CHARLIE SAMOLCZYK: Welcome to another Talking Technology podcast. I am Charlie Samolczyk your host, and today I am thrilled to be joined by Tim Rourke and Pardeep Bassi. Tim is a Senior Practitioner and leader in our personal lines practice, travels all over the world, meets with clients, does really good things. Tim, thanks for being here today.
TIM ROURKE: Thanks, Charlie.
CHARLIE SAMOLCZYK: And Pardeep is our Global Proposition Leader in Data Science. Similar to Tim, global role. Works kind of on the front and leading edge of the data analytics practice, helping clients both build up their practices, but also working with our technology and seeing how we evolve that. So, thanks for being here, Pardeep.
PARDEEP BASSI: Hi, everyone. Looking forward to the conversation.
CHARLIE SAMOLCZYK: So, in today's episode, we are going to delve into, actually, I think it's a really interesting topic. So, it's the growing importance of monitoring in insurance analytics. What I mean by that, when I hear that, and obviously we'll get into it today, but as insurers over time have built up their modeling estate and the sheer number of models that they have, I think it becomes more and more difficult to manage that estate.
And so we'll be looking at some of the tools and practices that you can apply. I think at least even on the operational side, to get the most out of your modeling estate, but also, I think importantly, where do you focus your team's energy and their time so that you can be as competitive as possible? And I that's where the real gold is and where we want to help our clients mine it.
We'll talk a little bit about some of the challenges that insurers are facing, and we are super excited to talk about our newly launched automated monitoring tool called Radar Vision. But I don't want to take away any of the thunder from Pardeep and Tim, so we'll leave that.
So before we get into it, though, I do want to do a little bit of an icebreaker. So to both of you, and you guys can answer how you want here, but what's the first piece of technology that you remember taking on that really changed the way that you worked?
TIM ROURKE: Good question, Charlie. I think actually when I first started working in the insurance industry many years ago, and I know that I can say this because at this particular point in time, I did not work for the organization that I work for now. But when tools like Emblem were being introduced, they were game changing pieces of technology for me when I was working in insurance companies. And the fact that they just brought to life the data in a way that we'd not really seen before was very important.
And back to, I suppose, the conversation we're having today, there wasn't a lot of models back then. And what you did model was quite powerful. And so yeah, that always resonates with me. It's got a special place in my heart as a client rather than being an employee in the business.
CHARLIE SAMOLCZYK: Yeah, good plug.
TIM ROURKE: Thank you.
CHARLIE SAMOLCZYK: Yeah, Pardeep?
PARDEEP BASSI: Tim took my one. I was about to say Emblem as well, but I'll go for R. When I started using R, it gave you incredible amount of flexibility and freedom. So, we could use more or less any open source algorithm and really leverage the latest cutting edge techniques in machine learning and AI. And then the visualizations and the customizability around all of that was very interesting back in, say, the mid 2010s for myself.
CHARLIE SAMOLCZYK: Cool. Thank you. I think if I answer that question, I mean, it's not insurance focused. I remember when in one of my first jobs, I got a pager. I just remember that feeling of like being connected and I used to wear it on my belt thinking I was cool.
So let's get back to the topic. We're going to talk about insurance analytics. We're going to talk about monitoring and managing your model estate. So first I think it would be helpful just to, as a little bit of context, I think, for everyone listening. So why are we talking about this? What are we seeing in the market? And what are insurers currently struggling with?
TIM ROURKE: This is a really important topic, Charlie. And one of the things that I've noticed over the last few years as model real estates have grown is the fact that the need to actually be running your business is really, really important. And if you've got a massive model real estate and a lot of your very scarce and valuable pricing teams are doing model admin, it doesn't allow you to run your business.
And this is something that has been a frustration of many C-suite executives that there's more time spent on actually doing the models and using them to effect good business change. And it's become a frustration. And then when you think about the complexity of all of the data, of all of the different methodologies and how data science and machine learning have kind of pushed the value and the kind of insight that you can get from models and bringing that all together, it's become a difficult and time consuming environment for insurers.
So this is something that we've picked up on as consultants. This is something that we've helped clients with in terms of how they can get the best use from the human beings working in their organization, that they're not just turning around the handles of model development. So yeah, this is a big problem that we've seen in the market.
CHARLIE SAMOLCZYK: And I think you said, Tim, the massive model estate. Just help quantify that a little bit. What is a typical insurer dealing with when they're looking at their modeling estate?
TIM ROURKE: It obviously varies. But you are going to have some insurers who are going to have retail pricing models and they can be abundant. You're going to get risk pricing models, fraud models, claims, analytical models, operational models. And you can be getting up into three figures with some organizations. And there are some that will have less. And there's obviously big decisions about how many models you want, and that's a science in its own right.
The number of models versus the value and that kind of efficient frontier. But yes, you are seeing a lot more model real estate than you have seen historically. And I think that the way in which data science has opened up these new methodologies and increased predictiveness and so on, that's added to that model real estate as well.
PARDEEP BASSI: So I don't think it's just the number of models, Charlie. I think it's the complexity of the models and the different model forms. So, insurers went through a phase where they tested more and more external data and just threw as much as possible as they could into more traditional model forms. They've then gone into more sophisticated model forms GLMs, GBMs, neural nets. There's a whole range and combination of models that insurers are using.
Quite a few of the more later used techniques, they're less stable over time. So what we're seeing is not only do you have more models, but the models are relevant for less of a time period. So the increased need of understanding whether they are appropriate and relevant to changing market conditions has increased.
TIM ROURKE: Yeah, that's a really good point, Pardeep. And the fact that people are worried now [that] they are making bad decisions off the back of outdated models. We go into insurers, and we'll say to them, ‘when was the last time you looked at this’? And they say, ‘well, I don't really want to tell you, but it's probably been about x months ago’, which is far too long. So, there is a recognition that some of these models might not necessarily be giving the best information that they can be, and that is a frustration. But again, when you've got limited resources, taming all of this model real estate, and as Pardeep said, the complexity of it and the changing nature of these models is hard.
CHARLIE SAMOLCZYK: Pardeep, I know you sometimes talk about this being a little bit of kind of a, not to be dramatic, but a future battleground, a little bit of where insurers are going to compete. I mean, certainly I think there's still right up there at the top is the modeling techniques, how efficient, how predictive are your models. But I think that this operation and this management and this kind of gaining insight from your modeling estate, maybe you can elaborate on that a little bit.
PARDEEP BASSI: Yeah, so, my view is we've gone through stages of building more and more sophisticated models, more models in general, breaking the problem down into more granular subproblems and throwing more data at it. Eventually, every single decision will be made, if not made directly, but heavily influenced via a model of some sort. In that world, I think the game will be very much who can understand the strengths and weaknesses of the model and react to emerging conditions as quickly as possible.
Because everything will be purely model driven, the game will be, how do you maintain that performance? How do you gain that advantage over other competitors who also have the same number of models, the same data? And it will be the ability to apply your judgment and understand what's happening as quickly as possible, which should give you a real edge over your competitors.
CHARLIE SAMOLCZYK: In the intro, I gave a little spoiler alert. Obviously, we're very pleased and excited about the release of Radar Vision. Pardeep, do you want to tell us a little bit about what it is and how we think it's going to help the insurance industry as a whole?
PARDEEP BASSI: So, I'll just start with this is something we've been developing over the last 18 to 24 months. We've developed this because of what we've seen in the market. So everything myself and Tim have just mentioned and what we've seen specifically within the insurance market. So a lot of insurers have really struggled with changing and highly volatile market conditions.
So what makes insurance different to other industries is primarily it's because you're trying to predict an event quite some time before it actually happens, and it takes quite some time for it to develop. So if you're building models and using models which are based off of the past, if something has changed more recently since you've built the model or towards the end of your modeling period, it's very hard to react and actually have the most appropriate model live.
So with that understanding, we've built Radar Vision seamlessly integrated into Radar. It allows you to take all of your workflows and analytics and models that you've built in Radar into the phase of monitoring. So what Vision does is it automatically analyzes all of your models, all of your emerging experience, and it tells you when something isn't as expected. And it does that driven by models, driven by AI, and it prioritizes that insight in terms of where should you focus your limited attention in the most appropriate place, and then it gives you a suggested view in terms of what you should do and what that's worth to you as a business.
So we're really trying to use modeling to give you the most relevant insight based off of emerging experience and tackling what I think is the single biggest problem with insurance modeling, which is the past doesn't always represent the future. How do you bridge that gap and stay competitive?
CHARLIE SAMOLCZYK: And you sneakily said the word AI in there. So I think that's a pretty exciting part of the proposition, I would think. Do you want to talk a little bit about that?
PARDEEP BASSI: Yeah, when we talk about AI, we mean everything from simple rules to machine learning models. And where we focus a lot of our attention at WTW is the development of our own custom proprietary algorithms. So with insurance in mind, with monitoring in mind, we've developed our own specific algorithms, which not only prioritize multi-dimensional segments, but they also gives you a view of how similar they are to one another.
So we'll automatically generate a range of segments, multidimensional in nature, and prioritize based off of the potential impact they're having in terms of the difference between actual versus expected, but also how similar they are to one another. So that's where we've really gone deep in terms of applying our large R&D resource that we've got available to develop something niche for the insurance market.
TIM ROURKE: From my perspective, and it wasn't that long when I was working in industry, being in an insurer is a difficult place. There's always external stresses coming into your business that are problematic, whether that be changes in regulation, whether that be changes in competitive behavior, whether that be changes in market conditions, inflation. All of these external stresses are something that insurers continually have to manage and understand. And again, if you can find problems very early, they're more manageable.
And for me, Radar Vision is a tool that allows insurers to do that in a much more controlled and governed way, in that you can spot problems early, using models as the sort of facilitator to do that. That's exactly the aim of this technology, that it gives you that sort of monitoring system of your entire business, spotting problems early, allowing you to remediate early, in a way that is in a lovely repository that's very visual and easy to use.
CHARLIE SAMOLCZYK: We talked about that it's not only the operational efficiency of managing the estate, but it's also I think that real competitive edge that if you if you do this well and you do it properly, you won't just be the same as everyone else. I think there could be that kind of assumed, you know what? If you just unleash the same tool, everybody's going to-- it's going to turn into this kind of homogeneous modeling estate. But I think that's not the case. This can be a differentiator from a competitive perspective.
TIM ROURKE: I 100% agree. If I look back at the last three years that motor insurers have faced and how difficult it's been and some of the things that have been missed, if you have a tool like Radar Vision, I think that those issues wouldn't have happened or those things would have been found a lot earlier. And back to your point as well, Charlie, this tool is not just there to provide everyone with the same facility.
Again, like Radar, how well you use it is going to drive how much value you get from the technology. And I think the same thing about Radar Vision. How you set it up, how you set the alerts, the monitors, and so on and so forth. All of those things will allow you to, as you say, spot the problems earlier. And, back to my point, that the insurance market is a tumultuous place and you're always having to keep an eye out on things. And using this tool to help you do that is very important.
I kind of liken it to, sorry to use this as an analogy, but it's quite an interesting one, the human immune system in that Radar Vision will be very good at allowing you to monitor things that you know, historical problems that you've seen and keeping an eye on them. And that's like if you catch a cold, you're not going to catch the same cold again. And Radar Vision will absolutely facilitate that for you.
But then Radar Vision will also allow you to spot things that you just didn't know were happening. So when you look at some of the horrible things that the industry has had to face from keyless entry theft to the emergence of electric vehicles being very expensive to repair and so on, and lots of other kind of examples that we can bring to life, I think Radar Vision will also spot those things that you just don't know are a problem, but it kind of highlights them as something that you need to look at and investigate.
And that for me, again, from someone who works in industry and was dealing with these external stressors, that's going to be a really powerful tool that will hopefully allow our pricing and underwriting clients to sleep a little bit easier at night.
CHARLIE SAMOLCZYK: So perhaps thinking not necessarily about the technology itself, but if you were an insurer and you're thinking about using some of these methodologies, what are some of the use cases that would probably come from a priority list? Where would you point this methodology?
TIM ROURKE: Yeah, that's a good question, Charlie. There's a couple of quick points, actually. One is that in my experience working with insurers across the globe, portfolio management, if it's done well, is such a key differentiator. It is the thing that allows you to remain successful for longer, to be able to ride out the underwriting cycle better. You've got a much more flatter kind of sign curve than if you don't have those kind of controls in place And really, really good portfolio management makes a massive difference. And I think that sometimes insurers can underestimate the effort and level of portfolio management that's required to be really, really long term successful.
With regards to some of the use cases, let's take health insurance. Again, an area that's very susceptible to inflation, an area that's very susceptible to new treatments, an area where you can see new diagnostics or new drugs that are coming into play. And those things can have quite a long term impact on a health insurer's profitability. Again, the impact of them can go under the radar.
So if you're looking at things like cancer and you've got models of cancer cost or cancer frequency, and the reality is your models are underpredicting in certain segments, then again, being able to spot that very quickly and understanding what that is and going and talking to your underwriters and your chief medical officers and saying, look, we've just seen some really weird things going on here. Can you bring this to life?
99 times out of 100, it will be recognized that there's an issue. And then you can, again, look at your pricing, look at your underwriting, look at your claims management pathways to manage those costs. That's really super helpful that even when your models aren't working particularly well anymore, it gives you really great insight into your business and what you might want to do in terms of other things to put the remediation activity in place.
CHARLIE SAMOLCZYK: I know you sometimes, Tim, talk about when models go wrong, that data is really valuable. So how models go wrong, where they go wrong. It's certainly valuable to know when models go right, but there's data in understanding when something's deviating from what it's supposed to be doing.
TIM ROURKE: 100%. This is something that is-- and I'm sure Pardeep can elaborate on this as well. But the fact that whether the models are going wrong segmentally or whether models are going wrong globally is really useful information. And having the ability to understand that at an early stage as well, so that you can glean all the insight from that information and then go in and put some changes in, that means that you can spot things, deal with things early on.
We talk a lot actually at WTW about the kind of war room mentality and looking to the future and trying to spot the problems down the line that could impact your business. And again, the insurers that do that very well tend to be more long-term successful. And I think having that great portfolio management and using the tools to maybe monitor things that you think could happen in the future. So if they do crystallize, you're in a much better position to deal with it. That's very powerful as well. These kind of portfolio management tools that we've been talking about, I think, are going to be very helpful in that sense.
CHARLIE SAMOLCZYK: So if you guys both worked for a fictitious insurer now and you had a tool like Radar Vision, what are some of the things you think that historically you would have found that we didn't find as an industry, you didn't find as your fictitious company, because that wasn't available?
PARDEEP BASSI: So the area I would focus on would be in claims. Not necessarily claims from a pricing perspective, but from an operational perspective. So if you had models which predicted the likelihood of a vehicle being a total loss, and behind that you had repair cost estimates, salvage costs, vehicle value costs, having all of those running in the background in terms of models informing you and giving you early insight, that could be hugely powerful.
So you could have Radar Vision running on your likelihood of a vehicle being total loss decision. And you could see that there's been a sudden increase or a decrease in the number of total losses, and the tool would automatically tell you it's because of higher value vehicles in a certain part of the country, and that it would allow you to go and try to understand that. Is it because of distribution issues with, say, electric vehicles? And I think it's that level of insight applied as soon as possible in the claim space where I would be applying a tool like Radar Vision.
TIM ROURKE: Yeah, I think that's a great example. Going back to health, actually, if you look at the environment pre and post COVID, there's been lots of changes in how health insurance is consumed. And again, having a tool like Radar Vision I think would have been very, very helpful to sort of understand what those changes were earlier on so that you can understand them better and you can change your prices, whether that be pricing, underwriting, claims, operations, control and governance.
Those things can be changed quicker and more targeted because you've got a better, more confident understanding of how the environment has changed post an event like COVID, for example, or some large piece of regulation. And again, going back to the motor, if you look at some of the regulation that we've seen in the UK specifically and elsewhere, things change. And, these tools will help understand that environment better.
PARDEEP BASSI: I don't think it's just looking backward to try to understand what's happened. It's to inform whether you're making the right assumptions going forward. So, if you take the recent drop in claim frequencies in motor in the UK market, if you're able to better understand what's driving that, say, in a multi-dimensional way, it's not just the weather, as some people are speculating, or because excesses are high and people are less likely to make a claim.
If you can understand those three or four different components driving that effect, you can start monitoring that going forward and saying, are each of those individual components developing in the right way? So, you can forecast future claims frequencies changes by actually understanding the underlying drivers rather than the high level effect that you're seeing. So, that allows you to make sure your prices in the future, 12 months down the line, have considered the actual underlying causes.
TIM ROURKE: And this is about confidence as well, Pardeep, you're right. If you see this stuff and you understand it and it sort of aligns with all the thinking within your business, your underwriters, you're more confident to take decisive action. One of the things that we have seen in insurance companies is the ability to be decisive and to be confident in your decision making can sometimes be difficult.
And I think that tools like this will just say that this all makes sense now. All of the components are adding up to something. We will make this decision with much more confidence than we would have done if we didn't have it. And that, again, when you look at some of the most successful insurance companies, that decisiveness makes all the difference.
CHARLIE SAMOLCZYK: Gentlemen, why don't we wrap it up there? It's been a really interesting conversation. Pardeep, thank you so much for your time today. It was great to have you on, and I can't wait to hear and watch as Vision launches and see how it unfolds in the market.
PARDEEP BASSI: Thank you, Charlie, Tim. Great conversation.
CHARLIE SAMOLCZYK: Tim, likewise. Thank you for joining. Really appreciate your time and your insight and your perspective on kind of the impact to the insurers that you're working with.
TIM ROURKE: Thanks, Charlie. Really, really good to speak to you both. Thank you.
CHARLIE SAMOLCZYK: A big thank you as well to everyone listening to Talking Technology. I certainly enjoy doing these and I hope you find them valuable and look out for the next episode.
SPEAKER: 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 Insight section of wtwco.com. This podcast is for general discussion and/or information only, is not intended to be relied upon, and action based on or in connection with anything contained herein should not be taken without first obtaining specific advice from a suitably qualified professional.
