Winning Executive Buy-In For Scaling AI in Insurance - with Ermir Qeli of Swiss Re

Foreign. Welcome everyone to the AI in Business podcast. I'm Matthew d', Mello, senior editor here at Emerge Technology Research. Today's guest is Irmir Kelly, head of Data Science and AI at Swiss Re returns to the program to discuss strategies for gaining executive buy in and scaling AI within the insurance industry. Throughout our discussion, Irmere highlights the role of data infrastructure, particularly for legacy institutions, in ensuring that data is clean, structured and distributed effectively across the enterprise. He also delves into how change management is crucial for driving AI adoption, emphasizing the need for collaboration from day one, upskilling teams and building trust in the technology from the beginning. Without further ado, here's our conversation. Irmir, thanks so much for being back on the program. Thank you for having me here. Absolutely. So we were talking in our last episode about kind of the large question for insurance leaders. You know, where do we start with AI? Especially tough question to answer in the insurance space for a lot of the reasons you mentioned in the last episode, changes, challenges surrounding data stacks and ways of at least driving that originally unstructured data. As you explained in that last episode, wondering if maybe the best way to kind of ask this question is now that we know Gen AI has been with us at least for a year and a half, it feels like especially through and through the culture, my parents know this stuff. There's no shortage of educational resources available to insurance leaders. But I'm wondering just from your perspective, especially coming more from the data science side of the business, you know, your background in retail banking, where should they focus their resources in terms of knowing what's best for the organization and driving those adoption workflows, Especially in the beginning phases, when you have a small team that's engaging in a larger scale, what will one day be a larger scale digital transformation. Thanks for the question. It's quite a question actually. And it's interesting because if I were sitting on this podcast a few years back, maybe actually the answer would have been a bit different. But especially if we look now with with all the advances in Gen AI and ChatGPT, I mean now my answer is much more clear and building a bit on what I mentioned in the last episode. So insurance and reinsurance tend to use a lot of unstructured data and if you look at from the perspective of of the industry, the pro, our product at the end of the of the assembly line is a contract which underpins a promise that we give to our clients in case things go wrong. We're going to pay, we're going to support them. Now the question though is for producing that contract. There's a lot of work that is going in there. And of course, there is also potentially a lot of questions that you might want to ask to this contract to understand what's the kind of coverage. So also from a retail perspective. So whenever I have a question whether I'm covered or insured on my homeowner's insurance, I have to go and read the document. So this is the kind of capability now that Genai is offering almost at our fingertips where you can basically query your, your documents at large and more in particular contracts and answer questions based on the content that these documents provide. And if we would look at that question probably two, three years back, this would be something very hard to solve. So you'd need a whole data science project to be able to answer just a simple question. For example, do we have war exclusion included in a contract? For example, from our insurance perspective. And this is something that now is capable to a large extent with large language model capabilities. So in similar spirit, I mean, the same capabilities can eventually help also to analyze claims at scale. So claims also involves a lot of unstructured data. So typically we'll start with either a phone call, an email, some other information that you'll send to your insurer. And of course, everyone is interested of analyzing this data as fast as possible because it means better customer service. And now we do have those models actually that provide you monthly model capabilities to analyze all this data at once. It's a really remarkable time. And I really appreciate kind of the layers that you brought in the example from homeowners insurance. But even like I think people can very easily imagine, you know, whether it's whether they're executive leaders listening to the show Insurance and beyond, or they're just everyday listeners, just consumers out there when they hear, taking that, that examination of that end product you mentioned, the contract, they're probably imagining. Not that I think the death of lawyers has drastically overblown as a rumor, to quote Mark Twain there. But I think they're what they're envisioning in terms of. Your last answer is, okay, well, I have a new homeowner's insurance contract and I'll be able to use an LLM to maybe instead of, I'll of course have a lawyer, but I'll have an LLM scan this document not only just for insights from a legal protection standpoint, but also insights for being a customer for staying advanced on policy or staying up to date on how your bills will change. So I'm getting a sense of that the depth of where they'll be able to take their perspective from these end products goes a lot deeper than just legal and compliance workflows. It goes into the entire customer experience. Where do we see that playing out? Just with what we're seeing also on the call center side, with building these customer profiles and the overall push throughout the industry to better know their customers, I think which we've seen reflected from our brothers and sisters in the. In larger financial services, of course. Absolutely. Indeed, the most popular use cases of large language model in financial services at large and insurance in specific are around better customer experience, a better call center experience, better chatbot experience. I mean, we all recall our chatbot experiences prior to ChatGPT. The answers seemed rigid. You had the impression actually you spend more time than if you had, if you did the stuff yourself. And ultimately they did not work so well. Now come chatgpt, most of that interaction seems seamless. The answers, of course, have to be controlled and have to be grounded to the data that an insurance company provides. But in a nutshell, that process has become simpler. And that of course will also change the way how the retail insurers interact with their customers. Yeah, I think there's a lot of broad implications there, maybe even a better way to kind of narrow down what AI is going to mean really in this space is we know at least from, and you come from that background in retail banking, which we talked a little bit about in the, in the first episode. We know from our friends in larger financial services banking, especially that customer experience, especially in the call center, is a great way to scale up AI. Does it have that same potential here in insurance as a great way to scale up, or is there more appropriate ways? Do you see stronger first use cases, front doors for that effort? I think it does. I think it does. And it connects on both parts. I mean, both on distribution side of things when basically when we're selling insurance products and typically that's a, that's a process that, that is, that has its particular sides, but also in the claims management, which are the two, the two parts actually, where the customer interaction is very important. This is of course more a topic for primary insurers, but is also as relevant for us because primary insurers are our clients. Absolutely. And I mean, we get into a lot of differences. I know we've, we're starting to separate, even in terms of our convers about data, the difference between, you know, retail insurance, commercial insurance, et cetera. But even, even just taking a broader step back and thinking about AI advice that applies to all kinds of insurance organizations. There's a lot of talk around a phrase that ends up getting called change management, as I've always understood the term and how we've mostly used it on the show. It sort of embodies everything around maybe a digital transformation or really just any change within the organization. If you're going through a merger or an acquisition, you know, any kind of large scale change for the organization, how are you managing that from an HR standpoint, personnel standpoint, strategy standpoint, how you're doing business, how you're going to keep services going for customers? Where does this discipline, this way of thinking, kind of take a different tenor in the AI conversation? And what should insurance leaders keep in mind when change management gets brought up in the organization hand in hand with digital transformations? A common theme that we see, and I guess this is not necessarily AI specific, but it's also relevant in context of AI that typically to build a prototype of solving a problem is easy. Also, maybe to run a pilot with a selected group of users might work well by being able to scale out and basically capture the value there. Where it is, namely all the people that should use a certain functionality is always a difficult problem because change management is hard by nature. Question is, why should you change a specific process? Why should you do things in a different way? And I think our task, both from a data function, but actually I guess across the businesses, is to kind of enter this journey together with our, with our business stakeholders and involve them at day zero, so to say, and run that change management together. And part of that includes upskilling, gaining trust in the technology because everything that is new is difficult to understand how it works. And I have a very good personal example of that. I recall when ChatGPT was released initially I was in denial for a few days, you know, because I couldn't believe how good the results were. And of course that's a point. We're well passed over it and including myself, I'm using that very broadly now. And if you take that and amplify it with a large number of users that should use the technology, you could argue then of course, to get the critical mass so that you get the right business impact. Right? This is actually the process that the firms need to look into so that basically they achieve the necessary adoption. Because let's not forget one thing. It doesn't matter how great the models are, if they're not used for the purpose they were intended to, they won't bring the necessary business impact. I think that's a lesson we're learning across the board and that goes into data cleanliness, data structure as well. If you don't have very well structured data, no matter how good your systems are, no matter how good your computer is, no matter how good your infrastructure is, it's not going to bring you the results that you want. Results is a very multilayered execution in a plan across the organization. And I think even how you've illustrated that in your last few answers, I think brings a lot of clarity for folks just in a sense of change management and really driving the scale. I'm wondering also what metrics really become important to really drive those conversations. Make it clear, especially to the management side, that as. As existential as ROI conversations can be around these technologies that they're. You're building systems that are serving core business goals. How are you proving that on the metric side? I guess the holy grail there is actually business impact. You have to achieve business impact. And the tricky bit though is sometimes it's difficult to project the future business impact of something that is new and is being introduced new in a process or in a technology. And of course you can start making, making projections, but those projections will depend on a lot of assumptions. Right. One thing that we have realized works well in practice is actually to look for proxies for business impact that will basically capture some of that value. And adoption tends to be one of them. Because if the technology is not adopted then it's not going to be that bring the necessary business impact. And adoption doesn't mean actually that you have to scale things to thousands of people at once. It can be also a handful of people, but the people that are running a business process and that that ultimately bring the necessary impact. Absolutely. I think as that conversation becomes longer, I know you see those metrics really change, but business impact I think ends up being an umbrella for a lot of con. A lot of the broader conversation and especially a couple of the metrics we talked about in the last episode that become important as that data becomes structured as you're realizing new visions. Like the example you talked about in the last episode about parametric insurance and taking data from the larger environment to bring a greater sense of responsibility and context to the prominent example was auto collision. I think especially to get there, it's going to be not just a change within the organization, completely self contained within that adoption team, but also a broader conversation with the organization. We see that a lot just under the umbrella of change management. So we appreciate you coming on the show and talking about it. Amir, these last two episodes, really, really great insight. I think our audience is getting a lot out of it. Thank you so much for being with us. Thank you for having me here. Don't forget to check out our previous episode with irmir that's from June 11th of this year, titled Lessons from Retail Banking on Solutions for Structuring Insurance Data. Ermir joins us to talk about the array of challenges facing insurance leaders at this stage of AI adoption across the sector and how the pace for which is marked by the fundamentally unstructured data at the heart of insurance tech stacks. Throughout the episode, Amir offers his perspective on scaling systems dealing with these and similar issues, offering numerous use cases in generative AI and beyond to assist in these processes along the way. If you enjoyed or benefited from any of the insights of today's episode, consider leaving us a review on Apple Podcasts and let us know what you learned, found helpful, or liked the most about the show. Also, don't forget to follow us on X, formerly known as Twitter Merge that's spelled E M E R J as well as our LinkedIn page. On behalf of Daniel Figella, our CEO and head of Research, as well as the rest of the team here at Emerge Technology Research, thanks so much for joining us today and we'll catch you next time on the AI and Business podcast. Sam.