InsTech - insurance & innovation
00:09:24 · Jamie Wilson
[00:09:24] S1: Yeah. Definitely. And it's this idea around where our company is actually investing with AI at the moment and kind of where do underwriters want support. And so what I mean by that is one of the questions we asked was asking all of these three fifty underwriters, where has AI delivered the most meaningful value your underwriting to date? And unsurprisingly, 51% of them said it was on saving time and on manual admin. So it's very much like lots of efficiency gains. That's where the AI has been adding value. Amusingly, 3% of them said they have received no value from AI yet. So there's probably 3% of CIOs who are very upset about that statement. But 21% said that AI had driven improvements in the quality of the decisions. So you've got effectively, you know, 50% saying AI today, it's all been about efficiency. That's where the values come from. 20% saying, actually, AI is helping us make better decisions. But we asked underwriters a different question to that one, which was what makes an underwriter excellent? So we got them to kind of imagine the best underwriter you've ever worked with. What about them make them exceptional? And the top things that came up time and time again were all expertise and judgment led. So top of that list was risk selection and appetite discipline, and then technical depth and complex risks or commercial instinct on price. All of these things were around technical underwriting, expertise, judgment. That's what makes a truly exceptional underwriter. Speed, efficiency barely registers on that list at all. And so it's this really interesting thing of the industry has effectively focused a lot of AI efforts on, okay, how do we do things that make us more efficient, which is effectively raising the floor. Like, everyone gets a little bit faster. But what underwriters really care about is exceptional underwriting. How can I make better decisions than anyone else? Like, how can I be the best underwriter? And that's in reality that decision, like, Like, how is AI driving value there? And underwriters, the kind of the response we're getting back is actually that's not where most of the value from AI today is coming from. So it's this idea of, okay, the floor is being raised.
00:11:28 · Jamie Wilson, Matthew Grant
Like, everyone's going a little bit faster, but the aspirations, the exceptional underwriting is that roof being raised of what could possibly be achieved today. [00:11:37] S0: So interesting. It's a bit of a torch in the hair in the sense that people are probably ever reacting to AI is all about speed, but recognizing you don't always have the expert in there. And so they're now saying, well, it has to be right. And then, of course, once you got it right, then it becomes about speed and efficiency. So you sort of neither is wrong, but you just need to be careful. You got one right before you get the other one, and then you can't get complacent because you got one right. I don't know if you saw it, but there's a fascinating video actually today in the Times in your robot Olympic games. And when you see what it can do with robots, you just think, well, the robot can do backflips and dance, and then then is underwriting really that difficult? I don't know if that was part of your survey where you've a view on that, but it's pretty extraordinary. [00:12:15] S1: Yeah. We didn't ask underwriters their views on kind of robotic Olympics yet, but I think you're exactly right. Like, kind of looking at how that's evolved over the course of the last few years, it's a nice indication of, okay. Well, you know, everything's up for grabs in the next few years as well.
00:12:15 · Jamie Wilson, Matthew Grant
[00:12:15] S1: Yeah. We didn't ask underwriters their views on kind of robotic Olympics yet, but I think you're exactly right. Like, kind of looking at how that's evolved over the course of the last few years, it's a nice indication of, okay. Well, you know, everything's up for grabs in the next few years as well. [00:12:30] S0: And then AI as well. We started to about AgenTic. We run an event on AgenTic. But can you just give me your perception of what you came out through the survey about this distinction between what it might be between the generic term of AI and a slightly more focused term around agentic. Clearly, it's got a slightly different meaning as well.
00:12:45 · Jamie Wilson
[00:12:45] S1: I go to a lot of industry events, talk to lots of people in industry on the tech and on the insurance side. Also, bit of a pedant when it comes to terminology. I think probably the joys of being an actuary by trade. I'm listening to talks and people are talking about machine learning, AI, large language models, agentic, and lots of the time these terms are used interchangeably when in reality they often have, like, very specific meanings for very specific context. The way I would describe kind of talking about AI versus agentic, I like using the phrase AI assisted versus agentic. What I mean by that is AI assistance is where a human is doing a piece of work, And occasionally, as part of that piece of work, they will call out to AI to help them with something. So it might be, you know, if I'm an underwriter, let's talk about that. That's what the survey was all around. If an underwriter... If I'm going through my day to day work, a submission comes in, I'm responsible for taking that submission all the way through to quote, it may be that I call upon AI to help me with certain parts of that workflow. So I call upon AI to help me ingest the risk, or I call upon the AI to help me check to see whether or not this risk fits within my underwriting guidelines, for example. That's an example of AI assisted. The underwriter, the human, is in the driving seat. They are moving the work along, but they just get AI to help them at various points. Agenic kind of workflows or Agenic underwriting is the inversion of that model. So the agent can actually move work along the workflow and can actually do the work and calls out to the underwriter or calls out to the human, whatever workflow it is, to get help when the agent can no longer do the work. And that could be because the agent either doesn't have the understanding or it's met certain guardrails that it can't go beyond. But, like, crucially, that's the difference between those two things.
00:14:31 · S2, Matthew Grant
[00:14:31] S2: The market is separating MGAs with strong foundations from those built for better conditions. At InsTech's upcoming one day MGA conference, we'll explore what that means in practice from capacity strategy and underwriting performance to technology, governance, and long term growth. To find out more and register, visit www.instech.co. [00:14:49] S0: I used to call that the humble algorithm where the critical thing there is it does know it, doesn't understand, and not get into that whole world of hallucination, which is what you're doing at HX. You released a product Hyper Operator, which has extended your reach. Can you just talk a little bit about what that is doing?
00:15:03 · Jamie Wilson
[00:15:03] S1: We announced the Hyper Operator in July of this year at our flagship event in in London, HX Live. It's kind of an extension of the general product direction in some respect. We've got to this point where HX, the platform, effectively an an AI assisted underwriter workbench, end to end underwriter workbench. I'm gonna use that term AI assisted very purposefully. The hyper operator is effectively... Sometimes we describe it as the AI agent for underwriting work. And it runs across the HX platform, and it has the ability to kind of leverage specific tools and specialist agents for whatever it's working on. And it can effectively act on the underwriter workflow rather than just advise or support on individual kind of like work points. And it does all of that within the platform's guardrails that the insurer can set up. So kind of that's a bit of a mouthful, but it's that conversion of agentic to AI assisted in the sense of the hyper operator can take a submission and bring it into the platform and move it through the workflow, take it maybe through ingestion, through clearance, get it through, like, some sanction screening and basic checks, get it to triage. And maybe at the point of triage, the guardrails are set up such that, oh, this is on the boundary of being an appetite. So I now need to call out to an underwriter to take a look and make an actual judgment decision or judgment call about whether or not we should move on to the next stage of underwriting, if that makes sense. And so kind of the underwriter or kind of the insurance company, they can set those guardrails. Those guardrails are really key because it may be that I've got one line of business over here that's really complex. My appetite for agentic workflows currently isn't there, and I want an underwriter to be involved in every single decision point. And so what I almost do there is I dial up the guardrails so that the kind of the agent's autonomy is dialed down in essence. And so the submission comes in and maybe the agent does ingest the risk, but someone has to approve that ingestion of data. And then the agent moves into clearance and populates everything, but a human has to get involved with that step. So you have the ability to effectively dial up those guardrails. You also have the ability to dial them down.