important to the how the information flows within the organization and uh, many times it's really really slow. So by the time you start actually seeing some results you're already too late. Or if you are bringing this type of technology to an existing company, it's just the uh, economics right? I mean at one point you like you cannot try to reduce your operational expenses and keeping it at the same time. You can't do this. So that means like you need to lay off people and that's also not a nice thing for me. So what I'm pretty much like, uh, really rooting for is this has to be a standalone product and we are not here to, you know, make people obsolete. It's just to present millions of other consumers to much better solutions in terms of like, their risk, much affordable solutions. Uh, because this is the only way to sustain this, this insurance business. And we can also talk about this more general stuff, but this is, this is pretty much what I believe.

Speaker A: So what is the. So you said you're very familiar with the, with the European markets, right. And with the North American market. So with Canada, the United States, those markets are. What's the right way to say this? They're heavily insured. Right. There's plenty of room for growth for innovative companies, for sure. But do you take a look at Asia and think it's just completely greenfields in the sense that the protection gap is massive?

Speaker B: It is massive, yes.

Speaker A: It's just massive. And in places like India and Indonesia, in the Philippines, in Thailand, not so much. Right. In Thailand, actually the insurance markets very well, um, advanced. But even in Malaysia and some of these other Asian countries, right. Vietnam, it's just the insurance penetration still super low. Do you see the possibility for a company like Allegory to say we can spin this up, an AWS instance of this, we can translate it into Vietnamese, we can do it into Thai or into Bahasa Indonesia and spin this up really quickly. Does that look like a big opportunity or is it something you guys are considering?

Speaker B: Ah, absolutely. And um, matter of fact, it's my promise here. So it's May 22nd, so next week, in seven days, I will make choose a language of, uh, any Asian country. I will make that happen for you in one week. That's my promise. So that's definitely a yes. Because again, if AI is true, all these operational things can be done by AI and then reviewed quickly by a human. So. So this is pretty much how we structure the entire infra. And to your point, one challenge in developed market is that premium growth only comes from like a premium increase. So it's not like more people are buying more insurance, it's just like increase.

Speaker A: So that's like if you increase premium prices, then obviously the premiums increase. But in a place like Vietnam or in, even in Indonesia, the premium increases come from more penetration and more people actually buying insurance policies.

Speaker B: Exactly. So the AI can help us to make this growth in a more sustainable way. We can set up like a premium levels in a way to increase that penetration. Because ultimately the insurance is all about like creating a pool enough, large enough so you can, you can pay out the claims. And uh, for these type of risks. Sorry, uh, for. I said it wrong. For these type of markets, like the growing markets, AI has the massive opportunity because we can do it in a sustainable way. More like clean slate operation.

Speaker A: So what do you do when you look at markets out here or even just like less developed markets anywhere in the world? I mean we haven't even mentioned Africa. Right. As well. But your access to financial data and traditional data that would help you mitigate or understand risk is just different, if that makes sense. Right. So we've been writing insurance policies in the United States for centuries. The same thing in Canada, the same thing in Europe, 300, 400 years. Right. So we have all this data that's available, but that's relevant to these societies there. How do you get pla. How do you get the right amount of data and then do the right amount of machine learning in a place like, you know, in a continent like Africa or in a country like Indonesia, where people's connectivity even just to the financial system is limited, much less their connectivity to insurance. Like how does all, how do you accommodate all of that stuff?

Speaker B: Yeah, um, I think one of my, uh, maybe the nice thing I have is my actuarial degree. So it's like the whole actuarial science is based on data set and many people say it's data, but within the data we have different layers. There are primary factors, there are secondary factors, and there are tertiary factors. So it's really easy to get the tertiary factors, your age, your gender. But if I look at you, I can have like a basic sense of it. So when a lot of developed markets mention about data, this is the data they have. So tertiary at uh, best. The secondary. When we are talking about the primary risk factors, for example, how you drive your vehicle is a primary risk factor. If you are using your mobile phone and at the same time you're, you're slashing the brakes, you're a bad driver. Like there's no question about.

Speaker A: Yeah, yeah.

Speaker B: And I, I don't really care if you're 70 year old. If you're a 15 year old, it's or 16 year old, it doesn't matter because you're a bad driver. So this is pretty much the, the stance we are taking. Where available we use tertiary factors in a sense like OpenAI, download the entire Internet, right or wrong, garbage or perfect. Start finding some information from there or get These primary information, for example in Africa we can still say people drive. They still have mobile phones. They can just drive for a week. We have the data. So collecting data now is the easiest part and by understanding this first principle because they will never change. If you are doing this without any external education or like external um, intrinsic or extrinsic kind of like uh, the value. So without doing any extrinsic uh, stuff, you will never change your behavior. You are going to be a bad driver. So in markets that we don't have no data, we can still have some starting points, whether it's telematics, Whether it's any IoT devices, whether it's just the generic uh, pension related stuff, uh, we can use satellite imagery now, uh, m much efficiently. So data. I'm also saying this to people like you have always hung up on the wrong aspect. Data is, you can collect data. Just, just look outside. It's everywhere is data and you can collect it. So pretty much data is not the point is how you use that data, how you understand it or the connect the dots. So there's always like a starting point. But now the starting point is, doesn't take, doesn't need you to take a few years, you can just make it in two weeks.

Speaker A: What happens to actuaries then? In other words, if you're building out this business that's going to get really large, right. It's going to fill some of this protection gap or close some of this protection gap. Obviously for more sophisticated policy underwriting you'll have actuaries doing this stuff. But I presume if you're underwriting policies and building new products, some human at some level is going to have to check just to make sure that everything's okay.

Speaker B: Yeah, yeah, exactly. Um, so that's a tough, tough question, honestly. Because one of my realizations last year was um, again as an actuary, I was asking the actuarial question to the machine and the responses was like shoot, this used to take three months of my time and now the machine is just doing it in five minutes. The worst two hours. Okay, let's give it a more, more maybe some, some sophistication. Five hours. It's definitely not three months. And that's the moment I realized, well I'm going to lose my job. Someone is just going to replicate me. That, that's pretty much like how I started, uh, also learning uh, coding as well, uh, and understanding more the AI. But yes, a lot of, a lot of things will change, but one thing will never change is the knowledge and the creativity part, so I have this saying is like, you can replicate knowledge. That's pretty much what all these companies are doing. But the creativity aspect is 100% random. It's not binary. It's just like in order to create something new, something fundamental, you need to create a lot. One thing is like back in the days, they ask Michelangelo, uh, is like, how did you create this? This? You're famous. I just chipped away all the storms that do not look like David. So it's pretty much like, like that. So you need to try a lot of things. So one thing absolutely ultimately makes sense. So I helped that in terms of these professions like knowledge have a professions, we are going to need more knowledgeable people. So data handling, the modeling part, like everything is done by machine, but that creation process and the review process, we still need human. So someone can make like a decision.

Speaker A: I want to ask you this as well, and only because it's top of mind for me.

Speaker B: Yes, please.

Speaker A: I don't know if it was last night or the night before, but my business partner and I were on a call with each other. We're just like catching up. And we were just having a conversation about this thing or that thing or some other thing. And then something popped onto my screen and we talked about that a little bit. And then the two of us had this kind of serendipitous creativity. You know what I mean? We're like, these three or four things came together that didn't seem like they were related to each other. And we were like, wait a second. I think we've just come up with a new product. We didn't even intend during that conversation to come up with a new product. And that's really serendipitous. Right? You know what serendipity is? Yeah, yeah. So how do you can, can a machine do that today?

Speaker B: Right?

Speaker A: Can agentic AI do that today? Like be involved in a com, in a two way or a three way conversation and then come up with a creative idea and make a decision about it in the same way that my business partner and I did last night or the night before? Or does that also require humans? Because you're right, you can get all the data, you can get the knowledge that already exists. That's easy. That's super easy.

Speaker B: Right?

Speaker A: But the knowledge that doesn't exist already. Can you create new things, you think in a way that is serendipitous, which is the way most innovation happens anyway.

Speaker B: Yeah, yeah, yeah, exactly. Um, we have this terminology in actuarial sciences like actual intuition. And it's also a valid thing in like when you're talking to the regulators, you can kind of like get around like that was my intuition. Same same thing. So yes, machine can do that. But machine will decide on these type because pretty much like we have all the agent, agent AIs, they form like a workflow. So pretty much it can run itself right now. But the main problem is at the time it makes that decision. Decision. It's the best decision it can come up.

Speaker A: At that time though.

Speaker B: Yeah, at that time. And that decision is the machine's decisions. So it made the decision because it thought this was the perfect idea. But the decision could be like, let's build a gun and start killing people. Well that's also a bad idea. It was a bad idea, but it's still a decision point. So we still need people with intuition. So that what that means as an end result, uh, we will never able to replace that. Yeah, because it's just like when you think about the knowledge or information creation, which is another, another science growing lately with after this European uh, mind project, like a lot of findings from that European mind project was like how we create information, how intelligence is being formed. And pretty much we realize that any machines, any, any form that can process information can. Intel can be intelligent. But what it means for the next phase is, is always like the creativity and the intuition part. Uh, it's just uh, it's just a lot of options out there and we always pick the right option that feels right at the time. So pretty much what happened to you is that you had millions of observations throughout your lives, but at that moment it just made sense.

Speaker A: Yeah, just an interesting concept to think about. Can we do this too? We spent a lot of time talking about process automation, automating a whole bunch of things using agentic AI. But I interrupted you kind of at the beginning of this conversation when you started talking about reinsurance.

Speaker B: Oh yeah.

Speaker A: What is the implications? Let's spend a few minutes talking about this too. What, what are the implications here for the stuff that allegory is building in the reinsurance market?

Speaker B: Yeah. Um, so in the same, in the same sense, uh, like the way we communicate people is just like getting much faster and faster. So this pretty much like an exponential increase. And now especially on the insurance and reinsurance aspect, it's more like maybe in the past an insurance company could sustain its financials by just insuring uh, a certain location, so renewing that portfolio because the risk is, especially in life insurance, there's this Phenomenal say in a given society the rate of cancer will never change because there are some societal aspects that drives the, your exposure to the cancer. So this is the phenomena in life insurance. That's why for life insurance companies it makes them more profitable the more people they underwrite. Because if you have like it's the same thing. Uh, but in PNC it doesn't work like that. So the more people, it doesn't mean like you're gonna make more, more profit. But in life you can, you can access this. So in the same sense now the risk is. Because the idea back then was like risk is more localized. So we can, we can just manage this portfolio. But now we understand that risk is not localized. There is like a whole atmosphere on top of us for the past billions of years and that actually changes things. And now we are also manipulating the dynamics of this millions of years of uh, the formation. So we don't know what is going to impact and how it's going to impact. So from a risk standpoint and especially on the reinsurance side, we really need to understand more on the mass changes on the environment because the reinsurance is not like uh, a 12 year business. It's more like uh, the long term decisions on the capital markets. So based on the current knowledge we really can't predict this because the hurricanes are getting tougher, the other natural disasters are getting stronger. The earthquake. Now we are understanding more about earthquakes because the periods of earthquakes, for example Istanbul, the in Istanbul location, the period of earthquake is one in 250 years. It's like five, six generations. But it has been happening for the past 13 million years. And what about the fault lines that we have never ever. Like for example in America maybe there is a fault line that happens every 1,000 a year. So we just don't know that. Like because we were not here a thousand years ago or we didn't have uh, said it wrong, we didn't have the technology to understand that or seismic changes. So these are the things now the AI can help us to understand better. And um, my guess is probably in 10 years the reinsurers will become more like the current insurers. So they will start underwriting much more risks in terms of like the collections, collections of risk rather than the individual ones. Because otherwise in terms of monetary, we won't be able to sustain this. There are also other dynamics we need to understand. Specifically the pension part or the aging population. We live longer, that's good. But who is going to pay the pension? That's Also another thing. So everything in the financial markets is pretty much connected to each other including and um, reinsurance is pretty much the ultimate one in the insurance uh, industry. So this is pretty much like what I expect to happen in the 10 years. Like the reinsurance will become more like the insurance and there will be of course consolidation.

Speaker A: So where else can. And maybe I'll let you go after this, although the conversation is super interesting to me. But where else can Allegory's technology be used that's not. Or that's either tangential or outside of the insurance space. Right. Because you've built something that sounds like it's pretty powerful. Can it be adapted to like stock trading or to other types of finance?

Speaker B: Yes. Uh, I think one of the power we have in the platform is we pretty much captured all the capturable data points. Uh like again starting point was like if there is a risk, land and water and then it stems from there literally millions of different simulations. So in that sense we can definitely apply this to stock markets which is on the plan. Uh, another outside of uh, the platform is like uh, probably you are familiar with the tools like Cursor, the software development tools and they are so uh, pretty much I started this whole AI project at the same time the cursor. They did. I already had my internal cursor before they had the investment.

Speaker A: They released it. Yeah. Just so you know, we've actually built our own AI tools as well.

Speaker B: Yeah.

Speaker A: Yeah. It's interesting. We haven't rolled them out to anybody else except ourselves, but we built an entire suite of AI tools. Sorry, go ahead.

Speaker B: Exactly. So, uh, pretty much like yes, there are tools like Cursor or the Windsurf to help accelerate. But I have been doing it since the inception of this AI project because otherwise it's impossible. Uh also that really helped me to create a much more cleaner code base. Aside from the data aspect, that's also another thing. Eventually someone needs to read that code. They need to understand this is another uh, angle that we can create much cleaner code. Uh, of course we are not going to uh, open AI chat and just having a chat. So we are way beyond that. Uh, by doing that you can create a software that works on your local machine. But if you want to make it like a cloud based infrastructure, security stuff, that's a whole different uh, profession. Uh but in a tangential way. That's the way how we can create software like even thousand times faster and better reinsurance. It took us 15 minutes to deal the MVP reinsurance. It took me it took me two days to review the code but it took machine to build it in 15 minutes and I didn't make any change. Uh the end result was I think it generated like 120,000 lines of code. Just code, no packages, nothing. It's just pure JavaScript, Python, CSS, Terraform and Shell scripts. That's 15 minutes.

Speaker A: That's amazing. So what does your customer base look like? In other words, if you had an ideal customer in North America, who would it be? And has allegory been out there long enough to actually have customers that are using the technology?

Speaker B: Uh so we pretty much launched this uh, the AI code engine earlier this year. Uh we also have uh, we are making it really clear like we do not have the license right now but people still, so we have an established SEO so people still find us. But uh, we started this let's say February 1st the AI code engine and up until now we have issued like close to 700 codes just fully AI generated, uh, all AI managed. And from that angle I can confidently that's like a one less than one person operational expense. So generating 700 codes roughly equates like 3 million dollar GWP had we had license but 3 million GWP at uh 1% operational expense. I personally have not ever seen. Have you ever seen something like that?

Speaker A: No. That's why I'm asking. No.

Speaker B: So that's one of the aspects like we are still learning because again we keep seeing people like selling this for $0.01 so we still need to learn a little bit. That's why it's running the other aspect right now we are actually uh working with a few clients uh on the enterprise side. Two of them are in Turkey. Uh there will be two in US as well. So what they are pretty much doing is we have 50 AI agents. Uh and sometimes people ask why do you, so why do you need so many AI agents? Because not everyone knows everything. So some people just know their area quite well. But I mean document generation could be like outside of their stuff. So they can use AI agent specifically for document generation so they don't need to write now create me a PDF look like this, look like that. So it's time consuming. So we have like templates. One company is uh, for example what they are considering right now is okay we are going to get this 10 AI agents. Interestingly one of them is AI actuary. Uh the IT team wants to use the uh, AI technology officer. Uh also the data science team wants to have our data analyst and code analyst tools. So pretty Much. The starting point usually for the company is let's get this, 10 AI agents for 10 people. And it's just like a monthly electricity bill for us. So it's just a productivity suit at that point. They don't need to make any integration at all. The data is theirs. We never ever sell the data. That's number one. Uh, we never do that. We never train data on customer input. Uh, and I can confidently say it's one of the most secure system we have ever built in the infrastructure side. So even some, even I can't see anything. Like everything is just encrypted or it's just like in the back end side or it's just like random gibberish. So there are a lot of things that uh, even sending data, receiving data, most of the time we just don't see anything at all. But the machine has the algorithm to encrypt and decrypt without anyone seeing it. Another use case is like we have this m, uh suite of APIs. They uh, can easily integrate one aspect of their insurance business. So uh, for example, we have this AI agent they can put on their website and they can just use it as an SDK solution. Uh, it's really like uh, I know all the problems in the insurance industry, so I started from there. What didn't work in the past was my starting point. Uh, and that starting point comes from like I build an insurance company, only insurance in the past. There are a few other companies I was involved like in the building phase. So I have this unique experience. I usually tell people that I had the opportunity to see what didn't go well. And in my opinion that's, that's the best knowledge and you uh, can ever have. Because, you know, you know the, what the adverse effect and everyone knows what needs to be done. But the main thing is what you shouldn't do.

Speaker A: I got it. Okay, look, I'm going to let you go. Ona Gungor, the founder and CEO at Allegory. This has been really fascinating for me. I hope you enjoyed this as much as I did. Thank you so much for doing this.

Speaker B: Yeah, amazing. Thank you so much.