AG Insurance CEO Heidi Delobelle on using AI to serve customers faster

Tired of hearing about "AI transformation" without seeing a clear path to implementation? This discussion takes a different turn. It sets the essential groundwork that makes AI truly effective in insurance. We engage with Heidi Delobelle, CEO of AG Insurance, Belgium's oldest and leading insurer, to explore how AI has been integrated into daily operations while remaining reliable, secure, and transparent. 

We focus on where the real value materialises. Heidi reveals that a successful AI strategy goes beyond cost savings to enhance customer experience, streamline distribution, and bolster employee support. We explore hybrid service models in which digital channels manage routine tasks, leaving humans to handle critical interactions. We also delve into how AI and data are reshaping call centres, moving past "press 1" menus to provide agents with real customer insights, enabling quicker and higher-quality resolutions. 

Next, we examine the foundations and risks of AI. Clean data, a unified customer view, a shared sandbox, and central standards prevent chaotic tech and help mitigate AI risks. Heidi highlights the importance of governance, cybersecurity, bias checks, and human oversight for significant decisions and external communications. You'll also hear about a specific use case where AI reduces SME underwriting research from days to minutes, freeing up capacity for growth, along with a five-year perspective on claims automation and prevention-focused insurance. 

If this conversation has sparked new ideas on applying generative AI, data governance, and responsible AI in your organisation, subscribe, share the episode and leave a review. 

Join us and listen to all episodes on www.pwc.be/aiunscripted

Transcript

Welcome To AI Unscripted

SPEAKER_00 0:00

Welcome to another episode of our AI unscripted series. The series that, as you know, focuses on AI but with a slightly different uh angle because we're trying to identify opportunities in terms of what you can do with it. What is the experience of the industry of thought leaders with AI? And today I'm going to have a discussion with uh Heidi Dolobel of AG Insurance. Heidi, welcome.

SPEAKER_01 0:31

Hello.

SPEAKER_00 0:36

Well, I think many people, if not all of the people that will listen in, will know AG Insurance, but maybe for the people that don't know AG, maybe a slot a short introduction of what are you doing? You're

AG Insurance And An AI Mindset

SPEAKER_00 0:49

an insurance company, of course. Yes, yes.

SPEAKER_01 0:51

So AG Insurance is uh the oldest insurance company in Belgium. Uh was created in 1824. So we celebrated our 200th anniversary in the United States. Older than Belgium. Yeah, we are older than Belgium, and we are market leader of Belgium. And we have a market share of 28% in life and 70% in non-life.

SPEAKER_00 1:15

Okay. Does that mean that you are you're looking at AI also already for a number of years? Because of course a lot of that you do is data, data-driven. Yeah. Is it top of the agenda?

SPEAKER_01 1:28

It is already top of mind since a couple of years. Like it was always the case in the past. We were uh one of the first companies that started an automation project, but long long ago. And so for me, AI it's another part in the technology that has to be embedded in our functioning, and uh yeah, it's uh evolving very fast, and uh we are very interested to experiment and to see what are the opportunities by measuring the risk as well, of course.

SPEAKER_00 2:08

And

Value Added Beats Cost Cutting

SPEAKER_00 2:09

how do you deal with that? Because we are recording it now in the end of January, the time that we're also at uh Davos uh economic forum. And there we are also launching our CEO survey, the global survey that we are doing, where we basically sound globally CEOs. And the sentiment is is is biased, because um the majority of the CEOs are indicating that they don't see the savings uh yet. Uh is that the same that you see here? But is it also the benchmark that we want to focus on? Is it really a savings discussion or is it much more a value-added discussion?

SPEAKER_01 2:45

Uh I think, first of all, it's a value-added discussion. I think AI can support us in many areas. Of course, it will help us to be to be more efficient, but it will also help us to improve the client experience, the distribution convenience, also the well-being of our employees. So all domains where AI can play an important role.

SPEAKER_00 3:13

And do you have different teams between of those AI projects that are focusing more on efficiency and then more on the indeed user experience, the market as such, or is it everything?

SPEAKER_01 3:24

No, no separate teams. No. It's the same. Yeah, it's the same.

SPEAKER_00 3:28

Okay. And how

Fear Change And Workforce Upskilling

SPEAKER_00 3:29

do you bring that message across then internally that okay, AI is important? Yes, as you say, for the two, but that it's also not a threat because that is something that we do see that people are starting to get a bit nervous, also in terms of their own future, which of course is fundamental. Um, that we yeah bring that into context. Is definitely for knowledge workers, and of course, you're also a service industry, so that's also the same thing uh that you are facing.

SPEAKER_01 3:55

So how of course there is some uh yeah, it's unknown for the people. They feel that it can have an important impact on their rules on their functioning. Also the unions are concerned about uh the fact that maybe sam bros will disappear. So I always stress that it is really important to stay competitively we have to embed AI. It's no option. Uh, but uh they don't have to see it as a threat but more as an opportunity. Certainly today. Uh I think co-pilot is a good world. That's a bit the way we use uh AI for the moment to help the employees to better serve the clients or the distribution. So, and it will evolve, it will take more and more decisions. But for me, it's very important to stress that we have to do it in a controlled way. Uh, we have uh to be trusted by the uh the clients. So it's very important because you're using control what we produce, and it's clear it will change their functions, but it will create also opportunities. Uh they will be able to serve better their clients. So it's not only about, like we said, efficiency, but it's better serving faster and in a better quality. So they don't have to be afraid. We have to do, we have to adapt, we have to upscale our workforce, uh, and also reflect on a strategic workforce planning for the future because yeah, the skills will evolve, and if you anticipate, then we don't have to worry about jobs of tomorrow.

Hybrid Digital Service With Human Touch

SPEAKER_00 6:00

And if you look a bit at the a different sector, but it's uh similar in terms of dynamic as a banking sector, where of course they are also heavily investing in all kinds of apps to indeed change the user experience. Is that something that you also see in the insurance uh area?

SPEAKER_01 6:16

Uh we see it evolving in the insurance sector, uh, but uh it depends a bit uh about what kind of uh insurance company you are. For example, at EG, we are not a direct insurance company, so we are working with uh BMP Paribas 40s and with brokers. So in the beginning we didn't have a lot of direct contract uh contact with the end customer, so there it was more the interaction with the distribution. But even in a model like we have, we see that end clients are coming towards the insurer. That doesn't mean that broker or bank doesn't fulfill their uh job. But yeah, it's natural uh if they have a claim and they they look for the easiest way, and then we saw uh and it was maybe also a bit efficiency driven, but yeah, when at a certain moment in time we saw the increase of incoming calls of the end customers, then we had to work on our efficiency, and that's the the moment we started to create first a website, now an app to enable uh clients to be in direct contact with us. But for us, our approach stays uh an hybrid approach and digital approach. So for the easy tasks, we will be digital via an app via website, whatever. But uh if it really matters for the client, it's important to still be accessible with human touch is still important. The human touch stays important, yes. But of course,

Call Centers Powered By Customer Context

SPEAKER_01 8:14

then AI can and AI can help a lot, and what we see, for example, in the call centers before, and we have a lot of data as well. So if you combine AI and uh data, you can serve the client better. What we saw before in the call center was yeah, we we were working with an IVR. If you're Dutch, then uh press one, press one, etc. etc. But now what we see, we know the client, we know the numbers, so we know uh what is the language he's speaking, but we see as well and we give this information as well to our employees all the interactions that the client has with the company. Was he searching on our website? Has he declared a claim? Uh did he call us all the inter all the transactions or interactions or on the slide or on the pop-up on their screen? So they know already in advance, more or less, why the client is calling, and so it's helping them to serve better the client. Uh so yes, with all the data we have combined with uh AI, we can really deliver a better service to the client.

Cleaning Data For One Customer View

SPEAKER_00 9:40

Which must be a massive investment, also, I guess.

SPEAKER_01 9:43

Yeah, but uh I think we have the advantage that we invested already a lot in the structuring of the data before. Otherwise, it would almost not be possible, uh the cleaning of the data as well. So if you uh know a bit the history of AG, we have three different business lines coming from other companies after uh mergers and acquisitions, other systems, and so we put a lot of effort in the past to first build a unique customer database and do a cleaning of the quality of the data to be sure that one client in one business line is the same client in another business line. So the unique customer view is very important. It's only at that moment that you can see what are the interactions with the client, otherwise it's garbage and garbage hands. So yeah, yeah. So and that yeah, I think that project to build this unique customer database took us three years from creation and cleaning before we could say that we have a clear view on the unique customer. Because in the Belgian legislation, insurance companies are not allowed to use the social security number. That's also a pity. But uh the foundation is also very important.

SPEAKER_00 11:24

Yeah, and do you have a team specifically focusing on AI or is it indeed embedded in the interesting team?

SPEAKER_01 11:31

Uh

AI Sandbox Standards And Shared Community

SPEAKER_01 11:32

we have uh a central theme for AI and also in the business line. Uh so it's very important that uh people share the ideas that they have. If you want to foster innovation, then you have to share the ideas. But there is a central theme as well to see that uh to structure everything a bit because uh yeah, the technology is evolving very fast already. But if every department is using its own uh uh applications, then it will become a spaghetti. So for us it's important that there is one team that has the overview has the standards as well. Define the standards. Uh they provide as well an uh a prairie um uh a playground. I don't know how they name it, uh, but it's an isolated uh environment where people can experiment with a sandbox. I was Jimmy. A sandbox where they can experiment and it's important to build it once, yes, and every can everyone because as you said, you are the biggest, but of course that also means that you need to be able to use the scale that you have.

SPEAKER_00 13:00

Because that's what smaller players struggling a bit, I think, because of course with AI, but you know what the philosophy is, but the tooling will evolve over time, and that is very quickly.

SPEAKER_01 13:12

Yeah, that's very quickly. So you need to. And that's something we see already. Yeah, if one business line starts with an initiative and another business line wants to do the same, there is already a chance that technology evolves and that they want to do it in another technology, or that an application uh package that they use provide already the same functionality or the same features. So it's evolving so fast that we have to have a central team to structure it and uh to keep us in the right direction because yeah, uh AI is creating opportunities but is bringing new risk as well. And so our AI uh community, we have even a community to federate all people that are working with AI, but they are in close contact with the risk-based governance bodies. Because cyber security is also very important and also very costly. Um, but so important, and it will not disappear as well. Testing the data quality. Uh remove bias in the kalibration all things. We have our guardrails, we have our policies. We have to obligate uh legislation as well. So, but it's important, it's new uh opportunities.

SPEAKER_00 14:51

But also for your reputation, huh?

SPEAKER_01 14:53

Certainly we we stay responsible, of course. And so that's why, in the first uh step we are more working on AI in a controlled environment where there is still a human validation, certainly for material decisions, because it's our commitment towards the external world. So if AI is used with an interaction with the external world, certainly if there are legal aspects in it, it will be controlled by a human.

SPEAKER_00 15:30

Which of course

Guardrails Ethics And Human Validation

SPEAKER_00 15:31

brings us to the question of the ethical dimension in a sense that definitely for an insurance company, I guess, are we going to use AI in an inclusive manner or exclusive manner? In other words, do we do do you feel that well for potential customers or customers, AI is really beneficial because the more data that you have, maybe the more people that you will be able to cover, because you have a number of data points that allow them to be covered. Whereas otherwise you might say, based on risk-based approach, well, you don't meet all those tests. But maybe, yeah, it is an insurable risk because you have data that you d otherwise wouldn't have looked at. That is, of course, for your sector, is very touchy, I guess, because you yeah, you can turn everything in in multiple directions.

SPEAKER_01 16:18

Yeah. So but uh in insurance we need a certain solidarity, mutualization of the risk, which is a big thing. Otherwise, uh insurance will disappear. So you can be more precise. Uh but uh if you want to do a segmentation in your pricing, the regulation obliges us to be transparent on the segmentation. You have to understand and to be able to explain why you do certain segmentation. So that's also important to keep in mind. So for the pricing, uh, there is enough enough regulation that there still is uh a certain mutualization of the risk. But I think for the underwriting, you have a lot more information uh where the quality of the underwriting is uh is better if you use the external and internal data.

SME Underwriting In Minutes Not Days

SPEAKER_01 17:23

And what we see, and there AI is uh coming into play as well for the underwriting of SMEs, for example, and non-life insurance. We need a lot of information because pricing is even less important than be selective in the underwriting. And then uh uh yeah, I see before the people that were underwriting SME they took a lot of time to gather all the data and they had to have look in legal data, but also all the reports that are available for uh for these SMEs, etc. It took I think two days uh to come with a decent report about all the risks linked to the SME today with AI, yeah. A couple of minutes a quarter half an hour, so they are gaining time uh and with a better quality, and so that means that you can underwrite more with the same people, and that's also efficiency. Absolutely. There are no people that uh are losing their job, but we can simply uh produce or underwrite more for the same so it's for us also a motor for growth.

SPEAKER_00 18:53

And is that in the key message because you're not a data scientist, I think, as a background?

SPEAKER_01 18:57

Maybe you are, but no, no. I mean actually, but yeah, close enough.

SPEAKER_00 19:02

But how do you then convince the board to really invest in that? Because of course you have the the short term and long term. It isn't uh well to some it's still an investment in the short term. Um, some are a bit underwhelmed by the short-term gains, also. I think you already put that in context uh clearly before. But how do you then convince the board that still it is an important uh area to to still focus on and that the short term is not only the best proxy?

SPEAKER_01 19:31

I don't think I had to convince the board okay they were pushing and maybe even pushing a bit too hard because they want directly to see what are the benefits. Yeah. So we are more putting forward some efficiency gains that we want to obtain in our multi-year budget, even without knowing how you will do it. And then we see with uh some initiatives where or pools where we can still gain in efficiency, yeah. And then we will think how we will build it, and it can be with AI, but if you know the solution in advance, then maybe linear programming is as well as as good as well.

SPEAKER_00 20:24

So you're steering basically by putting a stick in the ground, which is a bit midterm, and then working towards that, irrespective of white.

SPEAKER_01 20:32

And AI, I I never saw already a full process in AI. It has to be embedded in the core systems, in the processes, and yeah, a part of it can be AI, but uh it's not a purpose. So uh, but is it in some cases it's the best way to get to the result? And so there we we have a budget there for AI. Uh it's not leading in the total IT budget, but it evolves during the year. It depends on if it is the right tool to be used. But we spend a lot of uh money also to create the foundations for AI. So your data the data, the sandbox, the tools that uh are uh used, etc. etc.

SPEAKER_00 21:29

So it's a means, it's it's not a goal as such.

SPEAKER_01 21:32

No.

SPEAKER_00 21:32

Okay, I think that's that's fair.

AI Budgeting And Partner Trust

SPEAKER_00 21:34

And in other sectors you see that they're also using a bit of the dynamic that is currently taking place and the complexity around uh AI and other technology uh evolutions to create ecosystems, work with alliances with third parties. Is that something that you see also for AG?

SPEAKER_01 21:50

Well we have a special department that is always screening for potential partnerships. But uh it was more business relating. So that's a bit evolving. But I cann say that our strategy towards partnership has changed with AI. It's simply another possibility with um data and uh sharing data is for us playing with the trust of our distribution, playing with the trust of our clients. So if we opt to work with a partner to do something to get it, uh the quality of the data and respectful handling of data is for us so important.

SPEAKER_00 22:48

So then we put our own requirements also on the partners because we stay responsible and uh and do you see other players or other competitors then popping up again in reference to the banking, where you do see, of course, a number of technology players trying to enter that market? You indicated clearly you're highly regulated, um, so that's maybe not that straightforward. But still, is that something that you see or that you say, okay, we're not seeing it yet, but it may come in the future? Because a number of the AI consequences are, of course, that middlemen are being cut out.

SPEAKER_01 23:27

Um, it's still evolving. Yeah. Um, but I think a couple of years ago they were talking all the time about insure tech that would take over the insurance companies. What happened was that they came but they took some parts of the value chain of an insurer. Uh so uh that's one thing. We are watching the evolutions, of course, and it can change, but it didn't happen.

SPEAKER_00 24:01

Not at a pace that we anticipated.

SPEAKER_01 24:02

Yeah. Another thing is that uh digital insurance, direct insurance doesn't uh uh evolve a lot in Belgique, uh in Belgium. It is the case in other uh countries, but for insurance we don't really see a pickup in uh insurance uh uh in live insurance and neither in non-life insurance.

Why Direct Insurance Still Stalls

SPEAKER_00 24:30

Because customers don't like it?

SPEAKER_01 24:32

Uh the tools are there, yeah, but it's not used. Uh if you see uh the statistics of direct insurance in Belgium for a horror over a horizon of 15 years it didn't evolve. Uh so why is it? I think first of all, uh it's a complex uh matter, uh, insurance. Uh people want to have a good cover because it's fundamental in their life if they need it, and I think the proximity they have with the broker or the bank is very important. If they can call when they have a claim, their uh distribution partner, it's for them more helpful than uh using a chat, but uh it can evolve, and I think for simply claims it is already evolving, they are using uh already more and more the digital tools, but uh for underwriting I don't really see uh an evolution and it will change, but for the moment, certainly for life insurance, we tried or we have a direct product Yongo for younger people. Um we had already partnerships with Cake and Key Trade, and it stopped because there was no not enough interaction with the end customer. So we are uh trying because even trying is a good thing because it's only then that you can test the your digital tools, uh but the digital adoption is a second thing, and I see in insurance that it takes uh a lot of effort to convince clients to subscribe directly online. It is the opposite for the claims. There we see more and more claims declarations because that multiple or settlements uh for uh retirement. Uh yeah, there you see clearly um pickup. But what we are seeing uh with all our digital tools, uh it takes a while for the digital adoption, and it's not by sending an email to the client, yeah. You have to repeat it, you have to add incentives also for to capt their digital consent, it's important. Uh sure. Yeah, it doesn't come like that. Uh you have to make efforts, you have to promise that you will plant a tree. But yeah, it's like that.

SPEAKER_00 27:32

Maybe but moving then to the the role that you have as an employer because you're also an important uh employer in Belgium.

Training Everyone To Use Copilot

SPEAKER_00 27:42

Do you feel that the people are digitally equipped when they s come fresh from school, from university? Are we seeing a uh progression there and is it going fast enough?

SPEAKER_01 27:54

Oh, but the yacht people, uh young people, uh yeah, they are more digital than us, huh? Yeah. Um and I think they are equipped to to use the technical uh tools and also Ali I on that side it's okay. I think more for implementing AI within a company, uh it needs a little a lot of change management, a lot of training uh to take away the fear uh to help them how they can prompt. So we invested a lot and it was on a larger population, it was on uh all the employees of AG, so not only for the people that are working with AI uh Lunch and Learns uh e-learnings webinars to inform them about the possibilities and help them to with uh prompting etc etc. So but the younger people you're optimistic. Yeah, they try, yeah. They are not afraid, so they are more used with it uh than uh the it's uh older three person.

SPEAKER_00 29:20

And the tone from the top, so you're upskilling because you're right, uh we are of course of a different generation, which is not yeah, a criticism, but it is what it is. Uh how is that then? I uh is management also part of that upskilling journey then?

SPEAKER_01 29:36

Uh certainly, yeah. And some of the e-learns are uh mandatory uh for co-pilot to be organized and training for the executive committee. Okay. And maybe we needed uh more time than the younger people, but we did it, of course.

SPEAKER_00 29:55

Okay, good. Maybe a few questions, um Heidi, that you don't require a lot of uh thought, I think. Just first response that you have in the morning. The first app that you use. What is it?

SPEAKER_01 30:16

I have to say that the first thing that I do is uh launch my computer. That's already good. Uh not an app, but uh yeah, I will use my phone to take the app of the tet and listen to the post podcast of the C seven.

SPEAKER_00 30:38

Yeah.

SPEAKER_01 30:39

So uh yeah, that's maybe the first time I launch.

SPEAKER_00 30:42

Okay. Whether

Five Year Forecast For Claims AI

SPEAKER_00 30:44

you think that AI will be in general or specifically for your sector in five years' time.

SPEAKER_01 30:52

Oh difficulty because it's going so fast.

SPEAKER_00 30:56

But you can't be wrong.

SPEAKER_01 30:59

And uh we will not check within uh five years'. Wow, we may, we may. But I think there is still a lot of potential. Uh so we are only at the beginning.

SPEAKER_02 31:10

Yeah.

SPEAKER_01 31:11

I think that uh in five years time uh five years time we will have tangible use cases scalable with efficiency gains. Certainly in the claims handling domain, uh there yeah. No, I told you that we have also always still a human validation for the material cases, but at a certain uh moment we will trust the machines because humans can also make mistakes. They don't forget that sometimes we have to repeat uh yeah, okay, we are not perfect as well. Uh we will be able to um use uh real-time data from the clients uh to settle uh faster but also to to take uh all their data into account in our underwriting. Uh I think as well that uh using this data will help the sector to uh change a bit the needle and also be more active in the preventive phase. So now we react if we have a claim, but we will know some things almost in advance. So uh that's also something that will change a bit the nature of what we can uh which comes back to the point of that that you can also be more uh inclusive instead of exclusive, because indeed you can with prevention steer a bit behavior. I think that's a bit of a and yeah, we can reduce likelihood of claims or even the total amount of the loss.

SPEAKER_00 32:55

If you

Advice Cybersecurity And Critical Thinking

SPEAKER_00 32:56

would be sitting face to face with yourself, but then the version of 18 years old, what would you say to her? What should she study? What what shouldn't she do that you have done and that you say with hindsight that I I I would do different now if I were 18 than when I did it?

SPEAKER_01 33:16

Huh. I I don't know if I would do something uh else because I'm working with data. I mean actually so maybe uh but for example, my son uh studied maths and afterwards he studied uh um cyber security. I think uh it's a good choice. Uh job is for the next years to come. Cybersecurity is something that will never disappear. So but what I would say in to everyone, I think knowledge will still be important. So I hope that in the schools they will force people to build still on intellectual knowledge uh human knowledge because I'm a bit afraid that if we are not able anymore to um test and validate the data quality and maybe the bias of the data quality in the alimentation and the calibration of the machine machines, then we will have a problem. So I would uh encourage all the younger people to still invest in their critical mind, uh have enough knowledge to be stronger than the machines, and that's a challenge I think.

SPEAKER_00 34:54

It is, but I think and I couldn't agree more. So with these wise words, I would uh say thank you for the yeah, the interesting session.

Closing Thanks And Next Conversation

SPEAKER_00 35:06

And thanks uh to everyone for again tuning in. Um hope you found it interesting, and see you hopefully hopefully soon uh again on another episode of our AI Unscripted. Thank you and bye bye.