Charlie Samolczyk: Welcome to another episode of Talking Technology, a podcast about how technology is affecting the insurance industry. In this episode, we will be looking at machine learning and how machine learning and technology together are impacting the reserving practice. I'm thrilled to be joined by Joost Wilbers and Holly Layton, both of whom are experienced reserving practitioners. They sit in our EMEA business, and they lead our reserving transformation practice.
I think we can just jump into it, and perhaps, maybe to set some context, I think it would be interesting to hear from both of you as to what's going on in the reserving space. Maybe some of the themes and the trends that you're seeing in the market and where we're helping clients. And if you're OK with that, maybe Holly, I'll ask you to lead us off.
HOLLY LAYTON: OK, I'll start. So I guess the themes in reserving have been the same for a while now in the technology space, and that's that we haven't really embraced new technologies that are available to us. Often, when we talk to clients, they're still heavily reliant on Excel. They may even have data systems set up that don't allow them to scale their reserving models and don't allow them to access the granular data that they should be using in their reserving processes.
And that's definitely not a new theme. But it's becoming more and more important to change it, especially at the minute where we've had these additional pressures from things like IFRS 17 and other regulatory challenges, which has put this pressure on the reserving team, which means they can't necessarily meet what they are being asked to do.
And we're now moving into a softer market. And that really changes the way the value of the reserving function. Because we really want to be making sure that our results are as accurate as possible. And so improving the way we do reserving and making sure that we're not holding any additional margins in our reserves is even more important.
And obviously, the landscape of technology has changed. So now we're seeing more reserving teams using tools like Power BI, but that's been around since the early 2000. So that's not even like new technology that we're embracing. We're starting to embrace technology that's been used in other areas of the insurance company for years now.
But things like using more Python scripting-- again, not a new technology, but the kind of packages and things that are available to us is something that reserving teams are starting to look at. But definitely, it's not widely used and not widely understood, but the kind of skills that reserving teams need to implement these new technologies is changing. And so we're looking to get a different group of people to work in reserving than the old kind of typical reserving actuary.
JOOST WILBERS: And then we see that throughout the market, the priorities are shifting. Reserving didn't really used to be a priority in improving the quality, the granularity of the models, but also in the speed of being able to do another reserving cycle. And maybe IFRS 17 sort of triggered it.
But yeah, just in recent years, we are seeing a lot of insurers all focusing on changing their whole reserving process, making it leaner, more efficient, but also indeed more detailed in some parts to be able to reap more of the benefits from actually all the reserving that you're doing from quarter to quarter or from year to year.
Charlie Samolczyk: Holly, you talked about the skill sets changing. I assume as more people come into the business, the expectations of people coming into reserving maybe around the use of technology is changing. I guess that's one part of the question. Is it an exciting time to be in the reserving space now as things are changing?
HOLLY LAYTON: Absolutely. It's a very exciting time. I think there's a perception of reserving that it's just a function to meet your regulatory requirements. But actually, there's a lot more that people can get from reserving. I think people are starting to realize that. We talked to some companies who have great integration between reserving and the underwriting function. And people are actually reliant on the outputs from reserving to make kind of strategic business decisions.
And that's something that we have been championing for a long time now because we see how important the kind of analysis coming out reserving is. But unfortunately, like you said, it's not being at the granularity that the wider business can use. So that focus on trying to get reserving outputs at the more granular level in a format that's usable by the wider business is like a key theme.
And that comes back to the skills part, which is in order to deal with those volumes of data, you need to be using more data science techniques in some ways And having different skills and understanding of how to use data, which is where that comes in. But yeah, that makes it super exciting to be working in reserving now.
CHARLIE SAMOLCZYK: A bit of a culture shift, as well, probably.
JOOST WILBERS: Yeah.
HOLLY LAYTON: Yeah.
JOOST WILBERS: It's not only the old typical actuaries doing all the P&C reserving. But there's also a lot of new fresh faces with more of a computing science background, actually, to enable different techniques in reserving. So we also see the techniques being used developing very quick actually in the last couple of years.
CHARLIE SAMOLCZYK: Very good. So I've certainly heard a lot about Machine-led Reserving and the applications of technology in that space. I think it would be great. Maybe you can educate me a little bit on just kind of what Machine-led Reserving is and kind of affect the industry.
JOOST WILBERS: Yeah, I'll take that one, I think. Machine reserving we're is something that, I think, the name is something that we at WTW just came up with to give to one of our solutions. But it's using machine learning in the reserving space while still using also the traditional models that everyone knows from the P&C reserving.
And what it actually does it sort of fitting thousands of models to your triangles or to your data, and seeing which models fit best. So this enables you to do a lot of the reserving without any human intervention, getting a lot of insights from it. So for example, trying to remove any systematic bias because the machine will never look back at what you were doing prior to this reserving cycle.
Whereas, an actuary usually does the roll forward, keeps most of the assumptions similar to last period. Machine learning really enables you to have a fresh, unbiased view of your reserves but also challenge most of the assumptions that you are taking. So it's beneficial both for first line and second line. Also, enabling the second line to focus their areas of review during a cycle, which is-- yeah, it's exciting.
And it's very new. We actually are planning to release it any moment now, probably somewhere this month, but I'm not sure when the podcast will be released.
CHARLIE SAMOLCZYK: Very good. Yeah, we'll keep people in suspense. Thank you for that. I mean, is it an efficiency play? Is it an accuracy play? Is it both?
HOLLY LAYTON: It's all of it. I'm so pro Machine-led Reserving. I feel like it really is the future of the kind of things that we should be doing in reserving. It's a completely unbiased estimate of your reserves driven by data. And the lovely thing about it is, is the outputs coming from it look exactly like something a reserving actuary would do.
It's been coded so that it makes sensible decisions just like an actuarial analyst would do, but just driven by data. So it gives you, potentially, more accuracy because it sees things that we can't necessarily identify. But also, it is more efficient because it takes a matter of minutes to go and reserve all of your classes.
And I think, some people are a little bit worried about these kind of tools. Like, aah, does that mean that I won't have a job at the end of it? And that's completely not how we will see these tools being used in the market. They're very much not only a challenger to us, but also, something to actually support our analysis.
So we get that first data-driven view of our reserves, and then we can spend all of our time doing the more interesting actuarial work, adding in the market insights that the machine won't have picked up. And also, talking to underwriters and finding out information that we can then apply in our reserves that again, the machine might not have picked up.
So it's going to change the way actuaries are working. And it's going to make us have a way more interesting fun job with no kind of data processing or just doing boring robotic selections. It will be a tool that will support us. And I feel like every reserving actuary should want to get their hands onto it.
JOOST WILBERS: Now, if only to play what we have also been doing internally, the 'beat the machine' game. You get your classes, you put Machine-led Reserving on it, and you try to beat the estimate that the machine is getting. You can make reserving into game by using Machine-led Reserving.
CHARLIE SAMOLCZYK: So efficiency, accuracy, and bragging rights is what [INAUDIBLE]. And Holly I loved how you described it as lovely. So that's--
HOLLY LAYTON: It's my own opinion. You guys can be the decision.
CHARLIE SAMOLCZYK: Yeah. No, I mean, so both of you work in reserve transformation. And I assume Machine-led Reserving is one tool in the tool kit, but there's probably other things that people need to think about when they're thinking about transforming their reserving practice. I don't know if you want to touch on anything there. Not taking away from MLR, but there's probably other aspects, as well.
JOOST WILBERS: Yeah. Yeah. So I think MLR is sort of a tool within the whole transformation, but transformation actually starts at the very beginning. So having a look at your whole process, seeing if your process is efficient, if you're using the right amount of data, but also indeed, if you can eliminate a lot of the manual work, automate more of what you're doing.
So Holly was already mentioning Python as one of the tools to do that. We, of course, as WTW, also have Unify as a tool for complete process automation, also incorporating these Python tools, but also seeing that just like Machine-led Reserving, you also have other areas where you want to improve the quality of what you're doing in the reserving.
So that's not only your process, but it's also your analysis. Where are you focusing your analysis? Which segments are you focusing your analysis on? So optimizing your segmentation, that's all things that you would want to look at during your whole reserving transformation.
And yeah, as we said, this is really something that we see happening a lot now, mainly due to IFRS 17, but also in substitute context. It's a hot topic at the moment.
HOLLY LAYTON: Yeah, just like echo everything Joost said. I think with reserve transformation-- one thing I would say, as well, is just like talk to other people in the business. There are sometimes tools that other people are using that could benefit you in reserving, and also, people with the right skill sets to implement those things.
And just don't be static. A reserving process should not be the same for five years. It should be constantly improving. We should be looking for new technology. We should be keeping in line with the market. And not just the market, but other business functions, as well. Often, we see that pricing teams have way better visualization setup, or more use of Python scripts. And if you had done some collaboration, maybe we could have leveraged some of the skills and use those in reserving, as well.
So people should just keep moving their process on and don't get comfortable with stuff because it works. Just because something works, doesn't mean it's good.
CHARLIE SAMOLCZYK: Holly, you talked about benefits to the business. So we don't need to go into the nitty-gritty and specifics of how Machine-led Reserving works. But if you're an insurer, what are some of the benefits that you could expect?
HOLLY LAYTON: So from machine learning, it's that objective view of your results. And so that, again, provides the challenge on what you've done today. It also helps someone who doesn't know lots about reserving, provide challenge to a reserving team, who do know lots about reserving, and say, actually, why is this machine coming up with something so different? And every time it does the analysis, it does come up with something different. So can you explain that to me?
So it's that challenge, it's that objective view. And over time, as people understand how it works, it will also take away the reserving on a series of classes that are relatively stable and can be predicted by machine. There's always going to be a need for a reserving actuary. Maybe 50% of your classes could be done completely automatically. Again, that frees you up to do other more interesting work.
CHARLIE SAMOLCZYK: So Machine-led Reserving, the term, Machine-led, and you know, there's a lot of talk around artificial intelligence right now. Is Machine-led Reserving kind of the first step down the path of going towards more artificial intelligence being applied to the reserving practice? And sorry if that's kind of a naive question, but when I hear Machine-led Reserving, I think AI, but it's probably not necessarily the same thing.
HOLLY LAYTON: I think, in an actuarial context, having fully-AIed actuaries we're a while off it. And actuaries aren't known for being spontaneous risk-taking people. So I think it's going to take us some time to fully embed things like ChatGPT and other AI into our work.
And kind of what I said earlier, the nice thing about Machine-led Reserving, in the context that we're using it, is it looks like an actuary, and it feels like an actuary. I think that's going to really help people use it and embed it and embrace it. But it's not risky using it. You actually still has the power, the decision-making. It's just a tool. Whereas, some of the other things in the AI world feel like, how will we ever explain this to the board when we use it? How can we say that it's come from this tool, which is not the case for Machine-led Reserving? And I think that's why it is the first step for reserving, but it's a step that's very easy to take.
CHARLIE SAMOLCZYK: Yeah. Sorry. Sorry. I think that's really important. Not ease the adoption, but legitimately, give people the confidence that they can use this tool, and that it will fit into almost their current workflows in some sense. So they can use the output in a meaningful way, and take it to the board, or take it to management and make business decisions off of it.
Sorry, I cut across you there.
JOOST WILBERS: Yeah, no problem. On machine that we're serving, indeed, we have been using it internally already for quite some time now on reserve reviews that we do for clients. And we are not using it as the way to set the best estimate. We are using it as a sort of an extra to add value to your analysis.
So that's also why it's easy to adopt because it's not completely going to take over your whole reserving process at once. At first, it's an addition, and then maybe at some point, you can use it to indeed replace part of the analysis that you're doing on the stable classes, let's say.
So that's why it's so easy to adopt it at first. And to come back to it, is it the first step, or actually, it's one of the areas in which we do a lot of development at the moment within what we call the algorithmic reserving area. So this is just one of the-- I think we have a menu within algorithmic reserving in which we do quite a bit of development at the moment.
So optimized segmentation, which we already mentioned, is one of the other areas which we see is seeing a lot of development at the moment. The granular reserving or policy-level reserving is an area that's really exciting and is really making progress. And it's also something that we see as a focus area for the years to come.
And algorithmic is not necessarily always AI, but AI is definitely helping algorithmic reserving. So the use of algorithms within reserving space to develop faster.
CHARLIE SAMOLCZYK: Very good. So we talked about MLR, we talked about some of the challenges that the industry is facing and the seven areas of focus, some of the areas that we're investing in and we're looking at. And any commentary before we wrap up about what the future might hold, or some of the themes that we haven't got around to tackling yet but they're out there, and we need to raise them and give them some thought?
HOLLY LAYTON: In terms of themes, I think the main thing that everyone should be trying to do is producing a more accurate result on a more granular level. And there's multiple ways that you can get to that. So I think that that's kind of a theme that we're already noticing, but I think will continue in the next 12 months as people like try and improve that.
And I think just one other thing is, is that people should be investing in their reserving function. As I said, there has been a serious underinvestment in reserving, and it's now quite noticeable the kind of investments that people have made between pricing and reserving, let's say.
So I think making that investment and really seeing and trying to unlock the value from a function that you have in your business is key. And just embrace new technology. Don't be scared.
CHARLIE SAMOLCZYK: Joost?
JOOST WILBERS: Yeah, I think, too. What we see next on the journey of what we see coming up next, policy-level reserving is, I think, within WTW at least, a buzzword for the years to come. And as Holly said, other than just the reporting that comes out of it, but it can also indeed add to the quality of your pricing, or it adds value to your underwriting team. These kinds of things are on our agenda to address at insurers throughout EMEA.
CHARLIE SAMOLCZYK: Breaking down some of the silos between the different functions.
JOOST WILBERS: Indeed.
CHARLIE SAMOLCZYK: Excellent. Well, I mean, it's been super informative. And I think, we can probably leave it there. But thank you so much for spending the time and taking us through that. And wishing you all the best.
JOOST WILBERS: Thank you. Happy to do so.
SPEAKER 1: Thank you for joining us for this WTW podcast, featuring the latest perspectives on the intersection of people, capital, and risk. For more information, visit the Insights section of wtwco.com. This podcast is for general discussion and/or information only, is not entirely to be relied upon, and action based on or in connection with anything contained here should not be taken without first obtaining specific advice from a suitably qualified professional.