Building an AI-native MGA from scratch: inside Kinetic Insurance Services 

Robin: Welcome to this week’s InsTech Podcast. There’s a whole gang of us this week. I’m joined by Theo Butt, who’s the CEO of Kinetic Insurance Services; Andy Roberts, who’s the CTO of the same company; and Guillaume Bonnissent, who’s the CEO of Quotech. We’ve assembled this crew because we want to talk about Kinetic Insurance Services, launched in June by Convex. So this is the A team. Thank you all very much for joining me. 

Guillaume: Thank you for having us. 

Theo: Thank you very much for having us. 

Robin: Theo, I’m going to start with you, obviously, as the CEO. Convex launched an MGA business that sits alongside your underwriting business. It’s intrigued people a great deal. We are desperate to find out more about it. Tell us, what is Kinetic Insurance Services? [01:00] 

Theo: Thanks very much, Robin. Kinetic is something we’ve been planning for a while now, and it’s a core component of the next phase of Convex’s growth and the iteration of the business as we think about the next five years. Really what we wanted to do was find a way to increase our access to addressable market, stretch our risk appetite, and do that by utilising the strength of Convex, the carrier, and also by accessing third-party capacity in order to give us relevance and strength in the market. So that’s where the nascent part of the idea came from, and we spent some time thinking about the best way to do that, and this hybrid carrier-MGA model was what we landed on as our solution. We’ve spent the last year, really since July last year, putting the thinking together and putting the team together for the operational and technology components of that. It’s a fantastic opportunity for us to build again from the ground up, utilising the latest technology and agentic AI capabilities, as well as utilising a very forward-thinking, tech-focused platform in the shape of Quotech, which is really instrumental in enabling us to do the technology build that we want to do. [02:00] 

So it’s a platform for us to increase our access to lines of business that are adjacent and complementary to what we do in Convex, the carrier. We are not going to be competing with ourselves. We’re building a company that will form a strong part of Convex into the future — very closely aligned with, but slightly separate to. We’re very excited to be here to talk about it today. 

Robin: What kind of underwriter, what kind of entrepreneur are you hoping to attract into Kinetic? 

Theo: The advantage of an MGA with the support of a carrier is multiple. Firstly, we have access to capacity — that’s a big advantage when you’re setting up your own MGA. We hope to attract entrepreneurial underwriters who want to step into an MGA model but will also understand that we can provide them with access to capital much more easily, so they won’t have to spend as much time going out there looking for it. There’s also the infrastructure that we can build and are building around them. We can provide them with a cell to operate within and be entrepreneurial, but we’ll put the tech platform in, we’ll put the operational platform, the Quotech system, around them, as well as having the resources and capabilities of the carrier to call on from a claims perspective, compliance and all the other component [03:00] parts that we can bring to bear. So we want to attract people who are excited about building a new business for themselves alongside us as part of the Kinetic brand, which will stay part of Kinetic as we grow that collectively into the future.

Robin: Andy, that brings me neatly to you, because Theo’s pushed hard there on the tech benefits that Kinetic will provide to entrepreneurs. You’ve got the job of building this out. What is it you’re going to provide your entrepreneurial underwriters as part of the tech infrastructure they’ll get when they come on board? [04:00] 

Andy: Thanks, Robin. We have to stand up the core operation, and that’s what we’ve been doing over the last nine months — with a number of core systems, of which Quotech is one, which I’ll come to. But we’re also doing some really exciting build elements around it. Our vision is to eventually move to a world where almost all of our non-revenue-generating work is automated, and to bring insights to underwriters so they can generate more premium. How are we doing that? We’ve partnered closely with AWS and their systems integrator, Provectus, to help us build out AWS-native infrastructure. What we’re trying to build is an agentic-native factory — that’s effectively our core processes, all the way from the first email coming through, to generating our quotes, with the underwriter very much in control throughout. This is an opportunity we saw as agentic technology became available to us as we were starting up. It’s also a greenfield world we can play with. So we’ve partnered with the latest tech to build out an innovation platform that we can start deploying — not just the core processes, but what we’ll also call “superchargers”: all those repeatable tasks, the small data enrichments, where underwriters can say “this is what we want,” and we’ll have a team that can deliver it to them as quickly as possible. 

Robin: So the core components, if I’ve got that right, are an agentic workflow where you’re working with AI, which then links into a policy admin system. That’s the core of the [05:00] platform, is it? 

Andy: Yes, absolutely. That’s where Quotech comes in. We made a real effort to find something we could build with and around. We looked for something built truly with MGAs in mind — we were never going to build our own ledger, so we wanted to buy something with a super-strong ledger at the back end, and a policy admin system that covered everything from underwriting through to billing and collections, not just a shiny set of underwriting screens. That’s one of the main reasons we chose Quotech. But we also needed a really open platform — something API-enabled, moving towards MCP enablement — so we could build around that core as quickly as possible. 

Robin: Guillaume, you must be thrilled. This sounds like a dream gig for a man who’s been building platforms like this and preaching the case for new, modern platforms for a few years. How are you getting on? [06:00] 

Guillaume: It’s great. We’ve been working with Convex since about the start of Quotech — it’s always been a great partner for us. When they started telling us what they wanted to do with the MGA, at the time we didn’t know the name, so we called it “the Convex MGA.” We got really excited because we had perfect alignment with the vision we had of how a platform should be used and what it should enable. It reminded me of 20 years ago, when I was at Hiscox — we had a project called “Underwriters to Underwrite,” which never really worked, because the technology wasn’t there and we didn’t have the core system to be able to do it. So we spent a lot of time doing admin tasks. This was a really exciting project, and we’re very pleased to be involved in it, and to have won the RFP. 

Robin: But it’s not a standard gig, is it? It’s not like walking into a standalone MGA and selling your platform on an RFP. Here you’re going to have multiple underwriters, multiple cells, multiple requirements. Does that bring any particular challenges from your point of [07:00] view? 

Guillaume: Yes, it’s really interesting from that point of view, because at the moment we don’t actually know who the underwriters are going to be, or what kind of complexity they’ll bring. So we built our platform with that in mind, and since we started the project we’ve made it very flexible, so that when Convex hires a new team, the technology doesn’t become an issue. A lot of large MGAs, when they bring on a new team, need to look for capacity, and then need to configure the platform to work with that new team — generally that can take three to six months. We’ve worked to reduce that dramatically, to a matter of days or weeks. That’s been the really exciting challenge, and the team really took it on. It’s interesting because we don’t know if the products are going to be a super-complex reinsurance product or something as simple as golf all-in-one, but the platform can do both, and everything in between. 

Robin: Theo, I wouldn’t be doing my job if I didn’t challenge the model a little, because there are aspects to it that have intrigued everybody. The first challenge is this: is this a bit of a hybrid model? Won’t truly entrepreneurial MGA owners and underwriters be brave enough to go out, set up their own business, seek investors and claim more of the value they create? Are they going to be handicapped or compromised by the fact that you’re part carrier, part MGA platform provider? Is this for those looking for a safe option, or are you hoping to attract the truly entrepreneurial? [09:00] 

Theo: Hybrid is one way to describe it, and of course that is what we are. But as we’ve already described, that brings advantages in terms of capacity and other ancillary services. Kinetic is a distinct company as well — we’ve set it up as a separate company. It will have its own branding, adjacent but separate to Convex, and its own compensation arrangements in line with MGA compensation arrangements, so we’ll be able to incentivise people appropriately to build their profitable portfolios with us. On that basis, we should be able to compete with, if not better, any other MGA in the market. And it’s another — to use Andy’s phrase — “supercharger.” We can put people into a framework that lets them focus on the value-added work they do as an underwriter, rather than worrying about building systems, processes and back-end ledgers — all the things we’re traditionally not very good at as underwriters ourselves. That takes time, money, and can be expensive and not that effective. I think we’re genuinely offering the opposite: we’ll attract entrepreneurial people who can then get on with the business of building their business, rather than doing all the [10:00] work around and below it. It won’t suit everybody — some people will want to plough their own furrow entirely and take their chances. We’ve seen lots of people do that successfully, and lots of people do it less successfully. I think what we offer is a genuine differentiator, and that’s exciting. 

Robin: That’s fair enough. Are you in any way worried about potential clashes? You made a good point earlier about using this to extend your underwriting appetite into areas where you don’t necessarily have existing expertise. But what happens if you attract people where you do have expertise? Are you worried about the two businesses — Convex and Kinetic — competing with each other? 

Theo: One of the foundational early principles is that we’ll be complementary to, not competitive with, Convex. What’s been great is that Convex is a very nimble company, despite its relative size in the market — we’re still only 480 people, and I know every single underwriter next door in the Convex [11:00] building. We’re working hand in glove with those teams to develop opportunities they’d support, and where we find there’s a bit of crossover, we work through those areas together — I spent most of this morning doing that on a couple of potential lines of business we’re looking at. The nice thing is that everyone within the carrier is supporting this initiative. I stay on as a group executive member, so I’m very connected to what we’re doing across the group. Doug Howitt, the chief underwriting officer, Doug Howat and I sit down weekly to discuss potential opportunities and where they might best fit. So yes, there’s a risk, over time, that we tread on each other’s toes, but it’s one I think we’re well set up to mitigate given the way Convex is structured. 

Robin: You mentioned capacity at the start — you’ll provide capacity. Do you guarantee to provide it? Presumably, if you don’t like the look of an opportunity, you’ll turn it away. But in reverse, are people obliged to take your capacity, or can they come in [12:00] and not take it? 

Theo: From an alignment perspective with other third-party carriers, it’s very important that Convex is enabled to offer what we’re calling “cornerstone capacity” on every cell. Otherwise — if we set up a cell and say “we don’t really want to support that one” — why would anyone else? That’s one of the challenges of being so closely associated with a carrier, but it’s also the benefit, as we’ve discussed. So we’ll always offer cornerstone capacity, meaning the opportunities I bring in have to be structured in a way Convex is prepared to support, either with lead capacity or support capacity, depending on the line of business and our expertise to lead it. There may be the odd opportunity where we don’t provide capacity, where there’s already a 100% solution in place — if a cell comes along with its own capacity already set up, we’d consider that, subject to the capacity provider being happy to work with us. But the foundational principle is that Convex will generally be a cornerstone provider for every cell we underwrite. 

Robin: Thank [13:00] you. That’s the end of my official challenges to the business — I’m just asking the questions others, looking in from the outside, would want answered. It seems to me you might have a dream job: a greenfield site, the chance to tap AWS, to build something really exciting and fresh within a large, well-funded enterprise. That raises a question: with the very best of AWS’s agentic capabilities folded in, do you envisage situations where what you learn doing this platform might one day inform what Convex itself does? 

Andy: I think it’s a great question. There are elements of what we’re doing that Convex are already doing, and elements they aren’t — so there’s loads of learning to share across it. I think the thing we’re doing, building this truly from first principles, is looking at grown-up AI, which is the hard part: how do we build something for a regulated market that has clear explainability and auditability from day one, rather than bolting it on later once it’s built? So we’re designing and building the patterns for how we structure and manage context. How do we link in an email as soon as it comes through and route it to the right case? How do we automatically kick off the right agents? And, really importantly, always see who’s done what, when — an agent has done this, and why — and make sure there are serious controls over what it can and can’t do. We keep a human in the loop the whole way through; we’re in no way looking to replace the judgement of an underwriter. It’s about bringing them the insights and removing the operational work around it as consistently as possible. And then, of course, there’s the fun part later of how we change the skill set, so that people currently in operations become [15:00] more like agent managers — we need to build those skills. Our underwriters will be far more AI-native and tech-savvy than historically, and there’ll be lots of learnings, I think, on the operating model, that we’ll be able to share back to the mothership. 

Robin: Just be warned, we’ll be tapping you up for next year’s Agentic AI InsTech event, where you can come and share your learnings — I couldn’t miss the opportunity. Guillaume, how’s the rest of your business going? Convex and Kinetic aren’t your only customer — how’s business generally? 

Guillaume: Really good, actually — we have a lot more interest. I think AI has really triggered a lot of questioning in different boardrooms about the systems they have: how do they get rid of legacy systems, how do they future-proof their architecture? Interestingly, that leads to a lot of conversations where they realise that [16:00] a platform that’s API-first and has MCP servers ready is much better positioned to future-proof their architecture. Companies are also realising that the most important thing is the data — how do they record all the data they need, and how can they interrogate it? So the question becomes what architecture matters to support that, whether that’s a data lake or the right underwriting system. That’s where all our conversations at the moment are going. You should expect some really interesting announcements from us in the next 12 months, as well as the progress we’re making with Kinetic. 

Robin: I’m not going to let you leave that idea hanging — can you share a little of your thinking on what the next 12 months looks like? From those customer conversations, you must have some clear ideas about what you’ll need to invest in. Are there any themes to bring out [17:00] of that? 

Guillaume: I think it’s the integration with large language models — how do you let your underwriters interact with them? If you’d asked me that question a year ago, at our last podcast, I couldn’t have imagined users logging into our platform and seeing a prompt. Now, with the progress in large language models, and all the work companies like Anthropic have done with MCP servers and making them open source, I think the first thing that’s going to happen is that every user logs into the large language model interface given to them by their company. Why does it need to be given to them by their company? Because of the compliance angle — making sure they’re not in breach of any NDA they’ve signed, making sure the data isn’t being trained on, that it’s properly segregated, and so on. [18:00] 

Through the MCP server, they can then set up tasks. Say you sit down at your desk at 8am every morning — you want to see all your meetings for the day, all your tasks, every email that came in overnight. If you’re an underwriter, how many new enquiries came in and were automatically ingested into your platform? How many underwriting referrals do you have? How many queries from accounts? Then you go onto the different platforms. I think people are going to use things like Outlook less and less. Does that mean people will use our platform less? No, I don’t think so — but the way they access it will increasingly be through widgets on a large language model interface. Even I have to pinch myself thinking about the timeline — is it three months, six months? No, it’s actually doable today, and that’s [19:00] the extraordinary thing about the technology available now. Data ingestion isn’t a dream any more; bringing everything into your large language model isn’t a dream — it’s absolutely doable. So the question now is more one of adoption, and of legal departments working together to make sure they don’t say “no one can use a large language model,” but instead ask, “which one do we need to use, what package do we need?” And also controlling cost, because the last thing you want is a huge bill because every user is vibe-underwriting or vibe-coding. There need to be controls in place, but the problem at the moment is 100% adoption, not where the technology is. [20:00] 

Robin: That’s really interesting. When I said earlier I was jealous of Andy and the excitement of his job, all those controls make me less excited about it. Theo, over to you. Casting forward to 2027 — you’ve launched, it sounds really interesting, there’s a broad fascination and admiration for what you’ve done. Presumably now it’s about signing up some MGAs, some underwriters. What else is on the cards for 2027? 

Theo: It’s really about finishing the final components of the initial build — the day-one model. We’ll continue to build and iterate on that, and some of that will be part of what we do next year. From my perspective, it’s very much now about talent acquisition — finding the right teams to join us and bring complementary skill sets and underwriting appetites into Convex. We’ve got three cells we think we’ll have up and running in time for the 1 January renewals, with underwriters hired into those. Our first underwriter joined us in Bermuda at the beginning of August — that’s an emerging industry cell we’re setting up out of Bermuda, which should be up and underwriting in the next two or three months, dependent on regulatory approvals. We’ve got two further cells we’re working through, who haven’t joined us yet, so I won’t name any names, but we should have three or four cells up for 1 January. And we have ambition to add at least two or three more cells during 2027, leveraging the technology we’re building, and generating some underwriting income for Convex and our capital providers. So there’s a lot to do — we’ve got to find the right talent, find capacity to support that talent, and plug them into our systems and processes. 

Robin: Long term, is this the sort of business that Convex might spin off, or am I getting ahead of myself? [22:00] 

Theo: We have two MGAs, or MGUs, now. We built Convex US on the basis of an MGU platform, because that was the quickest, most efficient way to set up our operation on the ground in the US. We built that to be separable. However, it’s absolutely a core part of our distribution strategy to distribute Convex’s capacity into the indigenous US market. With Kinetic, it’s a slightly different strategy, in that we’re going to utilise more third-party capacity from the outset, to complement the Convex offering. But the distribution strategy is different, and it’s initially based here in London and Bermuda, as I’ve described. Eventually we may look to broaden that distribution base further afield. But we’re building it as a separate legal entity, owned by Convex, and we have no intention to sell the platform — it’s going to be part of our infrastructure going forward, and a core part of our next five-year distribution strategy. So when we look at premium in future, it won’t just be Convex’s own GWP — it’ll be gross premium managed by Convex. 

Robin: We’re [23:00] getting near the end, but Andy, my last question is for you, because I can’t work out whether it’s a job I’d love or a job that would terrify me. Are you enjoying yourself, or do you find the responsibility of the budget and the access to everything shiny deeply intimidating? What do you think? 

Andy: It’s a constant state of both, Robin — somewhere between exciting and terrifying. The opportunity’s massive, the technology’s ever-changing; the second you think you’ve understood it, things move on at speed. It’s been a hell of a learning curve, but an amazing experience, and more exciting than terrifying for now. Investing in the less glamorous stuff — reliability, trust, the controls and audit around AI — will make this less terrifying rather than more, over time. 

Theo: Can I add something, Robin? It’s a big challenge in front of us, and that can be intimidating, but the team we’ve put together — internally, and working with the Quotech team and with [24:00] Provectus, the delivery partner for Amazon — have made this a really exciting and enjoyable journey, because everyone’s pulled together. People see the benefits of what we’re trying to do, and are excited about building something genuinely innovative and differentiated. I’ve managed a number of large technology projects in my career, and I’m really appreciative of, and enjoying, how this one’s coming together. 

Andy: I agree. I’ve also seen underwriters get an early look at where we are, and they’re saying these are things they haven’t seen before. There’s a real excitement building around what this will mean for underwriters as they come and join. 

Robin: For what it’s worth, I love it, in the sense that your competitors are having to take this toolset and build it over existing legacy capability, which gives them enormous handicaps in how, and how fast, they can [25:00] deploy. People like me, who get to spout theory completely disconnected from the reality of doing business, have always asked why a big carrier doesn’t take a greenfield site and build this from scratch, getting the full benefit of doing so. This seems a marvellous way to get the best of both worlds — a well-respected insurance business, while getting to play with all the shiny new toys. The next two to three years will show the difference between those who’ve invested in this and know how to use it, and those who get left behind, whether because of the state of their data or anything else. [26:00] 

Theo: Going back to an earlier point about adoption — we’re not seeing as much of it yet as you might expect, and a lot of that comes down to the legacy issue. It’s much harder to put this technology on top of existing platforms than to do it the way we’re doing it. None of it is easy, but that’s part of the adoption challenge you’re seeing — it’s actually very hard to do at scale. You can do it in small pieces, but doing it at scale is much harder in a legacy platform world. 

Robin: I agree. Guillaume, you’re always a bit of a thinker in this space — how do you know this technology won’t be legacy in three to five years’ time? [27:00] 

Guillaume: I think the second-biggest challenge we haven’t mentioned is making sure that, in two or three years, what we’ve built isn’t legacy. We always thought that with our platform, if we didn’t change anything, we had about 10 years before it would become legacy. Now, with the pace of technological advance, I don’t know if it’s two years, one year, one week, a day or an hour — it’s incredible. I always describe it a bit like what Andy was saying: we’re running full speed towards a field that looks like paradise, everything green, butterflies — but if you turn round, the road is collapsing behind you, and if you stop, you’re gone. It really feels like that at the moment — incredibly exciting, but also nerve-wracking, because you’re constantly asking, “are we going in the right direction, are we doing the things that mean this will still be a great solution in two or three years?” We think the way we’ve built it — and because there was no legacy to build on for this project — means everything’s been done so that it’s flexible enough to keep improving and progressing. 

Theo: We’re going to find out. 

Robin: Is there anything else before we sign off — anything you haven’t covered that you’d like to? 

Guillaume: I think so. 

Robin: I wish you all the best of luck, and, as Guillaume says, I think we’d be remiss if we didn’t get you all back in a year or two to tell us how you’re getting on, because there are a lot of people watching carefully to see whether you make a success of this. Good luck, and thank you so much for joining me. 

Guillaume: Thanks, Robin. 

Andy: Thanks, Robin. 

Theo: Thank you, Robin — appreciate it.