Mark Twigg & Rob Agnew: Agentiv-x: Why neurosymbolic AI could transform insurance broking (408)

Speaker A: Well, this week's Incitech podcast is all about a new Incitec member, agentif X. I'm joined by two of their leadership team. Mark Twigg, who's the CEO, and um, Rob Agnew, who's the Chief Strategy Officer. Welcome both.

Speaker B: Thank you, Robin. It's good to be ah, here.

Speaker C: Yeah, thank you, Robin. Looking forward to it.

Speaker A: Well, thank you first for your commitment to membership and joining us. You're, um, exactly the kind of company we like to see. It's doing something exciting and that's why we were so keen to get you on the podcast. Let's first talk about the two of you. Mark, what were you doing before Agent of X?

Speaker B: Slightly odd background for someone running an AI business. Both of us came from financial services regulation and corporate communications and strategy. And having spent probably more than 20 years dealing with the consequences of poor decision making, primarily in insurance companies. We worked with banks and asset managers too, but there was a lot of work around mis selling misconduct, governance failures, compliance busts. And we saw the output of all of that in poor consumer outcomes and regulatory enforcement. Massive fines in some cases. So having seen what happens when poor decisions, decisions compound across an industry for clients and what that means for corporate trust and trust in markets, we just felt that there was a better way of doing that. And using AI and data, uh, to predict what is happening and prevent it just seemed like a sensible solution.

Speaker A: You are one of three co founders. Rob, who's joined us, is another. Then there's one other. Who are you missing? And then probably a little story about how you all came together, why the three of you and um, how did you meet?

Speaker B: The missing leg of the stool, if you like, is you'll hate me Describing him as that is Karim Derrick, who's our Chief Product officer and he spent about 20 years working in various tech roles, most recently with a leading insurance legal tech business. And he was creating technology solutions to support claims management in insurance, which centered on improving decision consistency, explainability and improving the governance layer around that. And as a threesome, we all provide very different contributions to the process. As I say, my focus was looking at the consequences of poor financial services decision making. My colleague Rob here, who's the chief Strategy officer, dealing with commercial and strategic issues within regulated markets. And then the deep insurance technology experience that comes from Karim. So we're all trying to solve the same problem, but we have different perspectives and different solutions.

Speaker A: Rob, your turn. Um, Mox set you up quite nicely there. Tell us exactly what agentive X does. Does.

Speaker C: Yeah. Sure. Agentivex is the uh, placement co pilot for commercial insurance. So we are effectively helping brokers get clients on the right cover faster. The product is designed using neurosymbolic AI. This is a form of AI that effectively uses decision science and we've got a proprietary decision science developed alongside the University of Manchester to capture the broker's view of a client's risk. And then we use agentic AI to actually review the company information and documentation that they receive and assess it against the policies that they receive ultimately from the underwriter. What you end up with is a very clear and evidenced client recommendation uh, at the end. And because we use the decision science to govern the AI agents, every step of our recommendation is documented. So it's completely auditable, a white box as the industry terminology now defines it. And that means that the broker uh, can always review and challenge and refine every part of that recommendation whilst we using the agentic AI we remove the manual work. By making everything explainable to the clients we reduce the E and O risk tremendously. And then from a broker standpoint the m most important part uh of this is if you're able to evidence the reason for the client recommendation then you are evidencing the quality of your advice beyond simply price and premium.

Speaker A: Just go back a bit to go forward. You talk about neuro symbolic AI which you push quite hard to do your website. I won't be the only one who doesn't know what that is. Tell us a little bit about that. I think two parts. What is it and why do you push it so hard? Clearly something about it that makes you think that this differentiates for you. Sure.

Speaker C: Let's start with the explanation, breaking it down into two parts. The neuro side is what people now mostly associate with large language models, your neural networks that can read large volumes of really unstructured data, spot patterns within it, et cetera. Everybody will be relatively familiar with that within the AI conversation. The symbolic side is slightly more and people will not be happy with me using this term but a traditional form of artificial intelligence. So the rules, logic, structured knowledge that we use for the evidencing of a decision and the reasoning behind it. So the neural side helps the system interpret the sort of messy real world data that the brokers are exposed to when assessing a risk. And the symbolic side helps the system reason over that information and explain its conclusion. The reason the term is becoming more prevalent is because of advances in both sides of this technology. We have developed a uh, proprietary data science layer called evidential reasoning with the University of Manchester. And that's what gives us the ability to weigh the evidence from the information that is extracted using the large language models. So in plain English, large language models do the paperwork, pull all of that into a decision framework where our decision science layer effectively helps the broker with the judgment across that decision and ultimately the client recommendation.

Speaker A: Got it, thank you. I'm going to make sure I can speak knowledgeably about uh, neuro symbolic AI. From here on in, you've gone after the broker side of things, which uh, has been relatively unusual. Most people going after the underwriting. Why did you do that? What was the problem that you think the brokers had that you set out to achieve with this?

Speaker B: I think the fundamental point is that the brokers have a lot of work to do in order to do their job properly. And if you think in complex commercial risk markets, you might be comparing 8, 9, 10, 12 policies and each policy might have several hundred words in the policy document and any supporting schedules. And the notion that you can review all of that and match the best policy to the client needs all the time is clearly going to be very difficult. And we know from a lot of the conversations we've had with brokers over the past few months that you have to take shortcuts and in some cases they don't provide a huge amount of detailed review of all of that policy documentation. And the danger with that is of course that you don't really understand what the danger is until you get to the claims process and the end insured client realizes that the COVID they thought they had isn't the COVID they've actually got. You end up with coverage disputes, you might end up with litigation legal fees. You may well have carriers settling claims that they shouldn't do simply to keep clients happy. So you get a lot of leakage. All of these problems are uh, widespread in the marketplace. And whilst the scope of broker businesses is very, is huge, from the large global companies through to the small mom and pop shops, there's not one size fits all. But I think there's a huge potential to capture information better upfront. And if you can get better information upfront, uh, about exactly what the client needs and ensuring that you've matched that need, then the ability to address all of those problems later on in the life stages of the policy is huge.

Speaker A: But that all sounds very compelling. But technology has been able to solve problems for brokers and underwriters for quite a long time without them truly embracing it. How enthusiastic are the brokers are you Getting good traction.

Speaker B: We're getting very good traction. We have a client lunch next week where all of the big brokerage firms are attending. But there's two ways you can answer this question, Robin. There are the incumbents and we know what the barriers are to take up amongst the incumbents. They might have an existing tech stack with existing SaaS providers. The brokers that they're employing might not be that engaged in technology. And I think you overcome that by having the best product on the market and create a compelling reason as to why they should be using it. I think if you can demonstrate clear KPIs that show that using our tool is going to help your business become more profitable, more scalable and promote better client relationships. I think there's another element to this and, um, AI itself is going to churn the marketplace. I think we are going to see a lot of new entrants coming to the market who are digital and AI native businesses. And being part of that conversation, I think will be truly revolutionary in terms of how the insurance industry services clients, the scope of people it can get to, the scope of COVID that it can provide access to the market. And that's going to be a real challenge to the incumbents. And I think being part of both of those conversations presents a huge opportunity for the industry, but particularly given the market timing for agentive.

Speaker A: Right now I see you're doing it as a lunch. Some things will never change. If you want to get a broker together, then you'd offer him some lunch.

Speaker B: We will also be at the BIBA conference in Manchester this week where we'll be offering them alcohol.

Speaker A: So of course, Rob. So over to you. I get the overall problem you're trying to fix. Is there anything that the brokers have attached themselves to? That's where I'm going to start. That's what I want to do.

Speaker C: Yeah. So I think the difference in our, uh, solution is that it's not applied to a workflow, it's applied to the way that the brokers work through a decision. And so what we're finding is when the platform is placed in front of them, it's rather intuitive. They understand exactly how the platform is trying to assist them and aid them in what is still a very manual process. I think they do recognize it as a new way of leveraging technology because in part it is the knowledge between their ears that we are encoding. It's about actually digitizing and creating an AI enabled process that plays back to them how they should make decisions and how to make those decisions more consistently and favorably towards the client. One of the great burdens for everybody that we speak to is the compliance burden. They are spending more and more time trying to move information through an increasingly governed process. And then also the other aspect is just the sales process. When you're particularly working in mid market SME clients, you are asking these businesses to part with a considerable amount of hard earned revenue and you are talking to them about risks that are becoming increasingly intangible to them. Until these things happen, businesses tend to struggle to value them and quantify them in terms of the hard earned revenue loss. So when you are able to demonstrate to clients with full explainability what a policy actually covers, that makes a real difference to your sales process. You're saying I've gone out to market, I have four policies in front of me and I can materially uh, tell you that this policy provides you with the most value for money because it is providing you with COVID across this set of incidents for the same price as these policies are covering to a lesser degree. So those are sort of the two main drivers for adoption.

Speaker D: Insurance has spent the last two years experimenting with AI. Now comes the hard question, where's the commercial value? On July 7th in London, Instec and AI Risk bring together 450 insurance leaders for a one day event exploring the age of agentic AI. From M Strategy to enterprise implementation. Here, real deployment, case studies, lessons from banking, live AI demonstrations and practical discussions on underwriting, claims, operations, governance and ROI. Ticketstar from £200 plus VAT register now on our website.

Speaker A: There's a question I haven't prepared you for but I'm intrigued to some extent you're selling to the brokers and I get that. One of the key features in all of this, and I know it's still early days, would be to get the client's perspective. Have we got to a stage yet where we've got some sense that this is something that brokers can sell to the clients? I can see that the more client advocacy there is for this, the more the brokers would like to get behind it.

Speaker B: One of the things we have been testing this on is the end clients themselves. Not as a means uh, of improving the intermediation process with insurance, but, but how they support their own risk management. And you have to remember a lot of commercial risks are not insured by insurers. They're carrying this risk on their own balance sheet. So there is an incentive there for large corporates, mid sized corporates, to look at the data they've got, use A tool like agentive and potentially improve their own internal risk management. One company we've been talking to, uh, early on in this process was a global infrastructure company that carries a lot of data internally in terms of the health and safety risks across its project portfolio. And that data is not being analyzed at all, not by their insurer, not internally. The ability to take that data, which may be incomplete, may have emissions, may be open to interpretation and apply that through our tool and create structured data across the entire enterprise that can help them better understand what the risk looks like. Yes, that can be helpful to a, uh, risk carrier, but it can also be helpful to the internal client. We're quite agnostic in terms of how widely this technology can be used. It can be used by brokers, and that's where we're pitching the product at this time. But more broadly it can be used in a wide variety of use cases.

Speaker A: So those are big points you're making there. But the more you make them, the more I think that the recommendations, the predictions, the decisions made by the technology, the more people rely on them, the more the whole value chain seeks to depend on them, the more they have to be right. And at some point aren't, uh, you going to come under some kind of regulatory legal scrutiny that says lots of people rely on this but show us that they can?

Speaker B: That's exactly where our thinking has come from throughout this process. And to that extent we've been working very closely in the background with the FCA to help identify and understand where and how the regulatory landscape is likely to shift and what the regulatory is looking for from compliant AI adoption. And I think this comes back to the point that Rob was raising earlier about the shortcomings of relying on large language models. And that is that they can't provide you with consistency because they are probabilistic models and they can't provide you with transparency because they're black box technology. And um, and the thing the FCA wants more than anything else is consistency and traceability. They need to know that whatever the AI tool is proposing is consistent across different consumer groups because they don't want different consumers getting different outcomes. And they want to know the why. Why did you make that decision? Why did you make that recommendation? And if you're using AI to fuel that decision making process, what process did the AI tool go through in order to come to that recommendation? And we've designed the product with that in mind. And my bugbear, if you like, with a lot of the insurtech market is that they're creating solutions that are being used in the insurance market that were not designed for that function. And I think that's the problem. There's a lot of tools out there that are being used for use cases that I think is potentially non compliant. Whilst the brokers will be relying more on AI, you still ultimately have to have a system where the broker can override the AI. So the ultimate decision is always the brokers. What we are doing is helping them to encode the decision making process and scale that decision making process. It's still a human judgment that's overriding everything, but the AI is simply helping them do more, do better, do it more accurately.

Speaker A: We had Glagogel, cio, hdi, our, uh, last agentic event. I asked him what he was most worried about as he rolled out this new generation of AI tools and he said that he was worried that underwriters would believe everything it said. In other words, they would over time become more reliant on it. And the more reliant they came on, the less they questioned it. And the less they questioned it, the more exposed they became to it. Rob, looking ahead, what next? Let's do this in parts. Uh, you've raised a little bit of money. Where are you with the fundraising and have you got any plans in the year ahead?

Speaker C: Yes, plenty of plans. A very successful pre seed raise completed end of last year, early this year, uh, at an outsized valuation and we are looking to raise our seed, uh, round in early 2027. Between those two milestones, we've launched our first product thanks to an incredible effort from our product team and RCARIM and we are looking to onboard approximately 200 to 300 brokers onto the platform by the end of the year to support that, uh, to ultimately support that seed raise. So, yeah, it's time for rubber to hit the road. Robin. We've got to now demonstrate that the commercial outcomes that we are espousing are a reality. We are very fortunate in that because we are white box. We are not KPI'd on benchmarks, we are KPI'd on real world outcomes. Using historical decisions, we can demonstrate the reduction in E and O risk if you had used our tool over the last however many months. We can also demonstrate increases in renewal, retention and new business conversion performance, as well as customer satisfaction within weeks. So we're very hopeful that the pilots that are in play at the moment will render really good quality results and things that just demonstrate a very clear financial use case back to these businesses.

Speaker A: Yeah, you sold an idea and got the money. Now that you're seed, you've got to sell a business with the, uh, parameters that make it highly investable. I can't see why you wouldn't. You mentioned the product side of things. You've told us where you started, any other products, anything that we can expect you to be announcing and launching in the next 12 months?

Speaker C: Yes. So we're quite fortunate in that, as we mentioned, our, uh, CPA came from a predominantly claims background. And so whilst we focused on the placement problem through our agentive handler product, we also recognise that claims advocacy is another big issue for brokers and they. The more we engage with the broking community, the more they talk about claims experience being a large portion of why they choose a particular policy. So we use the same technology to effectively demonstrate why a claim should be paid. It's a very simple process for us. By analysing the policy against the claims letter, uh, we are effectively able to isolate the wording within the policy that ensures that claim is paid swiftly. So we're in the midst of launching our second product very quickly as a result of it being on the same foundations.

Speaker A: Sounds very appealing. It's always the order of things, isn't it? Everyone starts. Most people actually start with underwriting, to be fair. And, uh, then only when they're up and running do they work out what the claims use. Case old claims people always come second in this. Look, we're nearly out of time. Mark, You've probably listened to a few Instec podcasts before. We always give an opportunity to give the audience something to think about. What would be your final message to anyone listening to this?

Speaker B: I think the market is ripe for disruption. There is a lot of client need that is under service. $9 trillion of protection gap globally. A lot of that's in the UK. At the same time, there's over $300 billion of advisory fees around the world and we calculate that maybe half of those clients are under serviced. So the market that we've got, I think does a fantastic job, but we are ready for a step change and I think AI can drive that step change, but it needs to be AI that is driven with the market in mind and operates within the context of that regulated marketplace. So I think I'm probably likely to see more change in the next five years than I've seen in the last 25 years. And I think that's hugely exciting and I think it creates massive opportunities for everybody, uh, and we want to be part of that journey.

Speaker A: I completely agree. Having been an advocate for the adoption of technology to make the insurance industry better. For 20 something years, it's always been a fight. There was a hostility associated with something that represented change. I've never had a landscape in which there has been so much acceptance of the fact that this new generation of technologies will make a difference and is the one to embrace. And, uh, the fact that there's great technology and a willingness to see what it could do seems to me like a real game changer. On that very profound note that we should go. You were very early in your journey when you came to us and you don't have a huge marketing budget. The fact that you chose to spend a good proportion of that on Instec is much appreciated. Thank you for your support. We'd love to support you further as you launch your product and you go raise your money. Thank you both.

Speaker B: Excellent. Thank you, Robin.

Speaker C: Thank you, Robin. Yeah. And I appreciate all of the support from yourself and the Instech team. It's been a fantastic community to be a part of.

Speaker A: Our pleasure. Thank you.

Speaker D: Well, if you've made it this far, then I'm pretty sure you found that as interesting as I did. The Instech podcast comes out every Sunday morning where we spotlight the latest news, leading voices and freshest updates across insurance that you need to know about. If you would like to take part in these conversations, head to www.instec.co to find out how you can join our network and, um, be a part of the insurance intelligence for the curious.