The GenAI Playbook: Guide to Insurance Transformation | Jacob Palmer, BCG

by Juan de Castro, COO Cytora

This a shortened version of Making Risk Flow podcast, episode: “The GenAI Playbook: Guide to Insurance Transformation". In this episode of Making Risk Flow, host Juan de Castro interviews Jacob Palmer, Partner at Boston Consulting Group (BCG). Jacob has over 13 years of industry experience in consulting, risk management, and insurance. In his role at BCG, he advises global insurers on everything from growth acceleration to the adoption of new technologies.Together, Juan and Jacob discuss a myriad of topics including why GenAI is a high priority for insurers, the ingredients for a successful GenAI strategy, the buy versus building considerations with new technology, and how the London market insurers are thinking about growth opportunities in the US market.

Listen to the full episode here

Juan de Castro: Hello, my name is Juan de Castro and you're listening to Making Risk Flow. In this new episode of Making Risk Flow, I had the opportunity to talk to Jacob Palmer. In his role as Partner at BCG, he advises global insurers in areas ranging from growth acceleration to the adoption of new technologies, and he brings an excellent understanding of their priorities and areas of focus. In the episode, we talk about Gen AI as one of those areas that are highest in the insurers' board's agendas, and we discuss how different insurers are making progress, the ingredients for a successful Gen AI strategy, the use cases they are exploring, and the buy versus build considerations. We also touch on how London market insurers are thinking about growth opportunities, specifically around the US E&S market. 

Jacob, thank you so much for joining me for today's podcast episode. For those of you who don't know, Jacob is a Partner at BCG and he's working with a number of insurers, both in Europe and the US, on many different topics, Gen AI being one of them. I'm sure we'll deep dive into that topic. So Jacob, thank you very much for joining me today. Perhaps we start with a brief introduction of yourself.

Jacob Palmer: Sure and thank you, Juan for having me today. It's really good to be on a new podcast. I'm Jacob Palmer, Partner in the BCG office here in London. I've always been in and around the insurance space. I started my career as an actuary, joining BCG originally back in 2016. I've been part of the London office here for quite a while. I did spend a bit of time outside of BCG. So I spent some time at CyberCube, where I was director of consulting, so I got to see the exciting world of cyber insurance. And there are a lot of tech developments there that will impact Gen AI at some point in the future. It's interesting to see where that develops. But I also spent a brief time at Swiss Re as well. I was part of the strategy team for reinsurance. And since I rejoined BCG back in 2020, a lot of different projects, mostly focused on Lloyds' London market, commercial insurance work here in London, and increasingly over the last year, as I'm sure any consultant you speak to will tell you, a lot of it focused on Gen AI, applications of Gen AI, and how it's really being used at anchor by insurers in general.

Juan de Castro: Definitely, Gen AI is a very hot topic so let's jump directly into it. So we are, let's say, 12 months into the Gen AI boom. Where are insurers today? How are you seeing your clients and other insurers making progress in this space?

Jacob Palmer: So maybe just before we get there, maybe just a couple of thoughts. I think the good thing about BCG is we get to see how Gen AI is being used, not just in insurance, but across all industries as well. So maybe a couple of comments in terms of how we're seeing that before we dive into the insurance side. I think it's well-known this is the fastest adopted technology, I think, in history. If you think back to when technology was incepted all the way through to where it's being used in practice by companies, this is the fastest development cycle that we've ever seen. And it's really interesting to see where some of the initial hype has actually translated into companies using it properly as use cases that are being embedded into value chains and we're starting to see some benefits on the back of it, and where it's been left in the hype cycle and companies have just looked at a few use cases, stopped development, and then gone from there. What's really interesting about insurance compared to most other industries is that the applications for Gen AI seem to be much more embedded, and there's a lot more value that we see in insurance, particularly compared to other industries. In fact, insurance has been our number one industry for Gen AI projects over the last 12 months. We can dive into why that is. But I think very specifically talking about bringing it home to commercial insurance, and particularly in London, Gen AI is very much an exec-board level discussion at nearly every commercial insurer that we've been in discussions with. We've sat on numerous EC sessions and other sessions where we've both explained Gen AI as a concept, but also looked through where the applications are going to be in practice. And I would say there are very few areas of insurance, in commercial insurance in particular, that are untouched. So we're seeing this clearly at the huge multinationals where there are millions, tens of millions in some cases, being spent on massive Gen AI development transformation projects, all the way down to conversations with individual Lloyd syndicates, where they're looking to how they can either buy Gen AI solutions or look to incremental small bills themselves to see how they can both augment and automate the different parts of their value chain. So for us, when we see it's touched every part of the sector, we see it at least being discussed across the board. I think the real question for us is where that then translates into full production of some of these solutions as part of the operating model within the insurance space.

Juan de Castro: A lot to unpack there. I think this is a great overview. I would at least deep dive into two or three areas you just touched on. So the first one probably is, why is it so relevant to insurance compared to other industries? Then we'll get into what areas it mostly applies to. And then the buy versus build is a great discussion. But let's start with the first one, in insurance you see greater opportunity for Gen AI, is it because there's so much opportunity to improve, let's call it that way.

Jacob Palmer: I think that's a very polite way of putting it. Yes, I think the reality is most traditional insurance companies have had a difficult time with tech development over the last few years. I think if you go back five years, the big topic was around modernization transformation within the insurance space. And in particular, if you look at commercial insurance here in London, those conversations are ongoing. Many of the major players still have modernization programs up and running. If you look at what's happening within Lloyd’s and Blueprint Two, the work that's happening there to digitise, modernise the Lloyd’s market, it's clear that the whole concept of tech modernization is a massive, massive topic. I think the reality is, though, insurers sit on massive legacy systems. The architecture is very messy. The data is very rarely centralised. It's often an amalgamation of different acquisitions over time that you've got to follow the flow of data all the way through from submission. If you're an interdiction insurer all the way through to bind, then you've got the past systems and the claim systems at the back of that. They're often very isolated. So you've got this perfect storm of huge data that is the lifeblood of the industry. You need data to survive. Huge systems are often very disaggregated and hard to flow data across. At the extreme, also have some elements of the industry that are still quite traditional paper-based, particularly if you look into some of the Lloyd’s space. And if you look at what Gen AI can do, and I've always bucked it into automation versus augmentation, the huge advantage versus previous rounds of this kind of technology is the data doesn't all have to be in one neat storage container to be able to see the value of the back of it. So insurers can cheat in a way, they can actually use Gen AI to avoid having to do some, not all, but some of the legacy modernization they would have to do, as well as supercharge some of the very slow processes that they currently have in a much quicker way than they would have done traditionally. And I think those two things have meant that they've latched onto this as a technology that they can use relatively easily to help really drive value.

Juan de Castro: So almost it's like it's a once-in-a-lifetime opportunity to leapfrog from quite an inefficient state into really a state of the art, right into this modernization target.

Jacob Palmer: In theory, it is, yes. And I think we can come on to a bit where some of those challenges in there are and some of those don't always work in practice. But in theory, yes, it is a way of making that happen. I think the other point to raise is that if you take this concept of augmentation and automation, Gen AI is really there, it is a super way of automating what you currently have. Again, automation has been done a hundred different ways over the years, but just the way it does it within Gen AI makes it so much more efficient than other versions of this we've seen in the past. But critically, it also augments, it also allows you to gather data from lots of different places in a very contextual way and then enhance what you're currently doing. Most technologies would do one or the other, and it's unusual to have a technology that does both together. I would say, though, that one other thing that people should bear in mind is that Gen AI is not an answer in its own right. So I think there is a bit of a danger sometimes in saying that Gen AI can solve all of the problems that I have from a technology point of organisation. The reality is Gen AI is one solution. You also have things like normal natural language processing. We have OCR, recognition software, and we also have traditional AI. And I think it's where we've seen companies use them together and use Gen AI as a bit of glue to bring those technologies and to really bring out the best of those other technologies that we're seeing the really fast process made. It's not the case that a syndicate can get ChatGPT, and then suddenly they're able to transform the way they do business.

Juan de Castro: So one of the things you said is it's got the potential to apply to several different use cases for areas of the value chain within insurance. Often we hear that insurance executives are thinking about where should they start? So what's your advice to an insurance executive around where to start and then how to start, I guess? 

Jacob Palmer: Yeah, so I think irrespective of Gen AI and any other solution, our advice on where to start is always to say, where is the value? So you really need to really look through your value chain and think through, if I were to change these parts of the process, where am I going to generate the most value? And going to the insurance mindset of loss ratio versus expense ratio, you can almost start to think about it for those two buckets. I think within that, there is clearly a lot you can do on what I would call boring operational transformation work. Again, I'm using that term slightly flippantly here. But you can look at operating or automating some of the background processes you have, core operations, etc. I think the reality is there's probably only a very low ceiling to the value you're going to get from using Gen AI in particular within those areas. Where we're seeing some of the most value being generated is where you focus it on underwriting or you focus it on claims. So you really focus it on the parts of the value chain that are the lifeblood of the organisation and really drive either that loss ratio or the expense ratio until within those parts, within those two. So for underwriting, there's a huge number of use cases. We were working with a syndicate recently and I think just across those two, there were at least 10 to 15 underwriting use cases they had, possibly more, similar numbers within claims as well. So I think two that I think worth bringing up, I think, first of all, anything to do with data ingestion and passing. So the ability to get data in one source, to extract the right information, then put it cleanly somewhere else. Again, if you think about the whole submission process, any part of that, we've seen a huge amount of value off the back of Gen AI. And that either is from an efficiency perspective, the fact that your underwriters can spend much more time focusing on the risk or the prioritisation of those risks as opposed to the manual process of getting data in for them to start with their day job. Again, quite a significant value off the back of that. And then the other thing that we've seen, and I think this is emerging more slowly, is this concept of an underwriting co-pilot. So the idea is you can almost create a cockpit for an underwriter and use Gen AI not just to bring the data in, but also to augment data externally, though the underwriter either would take a long time to gather or wouldn't necessarily gather first time around themselves, and really bring that together in one place for them to have that dashboard for them to make decisions about. So those are only two examples, but there are literally hundreds of use cases, and you can see if you break down the underwriting value chain into the micro-components.

Juan de Castro: In those two, you mentioned loss ratio versus expense ratio, right? In the end, do you see that the efficiency, so more expense ratio related, and the, call it, effectiveness or risk selection, loss ratio related, that both areas are often touched through the same initiative? The one you just mentioned right now, submission, ingestion, and triage, it's got an impact in both, right? It's a more efficient way of doing it, but also enables insurers to be more consistent in the risk selection. Is that how you look at it?

Jacob Palmer: Yeah, absolutely. I think there's consistency in risk selection. And again, I'm going to break my formula here, break my structure by adding a third element here, which is also an element of growth. You're able to write more risks by having the underwriters focus on the right areas and kind of bring more into the organisation as well. But I think to your point, yes, there is definitely an element of efficiency. If you just did the time analysis for how long it takes a risk to go through the kind of sausage machine and how many hours it takes to go all the way through to submission and bind, a substantial number of hours are saved just by using this technology to automate different parts of that chain. But I think exactly to your point in terms of risk selection as well, it allows you to focus on what's important. And I'm not going to pretend there's a magic way of being able to very specifically assess what the value there is going to be. I think the reality is there isn't. And I think there are different ways of estimating and approximating it. But the reality is that you have better information, you're able to make better risk decisions, and I think over time, you'll start to see that value really come through. But I think the critical thing for us, and I think this is something that I think the insurance industry should be focused on, this is not an either/or. It's not a human or a technology element. It's really where the two come together. It's where that human underwriting element is still going to be there. Particularly when you're looking at lead business, it has to be there in some form or another. I think it's where you can enhance that with the technology and really strip away some of these very manual, heavy processes up top of it, where some of that value starts to leak through.

Juan de Castro: And it's quite interesting the point you made about efficiency, loss risk selection and growth because again, growth is intimately linked to both of those two, right? One of the things we're seeing in the market is as you're more efficient and reduce turnaround time, your conversion rates also go up. So it's often quite difficult to split the two from each other.

Jacob Palmer: Absolutely. And I think maybe to the point that I raised before as well, maybe I think it's slightly controversial in the sense that you always still need that human in the loop. I think that is still going to be the case for quite a considerable period of time. I think that we've seen different innovations in the market. I think that there are different ways of automating bits of the value chain. But I think to your point, having a human underwriter in there that can just underwrite better business, that's really where I think the insurers should be focusing, on the underwriting side, really where they should be focusing their efforts. And to the point around growth, it's interesting, right? Because again, I don't have the numbers to hand. It'd be interesting to get your view from Cytora, is how often do you see underwriters just overwhelmed with the amount of risks they have, the amount of policies or amount of submissions that are coming in, and for them to be able to then really focus on the highest value ones for their business and be able to use technology to unpick those. You've probably got a better idea with the Cytora data than I have here. But I think that, from what we've seen externally, that again, has been a massive benefit from the technology.

Juan de Castro: Yeah, without getting into much detail, you're totally right. I think what we're seeing with our clients is that there's a benefit in terms of GWP per FT. This is the point you made around underwriters being able to quote and buy more, uplifting conversion from better broker service. And actually, one of the things which is not fully intuitive is there's also an increase in submission volume. Because in the end, brokers prefer to work with those insurers that provide a better service so that they share more submissions with those insurers. And there's a virtuous circle of those three, right? Receiving more business, being able to have more consistent reselection, being more efficient and accelerating role.

Jacob Palmer: Exactly.

Juan de Castro: So going back, part of the framework you were describing, let me play it back and see if this makes sense. Almost sounds like there's a first phase, which is about digitising the process to make the process more efficient. You talked about ingestion, triage. Then once you've got that and you've got all the risk data digitised, you can start getting into the co-pilot, so it's almost like a next maturity level, and start providing insights and recommendations to underwriters. And then eventually over time, you can start thinking about full automation. But I think that's farther away.

Jacob Palmer: Yeah, and I would take it with the next stage in the process as well, which is around policy production as well. I mean, if you think about what Gen AI is really there for, it's really the creation of content. If you dig into where Gen AI is, where the clinical purpose was. So again, being able to use that information, being able to then produce some of the policy documentation, again, huge, huge potential saving there. Even if it's a case of, okay, well, you want to take clauses from previous other risks that we have or standard clauses that are related to this particular line of business and bring that all together in a very bespoke, but very fast and efficient way. Again, just carrying on that value chain process. Again, the ability to extract knowledge, to put it in the right place, and be able to read very complex contextual policy documentation and pull out the right bits for this particular risk of the underwriter, is the next kind of natural evolution there as well.

Juan de Castro: And you also touched on claims. Often, for whatever reason, I think the industry is thinking about underwriting first, but then also starting to think about, okay, how does it apply to the claims workflow? So how are you seeing insurers thinking about Gen AI in the claims context?

Jacob Palmer: Yeah, so claims is an interesting one. I think there are a couple of areas where we're already seeing that come through. I think the first thing that we're seeing is when it comes to this kind of core claims management, and I'm trying to think of the right word here, almost the ability for the loss adjusters to get all the information together in one place. So claims can be massive, especially if you look at the specialty end, and massively complex. Claims adjusters and claims handlers have to go through huge reams of documentation, often very complex, very technical documentation, and synthesise the core information in order to make a decision. So what Gen AI can do is it can take the contextual information about what has happened in reality, it can also then take all of the policy documentation, then all the technical, even engineering documentation, depending on what the claim is, and really try to figure out whether this is a serviceable claim or not. Actually, one of the things we've just done at BCG with a client in the US for their engineering book, is actually develop this as a product for them. So essentially, again, we're not getting rid of the claims handlers. The claims handlers just put all the technical information into a bot. The information about the claim and what Gen AI will do is pull out all the right bits of information for them to much more quickly make an assessment on that particular claim area. I think hand in hand with that comes fraud detection. It's often very, very difficult for claim systems to manually detect fraud. Again, it's perhaps slightly less common in the large specialty commercial space, perhaps more in the smaller SME space. But I think either way, the ability for Gen AI to spot pattern issues and be able to see where there are items in this that perhaps could indicate fraud? Again, traditional Gen AI has been used there for many years in many different forms. We've just seen traditional Gen AI combine with Gen AI, and this particular case has been massively effective. And again, it does depend a little bit on terms of what we mean by claims and which part of insurance we're talking about. But again, it's seen as a huge advantage there. Perhaps the last one that I mentioned, this is often a bit left field, and I get mixed reactions when I talk about this, so I'll be careful with what I say. In particular, when it comes to the auto world, and you think about the US and the big UK players, where I've seen a couple of use cases for Gen AI is the ability for the technology to listen in to claims handler protocols, and to kind of hear and understand what is being said by the policyholder and then make an assessment about how likely it is, just based on the wording that you use and comparing it historically to other calls’ transcripts, being able to make an assessment about how high value the claim is likely to be. So certain words that are being used, which as a human, you wouldn't necessarily pick up when you're just on the phone to someone. Again, if you're looking back at a history of recognition over thousands of calls over the years, the technology has been used. And actually there's one major insurer in the US that has this in practice and has implemented this program. I won't get into the regulatory side of it. I think that's another discussion which we can open another day. But I think the point around all of this is that they are combining different technologies, whether it's Gen AI, OCR, NLP, or traditional AI, I think between all those, almost any challenge you have across the technology value chain, you can think about a combination of this technology to solve the problems. The only part of the challenge is there's an almost endless list of use cases you can come up with. It's really assessing what is going to be the highest impact, what’s the highest value driver for my company? And it's going to look completely different. So whether you're AIG or you're Allstate in the US, or you're Hiscox or Beazley here in the UK, or you're a Lloyd’s syndicate, highly value-driven for one is going to be completely different to another company. That's because it's so connected to what your legacy estate is, your architecture, your current operational processes, and where you spend money. So we don't necessarily often see a huge bleeding of use cases between companies. It's very specific to each individual insurer in terms of where they're going to generate value off the back of it.

Juan de Castro: I think most people when they think about Gen AI, they think about, as you said, documents, text. I think one of the biggest strengths in Gen AI is being more multimodal and then getting more into video, audio. I think you gave the example of listening to the first notification of losses, calls, or something like that. Is that an area where you're seeing insurers thinking more about moving the application of Gen AI broader than just analysing documents? 

Jacob Palmer: Absolutely. I think it's broader than just analysing documents. There's analysing documents, there's production of documents, there's being able to interpret visual cues as well as kind of written cues as well. I think it's a much broader, more broad spectrum there. I think the one thing that's worth noting is that, I think the way I've seen it described is almost a left brain, right brain challenge with Gen AI. And I think especially if you look at it compared to traditional AI, which is actually where Gen AI isn't quite as good. Gen AI is really very good at those context kinds of tasks, the creative side. It's weaker perhaps at kind of measuring things accurately and forecasting and predicting numbers. That's more of that left brain elements connectivity that predictive AI does very well. So it's really where those two things come together in an insurance context where the value is going to be driven. Gen AI shouldn't be used for core forecasting. I don't think that's really what the model is there for. So I think it's just important to understand kind of where it can be used in practice and where it needs to be used in combination with the others.

Juan de Castro: That's a very interesting point because, as a former strategy consultant, we always know that a good strategy is as much about what you focus on as it is about what you decide is not a good use of time. So what would you say to your clients about it: don't bother or it is too early or not applicable? What areas do you discourage them from applying it to?

Jacob Palmer: So I wouldn't necessarily discourage them from applying it to any areas. I think it's where it needs to be used in cohorts with other technologies that we would focus on. But I think if you just want to use it for kind of purely analytical processes, it's not what it's designed for. So I think that would steer away from it. Clearly also just make sure of the regulatory context where you're applying it as well to see whether it can actually be used for that particular use case. But I don't say we would discourage it. I think it's just, and again, maybe we'll get onto this point, it's how it's being used and the operational kind of structure around it that is as important as what is being used for. Because then you can generate as many use cases as you want, but if you don't have the core underlying capabilities and structure to be able to deploy it properly, you need to think about other ways of getting access to technology and other ways of trying to incorporate it into your operating model rather than trying to develop it yourself.

Juan de Castro: Okay, so let's talk about that. What's the right infrastructure operating model to get most of the benefits? And I guess two questions to that. One is, how are you seeing, first of all, organisations or insurers starting to make progress? And I think, as you said, this comes from the board. How is the board enabling the organisation to make progress? And then what is the operating model that is required to get value?

Jacob Palmer: I think the reality is Gen AI has created a hype cycle. So every board member that reads business papers or that use it in any form or another will have seen how it's being used by other companies within their field and then want to see how it can be used within their organisation. So I think differently from other technologies, the pressure has come from the top in a lot of ways for insurers to explore how they can use this in practice. There is some bottom-up push from the IT teams, technology teams and product teams, but a lot of it this time has come top-down, which I think has meant the reality is there is sometimes more budget being spent on it than other technology innovations in the past as well. That's not to say that it's end-to-end there. That's not necessarily a generic statement, but I think in isolated examples, we've seen quite a lot of investment being put into this.

Juan de Castro: Are you seeing boards and CEOs almost using a central budget to fund some of the initiatives?

Jacob Palmer: I think in some cases, yes. I think also the reality is because a lot of insurers are still on modernization journeys, this budget has already been allocated for other things for many years. It's been locked up in other areas. I think there are a lot of central resources and funding being used to experiment and see where we can use it and work with some of the tech players in the market, often co-investing to see where it can be deployed. So, yes, there is quite a lot of central funding being used. I think, again, because of the speed of adoption, it's been the only option because it wasn't in the budgets last year because nobody really knew about it. I think in terms of the operating model, and I would put equal, if not higher, weighting on a company's ability to think through the capabilities that are required to deploy Gen AI rather than necessarily the use cases which it can be deployed for. And I would split it into very traditional buckets, get nothing major here, which is what is the technology architecture you are developing on? So I think that's number one, what are the processes that you have for it and around it? So from a governance perspective. And then lastly, do you have the people internally to be able to develop it, use it, deploy it, and maintain it over time? Again, we'll get onto that point because that's a really important point when it comes to fast-moving technology. So if you think through the architecture the point is, first of all, do you have the right infrastructure to be able to use it? Are you already cloud native, for example? All of these solutions are cloud native. It's the CSPs that are really pushing the solution in the market. Do you have the right infrastructure or do you have to create almost a shadow architecture infrastructure just for Gen AI for you to be able to use it in anger? I won't go into details there. But it's important to understand what technology platforms you're going to be deploying it on. I think alongside that comes the point around how you're going to manage the process and the governance around it as well. So how can you get comfortable as an organisation that it’s being used in the right way, that it will continue to be used in the right way, even when it's deployed and fully into production? And also, how can you get your internal risk compliance regulatory stakeholders comfortable that you're using technology in the right way as well? And again, we've seen some companies jump a little bit too far ahead and start deploying use cases in ACA without really considering that point. I think it's important that all those function areas are taken along the journey, thinking through how they're embedded as part of use case development so that there's no surprise off the back of that. And then the last thing, and I think interestingly, perhaps one of the most challenging things and one of the biggest blocks to companies developing in themselves is do you have the right people to be able to develop and deploy and maintain this? It's a relatively new technology. I would say from what we've seen in the market, it's relatively straightforward for a data scientist or someone with a good machine learning background to work in a Gen AI environment, but that's normally the case, but not always the case. And the reality is there are some very specialised skills you need in order to be able to use this technology properly. Even things like, for example, on prompt engineering, just being able to kind of tweak and modify the Gen AI prompts over time so you get exactly what you need off the back of it. It's just a slightly different way of thinking from traditional coding. It's much more kind of human language orientated than it is developing lines of code. So I think just the reality of the situation is most of the people in the market who know this really well are being snapped up by the cloud service providers for huge sums of money. I think Open AI, Google, Microsoft, Amazon, they're clearly, to feed the huge organisations they're creating, they need the engineers, they need the developers, they need the product managers to be able to do that. There is a little bit of a mismatch in the markets where it's not always easy for insurers to find the right people either internally or even externally at the moment to be able to develop this out for them. And maybe we'll get on to how they're solving it. But I think just taking a step back, thinking through those three elements of what tech stack have I got, what the governance and process around it, and who is going to do it for me. Again, that thinking should go alongside any use case development in the yield rack.

Juan de Castro: This sounds like a very sound framework. So thinking about the technology, the processes, the people. Is it how you see insurers are actually thinking about it? Are they going through a rigorous process of understanding this in a fast-moving environment like we are in right now? Because it's quite difficult for them to step back and say, okay, let's go through these three areas before we make any other progress.

Jacob Palmer: It's a great point. And so I think before we get to specifically what we're seeing in insurers, I think for all those three, and I think very different to how that framework has been used in the past, it has to be a very agile way of thinking about it and it has to develop over time. So we just worked with a client to develop a generalised solution, looking at the follow market and how it can be used as part of the follow underwriting value chain. I think we were on the ground for about eight or 10 weeks developing something very fast, even during that time, the ChatGPT-4 Turbo came out, which not only made the project much faster, but also significantly cheaper. And that was over a very kind of short development cycle. So the key thing in all those areas is you're going to have to experiment and you're going to have to change. It's not going to be, we can decide on the tech stack now. It's that we have to design tech stacks right for our use cases and then iterate that over time as well. Same thing with processes. The processes will evolve as you look at different technologies, as you look at, you know, as different regulatory criteria come in as well. So all those are going to be a bit of an evolution over time. I think we're seeing some think about it quite rigorously, but I think the reality is a lot have jumped on, which use cases can I deploy it to thinking initially and focus more on trying to solve point problems, which I think has to be done because that's where you generate the value, but we would always encourage to do parallel. And when we do projects, we tend to try to do both at the same time rather than do one followed by the other.

Juan de Castro: Even this concept that you just explained about being very nimble in how you're thinking about this and things changing in the period of six to eight weeks, how are you advising your clients in terms of not just their tech stack, but also what LLMs do they work with? Because today it might be GPT-4 and tomorrow Google might be high performing. So is flexibility or is the ability of being able to use different models like a core component of how you're advising clients to think about it?

Jacob Palmer: Absolutely. And I think even if you break down that tech world into different components, clearly you've got the LLM, but you've also got other parts of the tech stack that you need to look at there as well. Within each of those, there's a huge number of choices. I think particularly when it comes to the model that's being used, yes, most of the time when we advise clients, we advise them to go with two or three to start with and iterate and test to see what the results are going to be. Again, I'm not going to pretend that I know all the models intimately, but from what I understand, there are probably like 30 to 40 out there at the moment, maybe slightly more in different forms. I think there will be a convergence, but I think at some point there'll be a convergence to which ones are the highest performing ones, but we're nowhere near that yet. We're nowhere near that point in the development cycle. So clearly keep it open. Each of those models has different costs. They have different pros and cons. They integrate better with different technologies. So do you go with a, if you're a Google shop, for example, you might be biased towards using Google products, but again, the advice has always been to be nimble, be flexible, try a few different things. You can have partnerships with a few of these different firms so that you can switch between them relatively straightforwardly, depending on what the use case is going to be. And that applies predominantly for the model choices. I think for other parts of the tech stack, perhaps less so, but particularly in the area where there's the most flux, I think the advice is always to be broad.

Juan de Castro: So then linking these quite fast moving environments with one of the points you made earlier about the scarcity of skills and people that are knowledgeable in this space. Obviously, the first question that comes to mind is buy versus build. What's your view on that?

Jacob Palmer: It's a tricky question. I think, first of all, you have to ground it in where the company is today. Is it a company that has big development teams, that has a modern architecture and stack, and are used to working and building this type of technology in-house? Looking back on the insurance industry, there have been many attempts to build from-the-ground solutions in many different areas, some of which have been massively successful and some of which have massively failed. So I think you have to assess whether as an organisation, you are at the level of maturity to be able to develop this in-house or with a partner. I think the other component is whether you're just looking to experiment and kind of dip your toes in the water and see where it can superpower your business or where you want it to become part of the fabric of your business going forward. So whether you want this to become a core technology asset within your business that can be deployed in lots of different ways in a tangible way. I think if you're in the former camp, there's definitely an argument for saying, okay, look at the tools that are out there, look at the different products, see what's off the shelf. Use that to start off with and kind of build your confidence in understanding how this technology works and how it's being used in practice and then start slowly to develop it. I think if you're the kind of organisation where you want it to become the core fabric of what you do, I think there you're much more into a built discussion and think through, well, how are you going to build out the teams? How are you going to build the core processes and develop all those tech partnerships yourself? So I think it's very contextual. I think it really does depend. I hate to give a halfway answer, but it really does depend on what your starting point is, but also your aspiration for the technology as part of your operating model.

Juan de Castro: Yeah. I guess what you're saying is part of the consideration is what's your level of maturity from a development perspective. Historically, one of the key drivers of buy versus build has been, is what you're building part of your core IP that will differentiate you from your competitors? Or is this something that you think the whole industry is going to do something similar about? And therefore, it's clearly inefficient for each player to develop their own solution. And obviously, I'm slightly biased, for full disclosure. Being part of the sector, when I think of things about, for example, risk document extraction or risk ingestion, that is a problem that is quite common across the whole industry. Whereas potentially, the co-pilot is something which is very specific for each insurer. So would you agree with that distinction?

Jacob Palmer: I think mostly, yes, I would agree with that distinction. I think that it's almost the utility tools versus the bespoke tools. And I would agree with that distinction. I think the only thing I would say is there are some insurers that we're seeing, in particular, the tech-developed ones who just want to own it themselves. And I think that there is a logic there. If you can develop the business case off the back of it, then that's fine. I would say those insurers are in the minority. Understanding where tech development is within the broader commercial insurance space, I think it's predominantly more the distinction that you've made that I think will prevail in terms of whether it's a kind of utility asset versus something that's core to your organisation. And even if it's not core to your organisation, breaking that down into different components and exploring where you can buy certain parts of that off the shelf, while still owning the core infrastructure. Maybe just one other point to raise on that. I think the other distinction is where the data goes. There are clearly a lot of challenges in terms of making sure data is being used in the right way, particularly if you're looking at core underwriting claims data. I think just the ability for that traceability element to make sure that the insurer always knows where the data is being stored, and how it's being used, both from an internal governance but a regulatory perspective as well. That may sometimes be the deciding factor when it comes to which way they go.

Juan de Castro: Yeah. We’ll wrap up. I want to make sure that we touch on something which is not Gen AI. We could be talking about Gen AI, especially because this is probably the most exciting phase of Gen AI, which is already kind of production-ready, but at an early stage, almost nobody knows what it's gonna look like in two, or three years, right? So I think we're living in a very exciting period, but leaving that aside for a second, other than Gen AI, I'm sure you work with clients on more traditional growth strategies, growth opportunities. What are the areas when an insurer comes to you and says, okay, I'm looking at how do I grow double digits in the next few years? What are the opportunities? What do you think about that?

Jacob Palmer: I think that it's a really interesting question. I think if you look back four or five years, all the topics, all the focus was around modernization, transformation, tech development. I think that's still there. The focus on growth has massively changed, in particular, given the market pricing cycles. I think what we've seen, particularly in property and account over the last two, three years, I think what we've seen in terms of reinsurance movement, where reinsurers are going higher up the ladder and focusing a little bit more on what the commercial insurers are retaining themselves. I think the growth topic is much more on the agenda than it used to be. I think the other thing that's interesting, particularly if you focus on Lloyd’s London market, a lot of those businesses are now grown up. If you look at the top ends of that market, you look at the big players within that space, these are now big organisations. And there is probably only a certain amount within London to sustain those businesses going forward. So we're having lots of discussions about the US, thinking through where we can tap into the E&S market. We're having discussions about growth in London, thinking through the market and how we can consolidate different parts of that, looking at some of the innovation in the market, particularly when it comes to digital flow. I think the other areas are perhaps getting slightly less focused. There is some focus on Europe. I think Europe has always been a slightly trickier market for non-European players, just because it's so disparate between different countries. And then lastly, in Asia as well. Again, we are seeing quite a bit of focus on Asia as a growth factor. That does go up and down a little bit over the years. We do see some fluctuation there. But definitely perhaps slightly more confidence in some of those markets than there has been historically.

Juan de Castro: This is probably a completely new episode - growth opportunities. It's very exciting. I think exactly what you said is that some London market syndicates are at a scale where if they want to keep on growing double digits, there's not much more to capture. So I think that's going to be a really exciting area to watch too. Jacob, I really enjoyed your views. It's very insightful, especially coming from somebody like you who talks with so many players in the industry and brings together so many perspectives on specifically how insurers are looking at Gen AI and how they are making progress on that in the final bit around the growth opportunities. So thank you so much, Jacob, for joining me today.

Jacob Palmer: Great. Thank you very much for having me.