Watch Christopher M. Wells, Ph. D., Indico VP of Research and Development, and Michelle Gouveia, VP at Sandbox Insurtech Ventures, in episode 30 of Unstructured Unlocked with Daniele Groves, Director of Product Management at Guidewire Software.
Christpher Wells: Hi, welcome to another episode of Unstructured Unlocked. I’m your co-host, Chris Wells.
Michelle Gouveia: I’m co-host Michelle Gouveia,
CW: And our guest today is Michael Duncan, senior advisor to various companies who provide services to insurance brokers and insurers themselves. Michael, welcome to the podcast.
Michael Duncan: Thank you. Happy to be here.
CW: Great. You have a long and storied career in the insurance industry. Why don’t you give us an intro to yourself and what you’ve been doing and what you’re doing right now?
MD: Yeah, perfect. So yeah, my background has been insurance the last 40 years. My last job working in the industry was with Zurich Insurance. I was a global head of underwriting excellence. My background has been underwriting predominantly all my life. So where I sit today is I advise companies who are providing services to the insurance industry, and I think it’s important to give you most of the little party pitch, which is I support companies who are focused on the digitalization automation of sales and distribution at underwriting.
MG: Well, that is the perfect background to talk about what we want to talk about today, which is very much focused on underwriting. So a little bit on the past, present, and future of what the underwriting role looks like within the insurance carrier. So Michael, on that, would love to get your thoughts on, and we have a lot of what the future could be, right? It could be all about the new generative AI type. It could be new solutions or capabilities to support the underwriting workbench, but we’d love to get your sense of where did that role start? What did underwriting look like in the past? And then we’ll just make our way through history into the future. So at Zurich or prior experiences, how has underwriting shifted from whatever period of time you want to start with?
MD: Okay. Well, let’s go back to the eighties, which is probably the best era of your music. In the worst era of underwriting,
CW: Bold claims on both sides. I love this.
MD: Maybe if I look back as I started at my underwriting career, the thing that would’ve struck me at the time was it was paper driven. Everything was about the paper. The submission that you got was paper. But interestingly enough, compared to today as an example, the information we got back then was quite limited. And maybe if I give you one example, I was a financial underwriter. I underwrote directors and officers liability, and I would’ve received with every submission, three annual reports the last three years, annual reports. And if you think about those annual reports, they’re sometimes 50, a hundred pages and you would just receive paper and then you’d have to wade your way through that. Over time, a lot of the information was well out of date, nothing was contemporary in what you received. And I suppose that’s the world where it all started. That was as time moved on, people have felt that the industry has really transformed over that period of time. But realistically, the only thing that really changed much of what happened in the underwriting world was Lotus 1, 2, 3, which I don’t think anybody on this will even know what that is. Oh, I
CW: Know Lotus. I know Lotus.
MD: Okay. There’s one. And that moved into Excel in the nineties. But that was where we started back then. And obviously as time went on, we had the internet and you could get some information off the internet for public companies. You couldn’t get anything for private companies. Nobody had a webpage. And it gives you a sense, it was all about paper. It was massive submissions, door stop, things that would turn up if you’re a property underwriter for example. And then you’d have to wade through it and large property accounts back then would take an underwriter weeks to wade through because there’s no way of putting it into a proper Excel spreadsheet and doing the slicing and dicing all that information. So I suppose the way I look at it back then, it was you had to use a lot of judgment, which is your own judgment in your head, machines, machine learning, none of that existed out there.
And if I just go to another area, pricing, this is just probably the bit that really gets me. My first job, I was given a piece of paper which told me that you find out the size of the company, you find out it’s turnover and you multiply that by a percentage to give you a limit for $1 million, $2 million, 3 million, 5 million, whatever. It’s that was it. And maybe most people would’ve got those pieces of paper from the likes of Munich Re or Swiss Re, and everybody used pretty much the same pieces of paper with adjustments here and there. So there wasn’t really any science to it, I think is the best way of putting it, nor were there any capital models, nor were there any accumulation models and there were no pricing algorithms. So that was the world. That’s the past of what handwriting was. So I’ll pause there for that one.
CW: Yeah, we talk a lot on the podcast about how streamlining your workflows, digitizing them can be a competitive advantage. We talk about how the analytics you can do on top of clean data can be a competitive advantage. What was the competitive advantage back then?
MD: Good question. I think there were two things. I think it was judgment. Some people were just better at making underwriting judgment than others, and it was relationship management. It was fine. You had to have strong relationships with your brokers to win the business because everybody was using pretty much the same rate sheet. But I think judgment was important. I don’t think you can underestimate the power of judgment, which by the way, I’ll call out early in this one is I can’t see certainly in my lifetime, and maybe that’s a bit shorter than others, but I just don’t see judgment being automated in the short term term. Maybe at some point will, beyond my timeline, it will be, but it was judgment and it was relationship management and understanding the customer. I think I spent much more time in the eighties and the nineties actually meeting customers irrespective of what the line of business was, as opposed to what happens today.
MG: Was that on the golf course, Michael or Martini lunch? Yeah,
MD: This is going to sound really weird. I was too junior to get onto the golf course. I wouldn’t have been invited to those meetings. So
MG: I don’t want to spend too much time on the past because I know we’ve got a lot to talk about for present and future. But one question on that. So everything came in on paper. In today’s world, there’s a lot of, okay, if this information is missing or we think it’s incorrect, we have all of these third party solutions that we can bring in to reconcile that information. If you had a submission come in, I’ll say, I’ll just call it not in good order, right? There was something missing. Was that something that automatically just said, we can’t underwrite this? Or was it phone call back and forth? How long did some of those things take just to get the right information in given how manual and how paper driven and how non-digital that the work process was at that time,
MD: Maybe the best way of describing is what actually happened through that whole process quickly. So I was a broker as well, but at one point, right at the start of my career, so a customer would ring me a new customer and generally that’s how it would happen. They would be, I worked for Marsh at the time and Marsh was no one as being one of the biggest financial lines brokers at that point in time in Australia, customer would ring you and say, I need a proposal form. So I would then post a proposal form to the customer and maybe a couple of weeks later that would come back to me missing some information. So I’d bring the customer right to the customer, tell me that, can you send me the additional information? Once that came in, then it’d probably take me, because you had a stack of paper, two foot high on your desk, I’d get to that submission after about a week, send it to seven or eight insurers, and I would’ve to fill out what they call a quote form, which was a carbonized, three copies, and I had to complete one for every single underwriter I was sending it to.
Then they would all get posted off to the underwriters and I could wait a week, two weeks, three weeks for a response to maybe come back to me. And then once I got that in, then I wrote to the customer, now to your point of if there’s information missing or the underwriter would send me a note, maybe this gives sense to my age. I was there when they working in the office, when they got us all together to show us the new fax machine. Nobody had ever seen one before, so it was a slow, slow process that was out there. So nobody got something that they wanted within a 24 hour, 48 hour period. It just didn’t work that way. So yeah, that gives you a sense of where, obviously it’s completely changed today, but it just gives you a sense of how the processes work then and mistakes were made. I think the best way of looking at it is things took a long period of time and just understanding the customer circumstances. You might have a six week period between start to finish and your customer circumstances have changed during that period as well.
CW: Fascinating. I want to start to transition to the present. We don’t have to rush it, but I want to make our way there. I was in and around the financial services industry for about 10 years, quantitative risk management stuff, and we provided services to insurance companies and for that whole time, insurance companies were talking about digital transformation, but it didn’t seem like anything was happening. And now it does by the way it does. We have a lot of good data points in the go back and listen to the old episodes audience. There are a lot of good data points there, but what do you think changed in the last, call it 20 years that now this is happening, that processes are being digitized, workflows are being reevaluated and streamlined. What do you see as maybe the top couple of things that changed?
MD: I think it’s a need to survive. It’s not a choice. If you don’t digitalize, if you don’t take expenses out through your workflows and the like, if you don’t get better at underwriting, more granular, respond quicker obviously to your customer needs, but respond quicker to the evolving landscape. And that could be trying to understand what’s happening to your capital consumption and so on. It’s that need of having to be able to react really quickly and that’s what this digitalization provides you. Access to that data in real time would be a really good example.
CW: Interesting. What do you see as the prime mover there in terms of what was the initial catalyst for pushing this? Obviously if someone does it, everyone has to it. That totally makes sense to me, but what really got the ball rolling?
MD: I’m just trying to think back what really transformed it. If I just think through what started, for me, the transformation at least it was when databases were brought in where I could actually work out whether I’d seen a customer in the last two years and quoted on it and I could see what I’d actually quoted on because in the past I had no idea. I may have seen it, but it was, who knows where that file is and everything else. And those databases, and it was DB four and things like that. If I remember off the top of my head, they get started to give us access to that information and go, okay, life is a little bit easier. I can understand things. And to me, that was probably the starting point and that wasn’t done on an enterprise wide basis. That would’ve been somebody in one of the operations in an organization.
To give you a sense, at that time I was working in Hong Kong and the rest of the organization had nothing. I think the other bit is email. Email was transformative in terms of what happened with insurance. And what’s interesting, my first email account was in 19 93, 4, a comper account. And the companies I was working for at the time was three, four years later before we had email. So we were all using our personal accounts. Governance would never get it, never passed governance now, but we’re all using our post email accounts because we knew it was more efficient. So if you start to see that was a real starting point of this change that was going to occur through the industry wasn’t driven by data at that point, it was just driven by efficiency. Things just worked a lot better. It wasn’t a fax going here or there. It wasn’t turning up to your hotel and getting a door stop of faxes that had been sent through to you and that sort of stuff. Now was it simple with 14 K dial ups, trying to download something was pretty slow. I can tell you and expensive if you’re in a hotel at that point in time.
CW: Yeah, I want to make a note of that and then I’ll hand it off to Michelle. I know she’s got a bunch of questions given that she’s much more knowledgeable than me. What you’re talking about there is really boots on the ground, grassroots innovation, not some top-down initiative. And I think that’s probably pretty thematic through the last decade or so.
MG: Actually, Chris, you’re right on my wavelength there. What I was going to comment was in the VC space, we talk a lot about innovation in the insurance industry over time and how do new technologies get adopted? And one of the things that we’ve as a team within our firm talked about is that insurers or the insurance industry is not technology averse. It is willing to adopt technology. That technology just needs to reach a scale at which it can be adopted effectively within the insurance industry. And I think that’s what Michael’s starting to talk about here is yeah, email was a brand new thing we started, it would’ve been great to have it roll out immediately, but the scale wasn’t there. Not everyone had email, right? And so you have to wait for that technology to get to a point where something as large and involved as the insurance industry can leverage it.
So that was just my comment, Michael, on what you were saying. The question for you is with new technology comes new challenges. And I would love to get your thoughts on, especially maybe as we’re transitioning to this conversation about the present day or the future of underwriting. It is all about data today. And when you think about the conversation of unstructured data and all this data that insurance carriers have locked in their systems and their databases that they’re not able to access, it feels like what you were just talking about of that transition to it’s not paper now it’s email. We can download things, we can file them, we can image them or P D F them. Was that the start of this challenge of filing away data that now people are really trying to get access to?
MD: And the thing is, we didn’t understand what back then, I don’t think we understood the value of data. We sort of did, but it wasn’t, and probably part of the fact is even back in the nineties there weren’t catastrophe accumulation models. So data, a lot of that stuff just didn’t, it wasn’t needed. They, yeah, the pricing models were pretty unsophisticated sort of things. It was starting to emerge and I think we started to understand the claims data much earlier than we did the value of the underwriting data. And so that’s always been lagging claims data, which is it’s better structured. The M D M for claims data is better in most organizations relative to underwriting data. And that made good sense because you needed that claims data to be able to do the pricing and then nobody was giving that much consideration to exposure data at that point in time.
So we’ve seen that. I think there’s also a point to make here is that during that sort of period, insurers got really excited about enterprise wide policy systems, transformational IT programs where you could spend a couple of billion dollars and you do all these wonderful things. And it’s fair to say it’s carnage out there of those transformational programs, I think you’d struggle to find too many that delivered what it said on the 10. I think that, by the way, the reason I mention it is because insurers became shy for a long period of time, of expending large sums of money on singular projects. They went down the route of more bite-sized things, see how that can be operationalized, see that actually the cost benefit analysis that was in the original document is actually being realized as time goes on. So I think insurers got a little bit nervous and certainly organization, pretty much every organization I’ve worked for would have at least one or two large IT projects that did not deliver anywhere near what they thought it was going to deliver.
CW: And was that digging a layer below the size of the project? Was it just an incomplete picture of how systems and processes interacted with each other? What was it that made it hard for them to deliver there?
MD: I think it was a whole range of things. Certainly I’m not an IT expert, but the way I would always look at it is there was a system that one of our organization used that was called IDE that was still being used 10 years ago. It was built in the seventies. There was another one called ID 90 built in the nineties. Then trying to put this fabulous architecture on top of it to make everybody’s life easier. The feedback I would’ve got it. It’s quite complicated to put a Tesla engine into a 1960 vehicle. That was the way it felt to me. It was out there. And again, I still think even back 10 years ago, the value of data, the completeness of data, the contemporariness of that data, it was still not a priority for insurers. And as Michelle you’ve mentioned, insurers had loads of data sitting in systems and the excuse, and this is probably fair, was we can’t access it. It’s unstructured and there’s no technology that allows us to get to it. And I think that was a fair statement. But then for the next 10 years, we just let unstructured data continue to exist within the organization. So we amplified the problem.
CW: So I think we’ve at least skirted around the edges of the opportunity with the unstructured data as that unstructured data has accumulated, what risks have accumulated with it in your estimation?
MD: I think where insurers are really good now is today is understanding is this is not a hundred percent complete, but certainly understanding what exposures they have in their business. At a certain point in time, it is not a hundred percent. And if you think about how it works, that piece of data that you get from a customer, it gets into sales and distribution goes into underwriting, works its way through claims, and finally ends up in the capital model. So that’s the way that little dot of data works its way through and where it ends up at the end of the day. And I think insurers recognize that one of the challenges they had is the speed at which that data moved through, but importantly, the quality of that data as it moved through, was it always complete? Was it correct and so on, and were they combining the right elements because it could have been unstructured, it could have been entrapped. So there were a number of things out there that insurers were sort of starting to understand. But I do, and as I said, I’m not a technology person, but I would’ve always been told trying to access unstructured data even in 2018, 19 was a massive challenge for most insurers.
CW: That makes sense. I mean the technologies we have today for doing that were nascent at the time and the people who knew how to use them, you could probably count on a few hands.
MD: And I think certainly the experience that I had that insurers tried to get in there and most insurers built huge hubs with data scientists in there trying to do things and in some cases trying to build technologies that would probably been better served, just going and buying it off the shelf from somebody as opposed to trying to credit themselves.
MG: Michael, we’ve been talking a lot about what’s happening inside the insurance carrier, right? Let’s get your thoughts on, so in the past there was mostly judgment, not a lot of guidelines, not a lot of structure. I would argue in the present there is a lot of that. You’ve got underwriting guidelines, you’ve got underwriting managers that are monitoring that accumulation risk as underwriters are binding business. How do you think about the transition from past to present and the component of the M G A or the M G U partner to an insurance carrier? How does some of that knowledge from the underwriting side get sent over what they’re developing, maybe improve what the carriers are seeing? What’s that data exchange? What are those guidelines like that structure around those relationships?
MD: Yeah, it’s a great question because earlier this week I had a call with a number of MGAs said they’re similar to MGAs in the us So program business for want of a better word, the speed at which that data is moving from say an mga, A to an insurer is very slow and still to this day, I think more importantly, the ability of the M G A to access things around an insurer’s appetite, which is continually evolving. So they may have appetite for California earthquake a month ago, that appetite changes as more exposure is loaded in and so on. So one of the challenges that exists both for carriers and for those program mjs is how quickly they can get those insights back and forth to one another. And that’s partly primarily due to how quickly that data is moving back and forth between the various parties. So I think it’s more efficient now that it’s still a challenge, especially in that M G A M G U sort of area.
CW: I worked in capital markets for a while and this kind of problem gets solved easily in a place where you’ve got a bid and an ask where I want this many of this stock or this sort of interest rate instrument, it strikes me that there’s no equivalent for appetite for risk on the insurance side or is that a naive supposition?
MD: There is an appetite that just hasn’t been a product. I believe
CW: There’s no marketplace. Yeah. All right. I got to go. I got a new idea for a business. Just kidding.
MD: Although people have tried to set up marketplaces and they haven’t been successful, but maybe that’s an offline conversation as to why they haven’t been successful. Okay.
MG: Michael’s already telling you why your business won’t work before you’ve even created it.
CW: None of my ideas are good. I know that. So before we transition to future, anything else that you want to lay down as sort of characteristic of the present state of underwriting before we talk about how do we go from here?
MD: Yeah, look, I think the last four or five years have been transformational to a journey, which is the industry is embarking on and it’s just going to happen to them. And what I mean by that is nothing’s really changed. If I think about it from a long period of time, we’ve got better models, we’ve got catastrophe models as opposed to a little bit guesswork here. We’ve got pricing models and the like. But if I think about have we really have workflows improved to a point where they are automated or they are digitalized and in the personal line space? I’d say to a large degree, yes, that is the case. I don’t think it’d be difficult to argue that segment of the business, a lot of things that happen there in the commercial line space, it’s been very different. And I challenge, and this is probably bringing us into the what does the future look like is that the industry is at the start, and it’s been at the start for the last four or five years by the way of what I suggest is an industrial revolution for insurance and it will transform the insurance industry.
It’s driven by two things. It’s driven by obviously great new technologies that are emerging that allow insurers to be able to access data. So if I bring us back to the first industrial revolution, our steam engine is analytics and our railway tracks is data. So if we think about it that way, you start to say to yourself the core, the found models and technologies that are emerging through generative AI and everything else, they’re all fed by data quality, proprietary data, and we can come back to that word later, but proprietary is very important from an insurer that they have data that they can trust and that they know the source of that data, which is critical in terms of if they ever want to contest or challenge a customer in a claim.
MG: I want to jump in on some, and this is probably a different topic for a different episode, but I’d love to get your thoughts. You mentioned what I believe to be accurately that personal alliance has been furthest along. If you think about the personal alliance versus commercial lines insurance and it’s quote innovated faster and it’s gotten to be more digital and more, much, much faster transactionally than the commercial. But we have this theme that keeps emerging of autonomous vehicles and obviously tons of data points now available if you’re connected with the OEMs and the manufacturers, et cetera. But you also have this question of liability. And so I wonder if as a result of that, that the future of personalized insurance, specifically auto insurance, will start to look more like having the challenges that some of the commercial insurance has had to digitize and innovate because it’s a brand new risk type that insurers aren’t used to or as a society we probably aren’t used to having to face.
MD: Yeah, look, I think it is to a certain degree. What I’d say is in that personal line space is really interesting and they stick to motor because it’s quite a big space of various products. An electric engine, it’s no different to bringing some carbon materials into a vehicle. It costs more to repair. It’s got all that sort of stuff’s in there. The autonomous vehicle at the moment, certainly from where I’m sitting, I’d be sitting there going, okay, God, there’s a lot of law suits against some of these manufacturers at the moment. So is it trustworthy to the degree as an insurer that you’d be really comfortable somebody driving 200 kilometers or a thousand miles or whatever it might be autonomously? Yeah, maybe it is. By the way, I’m no expert in that space. I think the challenge for personal alliance insurance is twofold, and we can bring it out on another session if you’d like.
The first one is the products are driven by regulation. They’re not driven by consumer demand. So if you just think this logically through, you buy property by householders, you might buy workers’ comp policy for people coming onto your property. You have an accident and health policy and you have travel policy. All you’re buying is property casualty and parametric through all of those products. But regulation says that you need to buy ’em all individually. So that to me presents the first challenge. The second one is, and I dunno the answer to this one, but it just intrigues me. Why can’t embedded insurance be the largest part of the market? Why is it not that everything you buy has got some level of embedded insurance and it’s value embedded insurance? I think there’s a lot of products out there that don’t create any value for the customer. So two things I don’t have answers to both of them, but they play in the back of my mind is to will personalized, be transformed by the way, through data that enables those products, enables insurers to go to regulators and say, we can do something which is ticks all the boxes. It’s just that the wrapper looks different to what you would expect.
CW: It feels like the limitation there is the consumer understanding that it’s value add, right? I understand that if I get my kids a happy meal rather than nuggets, fries and a toy, it’s a better value. But insurance is more complicated than chicken nuggets, I suppose.
MD: Yeah, maybe you should be giving them kombucha type things.
CW: Yeah, fair.
MG: But that is a good point because I also think that it’s a little bit of you have to drive it’s change management, right? Because historically this is how you’ve always bought insurance and there’s a little bit of generationally you have different buying patterns, but there’s a little bit of, well why, yes, they’re offering me insurance, but is it a scam? Do I really need it here because it’s probably covered under this other product I have or is it something that I really want insurance coverage for? Again, back to that we don’t understand the value side, but it is a little bit of, you kind of have to either go all in to drive that embedded insurance strategy I think, or if you keep offering both until there’s kind of a generational shift where everything is bought through an embedded channel, you’re going to have that back and forth of I’ve already got it here, do I really need it?
MD: Yeah, I agree. And I think that personal lines I think is a completely different, it’s a much broader subject in the better. Yeah, one to pick up. Definitely. Should we shift to the future?
CW: I think we have to. You got to take out the crystal ball.
MG: Michael. Let’s do it.
MD: Yeah, so I sort of started it with the industrial revolution. I, so I left that one for myself to fill in all the blanks. Maybe if I look at it this way and let’s stick to workflows because they’re probably more interesting if I think through what my experience is, and many people talk this, but I feel quite passionate about the fact that most underwriters spend and the percentages are not right, but 30 or 40% of the time is non-value add activities, even still today. And 30%, 40%, something like that is non-technical value add activities. So things that you could set the parameters and it could be flowing through a workflow and the workflow is a portfolio manager says, you should look at this, you shouldn’t look at that as opposed to the underwriter looking at everything because there’s no workflow that allows that to happen and the rest of it is all judgments.
What I’d like to see, and what I believe will happen is the majority of that first bit will disappear from an underwriter’s day job. And what they’re going to be allowed to do is to really make strong judgments on accounts. Now what I’ll call in the middle market space that is going to be fully automated all the way through and portfolio managers will be making that judgment. That’s what the future holds. So the S of Ss m e and the M of S m e probably then that’s happening already by the way, but the SS is being pretty much fully automated for most carriers or on the way to being fully automated the M and up into the large commercial. That’s where we’ll start to see this real shift in the coming years
CW: Just because of the uniformity of that business.
MD: Yeah, well, it doesn’t need to be uniform. It’s just that, A good example is why is an underwriter looking at if you get a schedule with 200, 300 properties and some schedules are 2000 properties, by the way, are they looking at the flood risk of every single property that’s there as an example? Probably not. Would it be better for it to be thrown that data of that property flowing through the workflow and saying, you only need to look at these two, these are the only ones that you need to look at and make a judgment call on because the rest of ’em are below the parameter that you’ve set and maybe you the underwriter setting, not your portfolio manager. You want to look at certain things. So you’re still using your judgment to craft your portfolio, but what it’s doing is taking up the value demand that occurs day in, day up for underwriters, all this work they do that they don’t need to be doing.
So I think the future is one where underwriters are going to be able to do that very easily and that will make their life a heck of a lot easier. And I’ll be clear here, there’ll be some efficiency gains. Clearly the customer broker experience will be a lot better because the turnaround times will be a lot quicker and they’ll be able to respond, but then downstream into capital modeling and all of those aspects, that’s going to move a lot quicker because that data is going, is much better quality coming through digital intake is going to improve the quality significantly very quickly of that data. They’re going to be able to access data within their systems. What I mean by that is exposure data that they have on a portfolio that is currently trapped in systems. There may be historical data that they can’t get to, but they want to see how things have evolved over time.
So the way I look at it as judgment at the point of underwriting will be transformed. They’ll be using more of that. The portfolio managers will have access to better tools, bit more data, and therefore we’re going to be more precise as underwriters in what they do and customers are going to be happier because they’ll be getting better results and so on. So I think that’s the future you could delve into. You could go down some very dark holes of saying that it’s going to be a hundred percent automated. I just don’t see that because I don’t believe judgment, as I said earlier on, is going to be something machines going to be doing in my lifetime.
CW: I think I’m with you on that, although I want to probe that you raised the specter of autonomous vehicles. I’m a huge skeptic there, but I also want to pull this other thread, which is, and I’ll frame it this way. So George Hagel, very problematic philosopher, but one of his best concepts was that history is cyclical. And so I want to tie this future, you’re talking about back to the past where one of the competitive advantages was judgment, and you’re saying we still need that judgment. A lot of the way that judgment emerged was people were deep in this data every day. It just became intuition for them. So in a world where you’re not deep in all of the data every day, where do you craft that judgment, which I think you’re saying is still a competitive advantage?
MD: Yeah. See, this is a bit where if I think of things that I did did back in the eighties, did that give me better judgment? And I’d say half of what I did didn’t do. It was just processing data. Did I become a better underwriter for doing stuff or hunting it down or no, it didn’t help. What helped me in terms of what improved my judgment was understanding over time the losses that I had on my portfolio, I became a better underwriter once I understood that you’re the best underwriter in the world for the first two years of your life, and then all of a sudden you find out that maybe your customers have losses and then you learn. So I don’t think judgment will change what’s interesting. You’re absolutely right. If the underwrite had no sense of the risk, they underwrote to the claims that emerged on that risk because it was all just flowing through models and they didn’t see it.
And the causation of the loss is where you learn as an underwriter, how did the fire emerge? How did the accident happen? That’s when you start to understand the risks that customers have. You understand the risk management activities they’re undertaking that will minimize the risk for you as an underwriter. To me, that’s how judgment gets informed, all those elements working through, is it a risk? Yes, clearly, because if it’s pushed too far, then you are saying that you lose the ability to get better judgment or receive better judgment where there’s no machine that’s going to be doing judgment for you. So that would clearly be a risk.
MG: I think it’s interesting too, going back to Michael how you were saying, and I agree with you that you’re further along on understanding the data from the claim side and when there’s a loss, right? That’s ahead of where you are from the underwriting side. And to Chris’s point, that data you’re collecting when there is a claim should help now inform the future your underwriting decisions. I think the judgment element is interesting because you could have a chat G P T or something similar internally saying, analyze the losses for the past year for all of this type of risk and identify the characteristics that were similar in all of the losses or not. And you may find that machine would just tell you purely on the data what’s there, but you may have a loss on an account that it was a freak accident. There was nothing that was inherently more risky on that account.
And so that’s where the judgment comes in to say, would I rewrite this again? Yes. Because there was nothing that indicated that there was more likelihood for a fire or a flood. And so there’s the level of what I’ve said is this technology will help get information to you. So then you still have to have your element of analysis on top to identify what does the data say and then what’s the actual behavior that we need to, or decision we need to make. Because if you leave it to just pure black and white data points, you’re going to lose out on a lot of great risks. And who knows if something may fall through the cracks that you normally wouldn’t write, but the system is telling you to, and then it may generate a large loss in the future.
MD: Can you reflect on that? I think there’s a real risk, but generated AI is going to be one of the biggest tools that will help underwriters understand what risks they’re taking on, and more importantly, their ability to drill in to something that they may not have been able to do in the past to be able to question and receive information that they probably would’ve got. They might’ve had to read through a file that’s got 200 pages in it or something, but they’ll be able to question a lot quicker. So they’re still getting that informative, that information coming to them so they can have better informed judgment. And so you don’t lose that skill.
It’s a conversation I had with some people recently, which was we were challenging my view that judgment wouldn’t be automated in the short term. One of the challenges, somebody said, well, if you look at the SS of S M E, the very small things that is flown through models, will there be risks that are taken on board in that space that probably shouldn’t have been accepted? Yes. Is the answer to that question, are there’s limit pig? And will there be sufficient algorithms going on in the background alerting somebody to the fact that it’s taken too many things accepted, too many risks out of appetite? More than likely, yes. The difference in that as you move up the chain, and where I think the risks become much bigger is just you’ve got large limited indemnity, you’ve got hundreds of hundreds of millions of dollars of exposure. You just got to leave that to an algorithm to go and say that’s the right thing to do, or the wrong thing to do is my personal view. I’m always happy to be challenged on these things.
CW: I would say the same thing. I don’t actually think it’s a technological limitation. I think you could have heuristics plus models that aren’t black and white make these decisions for you. It’s just at some point you’re going to want to fire somebody and it’s not satisfying to fire an algorithm. And it’s not just blame, right? It’s also regulatory. You have to have someone who can explain, even if it’s just, this was my judgment call, you have to have someone who signs their name to that.
CW: The other thing I would say is, Michael, I agree with you on the insights. You don’t need gen AI to do most of what you said, and I’m actually skeptical that in the next five years even insurance companies will be able to bring things like chat, G P T and their workflows just because you can’t put it behind your firewall. And that’s going to be a massive limitation.
MD: It’s an interesting one and maybe one for another conversation. I think there is a place for it. Maybe not chat G P T. Yeah, maybe it’s some of the other technologies are there that, I mentioned this earlier on. I think one of the things where chat G P T becomes a bit more tricky is the fact that it’s accessing information that you don’t know the source of that information. And from the underwriting perspective where you need to rely on disclosable information or at least know where it came from. So you can trust it a hundred percent check G P T can take it from wherever it’s at. And obviously all the copyright litigation going on at the moment in the US gives you a sense of this. This has got some way to go before it’s going to be unleashed commercially into things like insurance and probably banking and other areas. That’s my personal view.
CW: Yeah, I’m with you.
MD: Hopefully you’re not building chat G P T model on the side on weekends or something.
CW: I don’t have $60 million to blow on training.
MG: Michael, one more question. I think, so we’ve talked about the past, the present, and what technologies we think will help shape the future of underwriting. If you had to guess, and I’ll let you kind of go wherever you want with it, whether it’s personalized commercial lines, what will be kind of a standard workflow in underwriting five years from now?
MD: I won’t put a time on it. Okay. Because I dunno what the technologies will be like out there, but insurers won’t be asking for any data. The future will be insurers have access to all the data. You just need to know the customer’s name and what line of business they want to underwrite. I suspect this is 10, 20, 30 years away. But the ultimate future is insurers will have all the data. Somebody will ask for a quote for property insurance and it’ll be Chris Wells Enterprises is now the biggest IT company in the us, and they will have access to that data or it’s being granted. So therefore that’s one bit. But the other bit is insurers will know exactly what customers that they want to go after. Yep. They will have already modeled, they would’ve been using really sophisticated pricing algorithms that don’t even exist today because they’re emerging.
But what they’ll be using is they’ll be able to look at all the data that’s out there. Say I want to go after a segment of business. I know every company that operates in that segment. I’m going to model IT based and dial it up and dial it down for profitability, for capital, for whatever metric I’m looking at. And therefore somebody will come to me or I’ll go to the broker and say, I want Chris Wells Industries, you are the broker for that account. Here is the quote. Just tell Chris to validate the information’s correct. That is the future. Now, that future is probably 20 years away, maybe longer, I don’t know. But to me that’s what the future has to be because I know people have tried it in personal lines and it sort of worked, but again, it didn’t work because the ability to access the total universe, you need to be able to access the total universe of data for a segment to be able to then say, I can offer a no question, no data quote to somebody.
And that’s just not available at the moment. And then the other bit is you need some pretty sophisticated technologies to actually craft the IS interacting claims models, capital models, everything else. So you can literally dial it up, dial it down to see what output you’re looking for at any point in time. So to me, that’s the end game and everything that’s happening today or what we’re hearing about with data, what’s happening with companies that are just doing intake. So the workflows as they exist today won’t exist in the future. But I think that’s a long way away what I just described.
MG: Fascinating. Well, now that you’ve predicted the future of underwriting and Chris Welles’s future, love it. It’s the largest c e of the largest IT organization,
MD: I did say 20 years away. I think Chris may have pivoted at that point in time.
MG: This has been a fantastic look through what I’ll call the history of underwriting and insurance. But Michael, we’ll have to have you back and do a back to the Future part two episode and see how all of this emerges over the next little bit, especially as generative AI and things take off. So thank you to Michael Duncan, industry veteran and current advisor to InsureTech Companies for joining us on today’s episode of Unstructured Unlocked. I’m Michelle Govea,
CW: I’m co-host Chris Wells.
Check out the full Unstructured Unlocked podcast on your favorite platform, including: