Turning AI Agents Into Revenue Workflows With Outreach

Speaker A: What would speciality insurance look like if underwriters, carriers, and capital providers could all work from the same timely data? Well, my guest today is the CEO and co founder of Accelerant. We're going to talk about the risk exchange his team built to connect independent underwriting specialists with insurance capacity and risk capital. Now, yes, we all have an, uh, insurance policy of some description. I suspect very few of us, I include myself in this, know much about the industry and how it works. So Jeff will explain today why the traditional chain can be slow, expensive, and starved of useful information. But with the final risk bearer historically, uh, receiving only a fraction of the data collected at, uh, the start, something's got to change. And, um, thanks to AI and technology, everything is changing. So today we will discuss how, uh, portfolio diversification can improve efficiency, why smaller underwriting teams might actually have an advantage over established firms, and where agentic AI is already helping Accelerant organize incoming data and spot problem areas. And we'll also discuss who remains accountable when technology starts supporting an underwriting decision at scale and at machine speed. So, in short, we've got a lot to talk about. So let me introduce you to my guest right now. So thank you for joining me on the podcast today. Can you tell everyone listening a little about who you are and what you do?

Speaker B: Well, first of all, thank you, Neal, for having me on the podcast. So, I'm Jeff Radke. I serve as the, uh, CEO of Accelerant. I was among the group of folks that started it in 2018. Before accelerant, my career was reinsurance broking, reinsurance underwriting, and then the last sort of chapter was specialty insurance. And Accelerant is a continuation of that specialty insurance. So, like most insurance people, I've hung around New York, Bermuda and London is sort of the. If you think about the last 35, 38 years, that's where I've been.

Speaker A: Love it. And for people listening and hearing about Accelerant for the first time, you describe it as infrastructure for speciality insurance. That's incredibly cool. But what problems in that relationship between underwriting teams and risk capital? What was it that convinced you that maybe the existing model back then was in need of a rebuild?

Speaker B: As I said, I was among the group that started accelerant in 2018, and I describe it as a group of angry women and men. There were about six or maybe eight of us, and we had had decades in the specialty insurance marketplace. And what we knew was the process. You can call it a value chain if you want to be fancy, but the process of getting an insured policy was as torturous as you could possibly imagine. It was expensive, it was slow, and the data flows were anemic. It's the best I can do without swearing. Uh, you know, at one point we did analysis when we started accelerant. And the ultimate risk bearer, usually the reinsure, was capturing about 7% of the data that was collected on the front end. And we just knew that there was a better way to do it. So what we set out to create at accelerant, the infrastructure or the rails on, um, which specialty insurance runs. And what we wanted to do is we wanted to have a shared platform that everyone could use so you weren't reliant on your IT department. Right. A shared platform with complete transparency. You get thrown out of the insurance club if you vote for complete transparency. We don't want a data edge. We want the whole ecosystem to have the best data available. And we believe that having a, uh, complete bright, light transparency, that high degree of trust, increases speed and efficiency so much that that's the, our name of the game. Conscious of the fact that most of your listeners probably successfully avoided the insurance industry, maybe a little bit about accelerator, what we do on the supply side is we back inside. Individual insurance underwriters. Think of them as stock pickers, except they're insurance risk pickers. And they tend to be really specialized. One of our members does pickleball courts and another member does brownstones in New York City. And that's all they do. Right. So they're very, very niche. And what they need is they need two things for sure. They need the regulatory capability to issue an insurance policy. That means they need to be hooked up to an insurance company that has the ability to do that. And they need the capital to support that policy. Cause they're no longer an employee of an insurance company. That's one of the big changes. They're out on their own, they're independent. They've hung out their own shingle as an insurance picker. Right. An underwriter. And what we've done is we've tried to speed that up and make it much more transparent and much more efficient. Uh, I'll pause there because I'm in danger of going into the weeds. So Neil, uh, tell me, how did I do? Was that clear?

Speaker A: You did absolutely brilliant there. Because I think every single person listening will have an insurance policy or multiple insurance policies of some description. But very few will understand what goes on under the hood, how it all works. I think a great job of explaining it all there. And I'm curious, digging a little bit deeper in the weeds, how does managing general agent connect specialist underwriting expertise with capital? And where do delays, conflicting incentives or poor information, where do all these things commonly appear? What do you say there?

Speaker B: Um, I'm trying to get across the feeling as opposed to the detail. But before accelerant the typical value chain went as follows. Retail broker to wholesale broker, wholesale broker to mga, MGA to MGA or binder broker to a syndicate or an insurance company to a reinsurance broker to a reinsurer. Then the reinsurance company generally would have an investment banker who would go to the capital markets. That whole thing we would manage, we, the industry would manage to spend about 40 cents, maybe more on the small business that we do, at least 40 cents of the premium dollar or 40p of the pound on expenses, on overhead. Right. So uh, that's the old way. Now from an accelerants perspective what we have is we have that retail broker sometimes going to the wholesale broker but not so. And then they come to what we call the risk exchange. And if you want to think about the risk exchange as a little bit of a fancy manifold, right. So uh, uh, a risk comes in the front of the risk exchange and we route it obviously in an automated way. We route it to the right insurance company where that insurance company is already plumbed into the risk exchange. So a policy gets issued and that risk gets discharged distributed to a group of risk capital providers in a very efficient automated way.

Speaker A: Now if we go back to the early days of insuretech as it was called it once attracted considerable investment but often struggled to meet expectations. So looking back at those times, what did that first generation of companies misunderstand and fast forward to present day, what separates sustainable insurance technology from that attractive technology story?

Speaker B: Maybe one of our big partners is the Hippo organization and they're a really valuable great partner on the risk exchange. And their CEO Rick would describe it as, I think I'm quoting him reasonably well, is the first round of insurtech was focused on distribution almost exclusively and they tended to perhaps not have uh, the right people around the table worrying about the cost of goods sold or the expected losses that would come with those policies that were being sold. So I think that was the issue. Sometimes though I think people lose sight of the fact that we should all doff our caps, I think to those early insuretechs because what they demonstrated without a doubt is that we didn't have to do it the same way we've been doing it for 300 years.

Speaker A: Right?

Speaker B: The industry could change. And if you look at Hippo, for example, they're very successful now. Their loss ratios are very attractive and they continue to be very effective in using technology to sell. So, uh, I think that's what they got wrong. To generalize, that's what they got wrong at the outset.

Speaker A: Incredibly cool story as well. And now, of course, Accelerant operates this risk exchange that connects managing general agents with capital providers. So tell me a little bit more about how this model can improve alignment and transparency and how maybe concentrating activity on exchanges could introduce new dependencies or maybe even systemic risks there.

Speaker B: What did we get taught in economics class? The only free lunch is diversification. Accelerant's portfolio is incredibly diverse. Over 600 products, 22 countries, 300 and something. What that buys you is a relatively large, roughly $5 billion of premium, a relatively large portfolio of low volatility, small specialty insurance policies. At that point, what the industry has struggled with is handling that kind of business efficiently. Remember back to that value chain where I talked real fast and there were so many hops. Well, imagine if everyone has to earn a little bit, right? You know, it's death of a thousand cuts, right? Tick, tick. As you go all the way through. What we do with the risk exchange at Accelerant is in again, that broker contacts one of the accelerant members and that's done. Right? It's all prearranged as a portfolio. So instead of having program by program or product by product solutions, we have a portfolio solution which is across all 5 billion incredibly fast in relative terms, incredibly efficient in relative terms. So that's how the risk exchange works. That's the magic or the secret sauce, right, is you just, you're solving it on a portfolio level as opposed to a product level. Now, what are the systemic risks potentially that we're introducing? There's a single point of data failure which is Accelerant operating the risk exchange. If the platform, uh, it's kind of hard to imagine, but it's a useful exercise if the platform were somehow to stop being able or being willing to provide that data in a transparent way, the way we do across everyone involved, the whole ecosystem is relying on Accelerant and the risk exchange to publish that on the platform. Whereas if you break it down so it's product by product by product, it's incredibly inefficient. But I guess it's more resilient, right, because then you've got 600 paths as opposed to, uh, sort of one big path. The way we address that with our risk capital partners is we are just very, very clear about our processes and how we ensure that data flow and cash flow and all that other stuff that follows occurs. So it's a little bit like saying, geez, what would happen if NASDAQ stopped working? Well, it'd be a problem. It'd be a big, big problem. But the participants feel like all the incentives are aligned for that to be very, very unlikely.

Speaker A: Yeah. And before you join me on the podcast today, I was also reading that you've argued that data transparency is becoming a prerequisite for accessing high quality capital. So with that in mind, what information do capital providers now expect? And how can insurers improve transparency without exposing commercially sens? I would imagine it's somewhat of a balancing act there. But what are you seeing?

Speaker B: So I'll take a step back to make sure that the answer is as useful to non insurance people as possible. If you imagine that your job is evaluating, uh, a particular insurance risk, a hardware store or some other kind of shop, what do you care about? You care about where the building's located, how the building is constructed, what the building's worth, what kind of roof does it have? Is it in the path of a hurricane or some other sort of potential natural disaster? From a liability perspective, you care about what products they sell, how much do they sell, you care about what their store is like, how many square meters or square feet of display space do they have, and what sort of their foot traffic? And are they well regarded in, uh, their community? Because if they're well regarded, that tends to be a better liability risk than if they're poorly regarded. There's this host of things, and together, taken together, all that stuff we call those exposure characteristics, and they're little clues about how likely or unlikely a risk is to have a claim.

Speaker A: Right?

Speaker B: Uh, and historically in the insurance industry, it's interesting, is a little bit of a victim of being a first mover. In the 70s, they were the industry that moved actually much faster than investments into automation. And they bought these big giant mainframes and they built their businesses around something called the policy administration system. And a policy administration system, its job is to, well, you can guess, right, administer these policies. And, uh, when storage was very, very expensive, very, very dear, they had to make a decision about how many of these exposure attributes or exposure data fields they could store. And they stored way too few. They stored way too few because, as, uh, I said, it was expensive, et cetera. So on average, I think there's maybe eight to 12 exposure characteristics that are captured by the policy administration system we're capturing. And we cheated. Right? We started in 2018, so we already made all those mistakes. So I'm not saying we're geniuses. I'm just saying we started knowing the problem. I think the number now is we capture on average, over 60 exposure characteristics about each policy. Right. Um, and providing that to everyone involved in the transaction is really, really important. Just like in the mortgage market, providing all the details of the mortgages that underlie that bond, that collateralized mortgage obligations, does everyone look at the detail? No. When something goes wrong, does everyone look at the detail? Yes. And the fact of the matter is having that transparency, having that data availability allows people, and increasingly AI agents to do a much, much better job at selecting risk and then describing to the whole chain why we selected the risks we have.

Speaker A: The leading issue of agentic AI in businesses right now is ensuring agents act with compliance guidelines. And denodo applies guardrails across your entire data estate. By aligning your company's data infrastructure under one system, these guardrails perform consistently across your platform. So start scaling your business and start with Denoda. Simply visit danodo.com to learn more. And I was also reading before you joined me today that the divide between data capable firms and those relying on fragmented systems is also increasing. So can smaller underwriting teams close that gap? Or does the cost of modern technology risk only favoring the largest players? What are you seeing here?

Speaker B: Uh, funny enough, I think it's the other way, Neil. I think the smaller you are, as technology gets cheaper and cheaper, I think the smaller and newer you are, you have an unfair advantage, just like Accelerant did, of starting being a data centric company. Uh, the biggest hurdle is a mental one, always. Right? What's your mindset? Are you an insurance company? I would say if you start with that, you're going to fail. If instead you say, hey, we're a data repository, right? And we use that data, uh, to perform insurance services, I think you're much more likely to be successful. Smaller organizations, sometimes the budget is a constraint, but their mindsets are on average, right? So what we find is our MGA's managing general agents, our MGAs that we support. And as I said, there's over 300 of them now, but there's definitely a really solid cohort of experienced insurance professionals that understand that data and analytics is the future and they are excellent at it, even though they're small. What's really hard to do. And I had this job in a former life. What's really hard to do is to take a large organization that has built their processes around existing technology and to change those large organizations into more of a data centric operation. They're so process heavy that it's almost. Well, I had very limited success, which is code for none, Neil. I had very limited success in converting them from a, uh, workflow to a data flow mindset.

Speaker A: And I'm curious, from everything you're seeing and hearing, where is applied AI already producing measurable results in underwriting and claims? And how should maybe insurers test whether faster decisions are also fair, explainable and commercially sound? Of course, what you're seeing and hearing

Speaker B: here, and here's why our rating models are all deterministic, machine learning based. And what that means. I just said a bunch of jargon, right, Neil? What that means is if I submit the Neil Hughes Manufacturing Corp. If I submit it 100 times, you're going to get the exact same answer. For better or worse, you're going to get the exact same answer out of the accelerant risk models every single time. And if someone said, uh, well, why did you say no to the Neil Hughes Manufacturing Company? We would say, well, this is the things that they broke, right? It's very transparent inside the deterministic model. It's very clear. And if we accepted the risk and priced it, why did you price it the way you did? Well, here's why. It's one mile from the beach in Florida and that's quite a risky spot. And it has a of lot. Lots of the liability characteristics are high or low. What's the worst thing in the world? Right. Baby products. Right. As someone who produces products that children use and that's really risky as opposed to screws or some hardware that's much less risky. So I believe that the bit about AI and the rating models is much ado about nothing. I think the best companies, not just accelerant, I think the best companies are using, using deterministic models that can explain it. It's not like this secret thing where no one can, uh, really evaluate it. Where we are using AI.

Speaker A: Ah.

Speaker B: And where it's making an enormous difference is on the data intake, right? If you can imagine 600 different products, imagine all those different kinds of data. So we're bringing in Dutch motor policies at the same time we're bringing in plumbers in California. You know, how do you keep all that straight? The AI is keeping it straight. Agentic AI is keeping it straight for Us self healing, self organizing, makes all the difference in the world. Right. How else are we using it? With 600 products, there's always. You wish it was 5%, but it's usually almost 10% of those products that are on what we call the naughty step. You can tell where you. We started operations in the uk. A lot of our vocabulary is British. But the naughty step, meaning things aren't going exactly right. We find. We use AI to find the ones that aren't going exactly right much faster and much more effectively than we can with the human eye. Right. Because human eye has bias, human eye gets tired, human, all that stuff. Right. The portfolio construction is all driven by artificial intelligence. Again, agentic artificial intelligence. And then operations, the way it flows through the organization has been revolutionized. The team has revolutionized what we're able to do. So I guess what I would say is AI, agentic AI is making all the difference in the world for accelerant. And, uh, I have the perception that we're on the forefront. I think that's true again, because we're relatively small and relatively focused on, on achieving this. I guess it's not so surprising perhaps that we're more aggressively pursuing this. The one area where we're not pursuing it is in the risk selection and rating.

Speaker A: So many great points there. And we look at AI and obviously it can analyze far greater volumes of risk data than we've experienced in the past. And experienced underwriters, though, they will understand that context could be missing from a model. So how should firms divide that responsibility between technologies, technology and that human judgment which is crucial in understanding the context sometimes. And ultimately who remains accountable when the AI support decision proves wrong?

Speaker B: The underwriters, period. Sometimes I think we complicate things unnecessarily. But what did, what did my father say when I, when I was growing up? Right. He said, don't blame the wheelbarrow. Right. The, the problem is the person driving the wheelbarrow. Don't, don't blame AI. It's the, the problem. It's the responsibility of the person driving the AI. So who's ultimately responsible for underwriting outcomes? The underwriters inside the organization kind of period. I don't think that there's a simple, neat answer for what the blend looks like. I know we've changed a lot in just a year in terms of how much we rely on it. Increasingly are, uh, relying on the models. Rating again, for us is not AI driven. It's more machine learning driven. But we're having much more faith in those models than we did a year ago. We're having much more faith in the things that AI, the agents do than we did a year ago. But ultimately again, uh, it's Not Space Odyssey 2001 where it's some computer talking, it's Jeff Radke and his colleagues are responsible for the outcomes that happen at Accelerant. And, and we can use whatever kind of wheelbarrow, whether it's AI or machine learning or analog, we can use whatever kind of wheelbarrow we want, but we're responsible for the job.

Speaker A: Well, I've absolutely loved chatting with you today. Every day I try and get people thinking differently around the technology that is impacting our uh, world and our lives very often as well. And I think we all know insurance or claim to know insurance inside out. But getting that glimpse behind the curtain,

Speaker B: I'm 58 and uh, it's weird to me to say that I'm quite sure the next three years I'm going to see more change than I saw in the previous 40. Uh, I feel totally confident that's true. It's changing so fast and generally in a good way. I know that's an unpopular thing to say that generally it's a positive, but I think it is a positive if we're spending 40 cents of every single person's insurance dollar on expenses. Just think if we could bring that down, cut it in half. I don't know how many trillion dollars that'd be back in consumers pockets, but it'd be a lot, a lot of money.

Speaker A: Yeah. And there's so much to be positive around there. We're talking around AI augmenting underwriters rather than replacing them. Allowing better decisions at scale. Also, measurable improvements in underwriting, speed, risk selection, operational efficiency, claims outcomes. So much to be positive about. And for people listening that would like to dig a little bit deeper on um, anything we talk about, talked about today or find out more information about you and accelerant. Where would you like me to point them?

Speaker B: I think the place to start is Accelerant AI and then we're pretty accessible as a group. So by all means send us a message and we'd be happy to talk and can illuminate however we can.

Speaker A: Again, I've absolutely loved chatting with you. I will include links to everything that you mentioned there, encourage people listening to dig a little bit deeper, find out more. Information is a massive topic and there's so much going on in that space, so please check that out. But more than anything, Jeff, thank you for taking the time to come and speak with me today and talk about all this stuff. It's very complex world, but doing a language everyone can understand. Appreciate it.

Speaker B: Not at all. Thank you for the invite. I really enjoyed it. Have a great day.

Speaker A: I think Jeff's explanation today made a very complex market so much easier to understand. Specialty insurance depends on experienced people making very careful decisions. But those decisions become harder when information is delayed, fragmented, or trapped between multiple layers and intermediaries. So accelerant's answer seems to be simple. A, uh, shared risk exchange that gives underwriting teams and capital providers a very clear view of the exact same portfolio. And this technology matters. But Jeff's wheelbarrow analogy may be the line that stayed with me too, because the person using the tool will always remain responsible for the result. And, um, that principle, I think applies far beyond insurance, especially as more and more companies are adopting machine learning and AI. So a massive thank you to Jeff for explaining how data, capital and human judgment can work together. And remember, you can learn more about accelerant at Accelerant AI. But over to you. If you work in this industry, could better information help return more of your insurance dollar to the customers in the years ahead? Let me know techtalks network.com but we are out of time now, so I'll be back again tomorrow with another guest. But thanks for listening as always. Bye for now.