Kyle Geoghan

The through-line

Everything Kyle Geoghan argues comes back to a single complaint: insurance is hard to get at. Either you are "interacting with a legacy system with limited technological capabilities or you have to talk to a human and that's it. there's really no here's the perfect way to get everything I need in terms of insurance information" [2]. He started Indemn in April 2021 on the thesis that "web forms are terrible and nobody likes them and there's a lack of access to information and insurance products themselves and AI can help" [1], and he describes the product consistently in that frame: a conversational insurance platform, or in plainer terms, "AI agents that you can call, email or chat with and uh they answer most frequently asked questions and access all the essential workflows" [1]. He positions it as "a new core system, sits in front of all of the legacy core systems" for MGAs, carriers, wholesale brokerages and their customers and brokers [1][2].

What has shifted is not the thesis but the vehicle. Indemn began by proving the idea on itself, selling Markel-backed wedding and event insurance through its own delegated-authority MGA, EventGuard, and has since sold that subsidiary to Jewelers Mutual to concentrate on the software platform [1][2][5]. Geoghan frames this as a maturation rather than a retreat: the lab did its job, the software customers arrived, and running the MGA became opportunity cost [1].

On why access, not price, is the bottleneck

Geoghan's ambition for AI in distribution is measured in volume, not efficiency. He points to equity markets as the analogy: there are "50 times more stock purchases every day today because the internet and some apps exist," and the charts and price-to-earnings data on those apps "make you feel comfortable to press buy now" [2]. Insurance will not see that multiple, but he thinks "we could 3x the total premium written globally if it's just easier to access and trust the products themselves. If you don't have to wait to get answers to your questions, if you want to look at a different option for a quote and you just ask and you get that's a very different interactive way to access insurance" [2]. The obstacle is that insurance is genuinely complicated, "which is why just an app in the internet wasn't enough to really make a difference" [2]. This is a continuation of the mission he joined CoverWallet for, and of something older than that: "since I was a kid, I've been trying to make it easier for my customers to understand and buy their insurance" [2].

The same logic explains which products he thinks are ripest. Simple lines where "if an insurance agent's working on them at all, they're already losing money" are the natural first targets, he says, naming umbrella, renters, pet and travel: "there's a lot of these products where an agent would love to provide the best option and all the education around it. Uh but they just can't do that and keep the business running" [1].

On needing equal parts insurance and technology

Asked the old insurtech 1.0 question of whether Indemn is an insurance provider or a technology company, his answer is flatly "We're both. You need equal parts" [2]. He is critical of the first wave on exactly this ground, describing CoverWallet as having an exceptional technology team but also being characteristic of an era that would "worry more about technology than insurance itself with its own challenges" [2]. His own background runs the other way: third-generation insurance family, filing surplus lines taxes in his father's Long Island agency at twelve because he was the only one who could navigate each state's janky website, an independent agent in his twenties before joining CoverWallet in 2017 [2][4].

The practical argument is that domain knowledge is what lets you guess well. Natural-language quoting and real-time underwriting carry "all kinds of compliance risk" and "all kinds of distribution challenges" across brokers, embedded platforms, direct and affinity, and "if you don't have insurance experience, you're not going to build that platform to scale the way the insurance companies need it to" [2]. He is unsentimental about how much of this is trial and error: "you really do have to create the future. You have to have a perspective on what it's going to be and try it and half the time or more it doesn't work. So then you try the next thing and you just have better guesses and more attempts if you have more information on the industry" [2]. He is correspondingly impatient with the handwaving, noting "a lot of people saying, 'Hey, I can do XYZ'" [2], a scepticism the hosts extend to firms founded in 2018 and 2019 now claiming to be AI native [1].

On the MGA as a lab

EventGuard was never a separate business in his telling; it was the test rig. "The MGA is fully built on the Indemn AI agents platform and it served as our lab. So whenever we wanted to test a new model or a new workflow or a new distribution channel uh we would start with event guard" [1]. Embedded relationships, affinity partnerships, changing the context when distributing through agents: each primitive was built on the platform and then proven "using the real world" [1]. The original plan was faster and more naive, he admits: sign with Markel in April 2021, expect three months, replicate across products important to millennials, "and we'd be an MGA for millennials." It took a year and a half of learning, and for most of that stretch "people said, 'I don't care. I don't like chat bots. This is not the future'" [2]. The breakthrough came with a human in the loop and a better reinforcement learning training model, just as ChatGPT launched and the market arrived on its own [2].

The decision to sell was about focus rather than performance. Roughly 95% of the customer experience at EventGuard is AI agent driven, it grew mostly through referrals, and it hit a million dollar run rate premium in March, at which point Mitchell Busy, the underwriter they had recruited from Markel, asked for investment and headcount to take it to six million [1][2][5]. Geoghan's read was that the software business was growing too fast to justify the distraction: "Mitchell, this is too much opportunity cost here. What we're going to do is look for a strategic partner that will invest in it and is also potentially a customer" [2]. A competitive process followed, and Jewelers Mutual was, in his words, "the best possible partner" [1]. The broader lesson he draws for others is that you do not need a war chest: "there are a lot of ways to get started with a lower investment and upfront time and cost and we did that" [1].

On what the AI agents actually do

He is specific about the current capability, and honest about its limits. The starting point is deflection of repetition: "right now we can replace 30 to 40% of the repetitive questions that a underwriter or CSR is fielding every day," achieved by tracking what people ask and what humans answer and handing that to the agent, "and now you don't have to answer that same question 20 times" [2]. From there the agents move into workflow: asking the questions needed to hit an API and return quote indications, walking step by step through an application and validating underwriting answers, generating certificates, and handling FNOL [2]. Broker service is a live use case, with the agent calling policy administration systems in real time so it can report that "the customer has a policy enforce, but they've got a bill of X dollars and an endorsement request and process," or tell a broker "the policy is enforced, but you owe $300. Here's a link. Click it and pay" [1][2]. He frames all of this modestly as "really just access to information, the primary data sitting in those systems" [1].

The destination is self-service: "the ultimate goal for me and my company is to let my customers build and manage their own AI agents through our technology platform. And so we're about halfway there. Our customers can do it, but we're still early in terms of capabilities, knowledge, and we're changing the platform every day" [2]. Where he sees the commercial value for specialty insurers, MGAs and wholesale brokerages is at the top of the funnel, "particularly upfront with engagement and education. So, you can get more leads, uh, right away. Uh, you can get answers to new brokers that you sign up" [1]. Deeper integration with core systems is the stated 2026 priority, including partnering with "great core systems like Loro" so that an AI agent sitting in the corner of a policy administration platform can answer the most frequently asked questions, act as a feedback mechanism, and "deliver easier natural language access to all the pieces of the insurance workflow puzzle people work with every day" [1].

On advice for new MGA founders

His counsel is to start immediately and start small. "You don't have to wait to get started. Um and there's a lot of lowhanging fruit across the business" [1]. He offers his own habits as the baseline: he no longer types emails, he dictates by voice and asks a model what to do next, and research has become trivial [1]. The first strategic decision, in his view, is distribution: pick the channel you will target, and if you are good at it you pick the right primary opener for that product, whether that is agents, embedded channels or affinity, which "really depends on the product" [1]. On build decisions he pushes founders to at least ask the question: yes, you can hire someone to build a web form, "but now you've got to also explore, well, what does it look like to do that with AI? And we're still very early days there, but you have to start asking those questions. And as soon as you do, you start to see what the technology can do" [1]. He applies the same test to vendors and teams: "if you're working with any technical teams and they're not using the best tools and showing that they can be much more productive with AI, then you're already behind kind of what the cutting edge is and what the potential is" [1].

Taken together, this is why he is bullish on new entrants. AI will let new MGAs launch at lower cost and reach distribution channels more easily than before, alongside a growing ecosystem of capital, product development and underwriting support, so that "we're actually entering into a new renaissance for insurance" [1].

On the shape of the technology shift

Geoghan reaches for infrastructure analogies rather than product ones. Generative AI is "a clear technology paradigm shift for the whole world and every industry," comparable to early electricity generation in the 1900s or, more usefully, the internet: "What we have are pockets of intelligence you can train for your business or your personal activities and go back to it and improve it. And you know that's a real different type of technology interaction for humans" [2]. He extends the internet parallel to adoption curves, arguing that just as businesses went from one website to a handful of SaaS subscriptions to hundreds of tools, either your existing tool gets better by implementing generative AI or a new tool arrives built on it, and whole industries will exist that could not have before, the way Google, Facebook and Amazon could not exist without the internet [2].

He also flags where insurance sits in the queue. His read for the coming year is that "more industries outside of insurance are going to have more salesbased AI support and I think that will bleed into insurance might take a little bit longer" [1]. Tooling gains compound the effect, and he cites running ten terminals at once in Claude Code as "another order of magnitude improvement in coding and development" [1]. He is aware of sounding like a broken record and says so, but holds the line on the magnitude: "if you can get into the deeper conversations about where it'll make a difference and what the world might look like in three, five, and 10 years, you probably cannot overstate how the world will change. There's a lot of ways we can predict the world is going to change with this technology and I think there are many more ways we cannot" [1]. That said, he is clear that AI has to be pointed at a defined problem and a defined value rather than applied generically [2].

Takeaways

• The core problem is access, not price: today you either fight a legacy system or wait for a human, and Geoghan argues easier access and trust could triple global written premium [2].
• Indemn positions itself as "a new core system, sits in front of all of the legacy core systems," delivering AI agents reachable by call, email or chat for MGAs, specialty carriers and wholesale brokers [1].
• Realistic near-term impact is deflection plus workflow: replacing "30 to 40% of the repetitive questions that a underwriter or CSR is fielding every day," then quoting, underwriting validation, certificates, FNOL and live policy status lookups for brokers [1][2].
• Insurtech needs domain depth: asked whether Indemn is a technology or insurance company, his answer is "We're both. You need equal parts," and without insurance experience "you're not going to build that platform to scale" [2].
• EventGuard was run as a lab on Indemn's own platform, reached roughly 95% AI-driven customer interactions and a million dollar run rate premium, and was sold to Jewelers Mutual because the software business made it opportunity cost [1][2][5].
• Simple lines where agents already lose money, umbrella, renters, pet and travel, are the natural first targets for conversational distribution [1].
• Advice to new MGA founders: start now with low-hanging fruit, choose your first distribution channel deliberately, and treat a technical team not using the best AI tools as a signal you are already behind [1].