The future of insurance and analytics

RAVI SHARMA: We also have Laura Doddington, the head of personal lines in WTW's Americas Insurance Consulting and Technology practice. Welcome, Laura.

LAURA DODDINGTON: Hi, Ravi. Thanks for having me today.

RAVI SHARMA: Well, it's great having you both on the podcast. I like to start things off by learning a little bit about our guests. Pardeep, I'll start with you. What's your favorite sport to spectate? And your favorite team in that sport?

PARDEEP BASSI: I like a range of sport. And it's a tough question. I'm currently watching a lot of tennis. Wimbledon is on here in the UK.

RAVI SHARMA: Oh, it's exciting, watching the tournament. So it sounds fun to me. All right, what about you, Laura?

LAURA DODDINGTON: So I think I have a couple of answers, Ravi. Big Formula One fan. So big Mercedes and Lewis Hamilton fan. But I've been in North America for almost 10 years now. And so I feel like I've grown to love basketball. And I love watching it. And the Raptors are my local team. So basketball and the Raptors

RAVI SHARMA: Well, the Raptors can certainly be exciting to watch as well.

LAURA DODDINGTON: Yeah.

RAVI SHARMA: Well, now that we have that out of the way, let's jump right into discussing the future. I'm going to start this off with a general question. And maybe, Pardeep, you can start. What are the trends that we're seeing within the industry when it comes to advancing analytical capabilities at insurers?

PARDEEP BASSI: One of the biggest trends that I'm seeing is data science and analytics is spreading across the whole insurance value chain. So traditionally, there's been a big focus in the pricing and underwriting space to apply advanced analytics, data science, machine learning, any of those terms which are used. We're now seeing that spread across every single decision and influencing as many decisions as possible. Hundreds if not thousands of predictive models influencing all key parts of the insurance value chain.

RAVI SHARMA: Great. Laura, are you seeing anything that you'd like to add to that?

LAURA DODDINGTON: Yeah, so I mean, I totally agree with what Pardeep was saying there. And then in addition to that, I would say there's an increasing focus on how do we make that analytics more efficient? How do we make it more effective?

So how do we ensure that we're able to apply analytics at scale, for example, and in real time, and getting it to the point of sale, or the point of quote, or wherever it is that we need the analytics, and also an increasing focus on the governance around it. So starting to recognize that actually analytics brings with it new risks and things we haven't thought of before. And we need to have the right level of governance around that.

RAVI SHARMA: That framework that you both have laid out of really deploying analytics across the entire organization with increased governance and increased sophistication. Undoubtedly, you're running into a lot of roadblocks and hiccups when you're working with your clients on trying to implement these solutions.

So I guess the next question I would ask you both is, when working with insurers across the industry, what are some of the roadblocks that the organizations are facing that are preventing them from advancing their analytical functions?

LAURA DODDINGTON: From my perspective, I think, having actually recently moved from a carrier into WTW. And I think one of the challenges is thinking, there's so much analytics you could do. There's so many different problems that we could try and tackle with analytics. What should we try and address? And is it solving a real business problem?

So it's nice to go build a really cool fun model, but is it really solving the business problem that's going to drive value for the organization? And so I think organizations are trying to think about, how can we ensure that we're tackling the most important problems with our analytics? And that we're then able to deploy that effectively in the organization.

And that can be really difficult to do when there are so many different things that you could be modeling. And if you've got silos and your actuarial team or your data science team, for example, are not connected to other parts of your organization, it makes it really challenging to actually embed it in the business. And so I think organizations are grappling with, well, how do we actually start to make this be part of our organization's DNA rather than something we do on the side?

RAVI SHARMA: And what about you, Pardeep?

PARDEEP BASSI: So building upon the focus on the business. Insurance is quite a unique business. And being able to understand the constraints, the regulatory environment, and the insurance specific problems, that plays a key component in being successful in this space.

So not only do you need to have a deep understanding of insurance and where to apply the techniques, which techniques are most appropriate. You need to have an understanding of the advanced techniques. You need to have the right technology to support you, whether that's proprietary software, leveraging cloud compute, or open source, and a combination of those all together.

But you need the right culture, which allows you to bring these different skill sets and knowledges, which usually sit across different individuals and different teams. How do you bring all of that together to actually have that transformative impact, which the promises of data science and machine learning across the whole of the insurance piece.

LAURA DODDINGTON: And I think one of the keys with that is, how do you make it more transparent? There can be a risk-- as analytics gets more and more complex, there's a risk that it feels like a black box. And so now if I'm sitting as a business leader, for example, I really don't understand what this black box is doing, and whether I can trust it, and how it's going to impact my customers or my teams. And so I think that's a challenge in organizations.

And it then becomes important that our actuaries, our data scientists are doing everything they can to open up that black box and show transparency. So they can get buy in internally, but also increasingly externally so that regulators, for example, can understand what it is that these potential black boxes are really doing and what the impacts are. So that we don't end up with, for example, unfair discrimination within a model because it's sitting within a black box.

RAVI SHARMA: I can see that need for the increased governance that you mentioned in the prior discussion here with the increased regulatory filings that we're seeing, specifically, in the North American market as insurers are getting more sophisticated and the products are getting more sophisticated.

And the need for understanding what's supporting a model, going into the model, and how it's being deployed within the insurance product is, like you said, external parties are now more interested than ever in that. Another curious item I'd ask you both is, the demand for analytics comes from very siloed areas of an organization.

And so how are insurers overcoming-- underwriting is asking for this specific data set or type of analytics where IT might not be there ready to support that. And they don't see the value of it, but underwriting is still asking for something that would be very valuable to them. And so that's a roadblock. I'm just curious how you're seeing insurers overcome that.

PARDEEP BASSI: The biggest thing here is the ability to collaborate and share your expectations, wants, and needs and being able to translate that. So an underwriter may have certain needs, but they need to work with the right people, whether it's in the IT team or a data scientist to translate that into language which is common across all three.

So you can understand how the different departments, individuals, how they can work together, because it is that combination of all of those separated components, which will give you the ultimate solution which will work. You can't have your underwriter pretending to fully understand the IT requirements. And your IT team will never understand what an underwriter is looking to do. So it's being able to work together, which unlocks the value.

LAURA DODDINGTON: Yeah, and I think part of working together is also about having joined up objectives. So the objective isn't data science are going to build an amazing model. Underwriting are going to apply it. IT are going to ensure that they have the appropriate governance around it. It's actually, for all of those teams, they have one single objective, which is that we need to, whatever it might be, we need to have a better client experience, with more straight through processing for our underwriting applied in a safe and secure way within our IT environment. But they're all working towards that, as opposed to each person feeling like they're just working towards their bit of that.

PARDEEP BASSI: Another area where we see this separation or difference in understanding really stand out at the moment, it's around the risk governance and ethics. So the current insurance risk framework which we're seeing insurers adopt and successfully use for many years. We're seeing that no longer being completely appropriate for this new world of open source, data science techniques. There are existing risks which become more prevalent, but there's also new additional risks which those who are individually and responsible, they may not have that complete understanding. So the two different groups here are your practitioners who are using the latest techniques. And those who understand the insurance risks.

LAURA DODDINGTON: If you think about, say, a second and third line risk team for example within an organization, they have probably, across that team, hundreds of years of experience in understanding underwriting, understanding claims processes, understanding pricing. And so they're really well equipped to be able to talk to those teams and appropriately understand the level of risk there and appropriately challenge, which is their role to do within the organization. Within data science, if you don't have people who have that data science experience within second and third line, they may not be so well equipped to be able to provide the appropriate level of challenge that they should be in that organization.

PARDEEP BASSI: Another example here is open source. So there's a whole new movement. It's not new. This has been going on for 30 or 40 years. It's just growing exponentially at the moment. Where there's huge advantage of being able to innovate the flexibility, the speed which open source offers. But there's a whole range of challenges which need to be understood and mitigated. So of those challenges, you've got an increased risk of malicious code being introduced to key decisions at key decision points within your organization. So the whole governance, security piece, the maintenance and support of your models, where you have potentially key person risks not being able to understand issues, concerns with the performance of your model and then the stability of it as well.

There's huge challenges in adopting open source. And we're spending a lot of our time, in particular WTW, to understand how we can combine the benefits of open source with our existing Radar functionality, which gives you that complete security, stability, and that long history of successfully providing this service to insurers. So bringing open source into that secure governed manner is a big, big focus for us at the moment.

RAVI SHARMA: I'll never forget when I was working with an insurer and there was a process that we were doing that you had to delete the files in a folder. And so one of the analysts in the department thought it would be a great idea to write this macro. So they didn't have to go in and manually delete these files.

Well, they wrote a macro that would just delete anything that was in a folder. So if you put that macro in a parent directory, it could have obliterated a whole network drive. So just the governance and integrating the processes with the appropriate controls is just so important. And sometimes, the people that are making these decisions and writing great code, sometimes they don't realize the risks they're taking on.

So carrying on the same thing, I want to ask both of you, how are you seeing insurers starting to address some of these challenges? I know we've already jumped into this a little bit, but I want to expand this a little bit more and really ask you specifically, where have you seen it done well? And what sets those organizations who are doing this change well apart?

LAURA DODDINGTON: So Ravi, I think there are a few things. I think tools are obviously really important. Critical that you have the right tools in your organization that allow you to have the appropriate governance, the appropriate ability to deploy so that your actuaries, your data scientists, your analysts are able to spend their time thinking about the areas where they can add the most value.

And where they are able to, for example, build models that are solving real business problems. And they don't, then have to worry about exactly how are we going to deploy this because they've got best in class tools that help them to do those things. So tools are really important.

And processes and structures are really important. So actually, how are you making decisions about what to do next? How are you structuring your teams in a way that, as much as there will always be silos, you're trying to break those silos down as you think about the governance that you put in place, the structures that you put in place so that the teams are able to work much more effectively together. But I think the third component and sometimes this is not given as much attention as it should be is culture. And culture is really critical in all of this. To have a culture that is embracing analytics, looking for ways to really use this effectively in the organization is so important so that you have that buy in throughout.

PARDEEP BASSI: For me, the one that really stands out is the focus on business value. So rather than building theoretical models, using the latest techniques, it's understanding where you should be focusing your attention and using the most appropriate technique for the problem you're solving. So you may even want to compromise the predictive power of certain models and approaches that you use, if it means the realization of value increases.

So for one small example being, from my own personal experience, you go from models which are more interpretable and potentially slightly less predictive just so that user gets the confidence that the model is working in the way that they want and expect it to be.

And then you can move on to the more predictive, more sophisticated models at a later date. So I think it's business value focused first from everything, from the technique, your approach, the engagement from the data scientists, analytical community into the domain experts who have had decades of experience in insurance.

And having that throughout the whole process of, where is the value? How do we do this in the most efficient way? It's not just a personal pet hobby where we want to do everything, writing our own code. We don't want to use any other software. It's business and efficiency focused throughout. And that even goes into the governance, and risk, and ethical perspective.

So rather than having a mindset of we'll build these analytical models and we'll deal with the governance and risk and ethical components afterwards, having the thinking of what does this mean for the business given the regulatory environment, given the ambitions, and the goals, and the strategic nature of the business. How can we introduce the considerations by design, even into your model building, your data analysis phase of the predictive modeling life cycle.

So governance, ethics, risk by design fully integrated throughout is another example of putting the business needs first and having that really, strong, strong focus of, is this adding the value rather than this is interesting for me to work on?

LAURA DODDINGTON: And I think, Pardeep, to add to your point around-- that actually it's OK if the model is perhaps less than the most predictive one if you can deploy it is more valuable. I think to add to that I would say, there can also be a temptation for us actuaries, and data scientists, and people to really target perfection.

I want to build the perfect model. And actually, it would be better to build something that moves us in the right direction, that we can do quickly, and then build on that. And then make it better and have a test and learn framework, where we are constantly trying to evolve things as opposed to, I can't do anything until it's perfect. And so I think, thinking about, when is something good enough that actually you can start to test it and see what it does rather than always aiming for perfection is important.

PARDEEP BASSI: And it's not just about putting models live. It's once they are live, how do you ensure they continue to perform, your ability to actively manage your portfolio, the management of uncertainty, complexity management. These are all the things that you need to think about by design throughout the whole process, rather than we have a model. It's live. We'll do all those issues afterwards.

And it's that full understanding of where that business value is. And like you said, Laura, it's not about just getting a model live which is very sophisticated. You may have more complex models which require more effort at a later date to maintain. And it's that thinking throughout which adds the real value.

RAVI SHARMA: Well, this has been great. I think we've covered a lot of ground today. But one thing I like to do is just recap this for our listeners. So Laura, I'm going to start with you. 30 seconds, if you could give our listeners one take away from today's discussion, what would it be?

LAURA DODDINGTON: So I am really excited about the future of analytics in insurance. I think we're at an exciting time where it's rapidly evolving. There is more access to data than there ever has been, more access to computing power than there ever has been. There's so much potential.

The message I would leave to our listeners is to really think about, how are you going to maximize the value of that? And maximizing the value doesn't mean, I'm going to build like the absolute, coolest machine learning model that has ever been built. It's all the things that go around that.

So be really thoughtful about how you are deploying, how you are choosing what to build so that you're adding the most value to your organization. Always start with that focus on, what am I doing this for? And how can I do that to maximize value?

RAVI SHARMA: What about you, Pardeep? What would you like to leave our listeners with?

PARDEEP BASSI: For me, it would be, if you were 10 years in the future and you were looking back, would you feel comfortable that you're doing enough, enough to firstly survive, compete, and then ultimately win? Because insurance and analytics are so tied together that I don't think you can think of, we're an insurance company which does analytics. It's almost we are an analytical-driven insurance company.

RAVI SHARMA: Well, Thank you both for joining us today. Laura, thank you so much.

LAURA DODDINGTON: Thank you for having me, Ravi. I always love talking about analytics. So thanks to you and thanks to our listeners.

RAVI SHARMA: Of course. And Pardeep, thank you as well.

PARDEEP BASSI: Thanks for having me, Ravi. It's been a pleasure.

RAVI SHARMA: As always, we'd like to thank our listeners for joining us. For Laura, Pardeep, myself, we look forward to seeing you next time on Rethinking Insurance.

NARRATOR: 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.