Key takeaways
- Insurers carry decades of legacy systems from past mergers, and the pressure to modernize now comes from the business: a CEO who has seen what is possible, pushing IT for one view of the client and one for the employee.
- Claims are where emotion runs highest. Agents take in everything at once, so a claim can open with “what happened?” instead of “what’s your policy number?”, with people kept in the loop where the insurer and the regulator want them.
- Straight-through processing still runs on rules, but AI sends only the likely fraud to a person and picks up rule changes from its knowledge base immediately, where recoding used to take weeks or months.
- Accuracy is earned and shown: one insurer ran 1,000 identical cases and got the same answer every time, and every confidence score explains the rules behind it, a glass box instead of a black box.
Speed, legacy and a CEO in a hurry
Mike
Welcome, I’m your host Mike Parsons. And once again, we’re here in the FlowX.AI headquarters. And today, we’re getting into mission-critical enterprise technology. This is the Premier League, this is the real stuff. And what we’re doing is we’re unpacking a number of different verticals and industries: banking, insurance, and logistics. The reason we need to get into this is things are moving fast. There’s a lot of noise, and we need to find the signal. So how are we going to do that together? Well, what we’re going to do is we’re going to talk to experts, people who are working with this technology, with AI every single day. And we’re going to get into it. We’re going to ask what’s happening now and what’s happening next. And by talking with these experts, what we’re really doing is having a conversation. We’re connecting the dots because for all of us, if we can figure out how AI technology is turning up in mission-critical enterprise applications, then we can do a great job of accelerating the enterprise. So in this session, we’re going to talk about insurance. We’ve already covered banking and logistics. And I’ve got two fabulous experts, the gurus. One might say the first ladies of insurance here at FlowX.AI. We have both Miruna and Tanja joining us. Welcome.
Tanja
Thank you.
Miruna
Thank you.
Tanja
Thank you for having us.
Mike
It is a pleasure. I’m very curious about the world of insurance and how things are playing out right now. Perhaps, Miruna, you spend a lot of time I’ve seen you just today on the phone working with insurance companies. Tell me about the kind of work you’re doing with insurance companies. What are the topics you’re talking to them about?
Miruna
Well, here at FlowX.AI I’m a field strategist. So here I’m in charge with actually having the the conversation, the exact conversations with the customers to understand their needs and their pain points. And to actually try to accommodate that in our agentic solution in order to be presented to them and to convince them that they really actually need that solution to nowadays in their companies.
Mike
And when you have that conversation, Miruna, what are generally the biggest challenges they’re trying to figure out when they’re talking to you?
Miruna
So the challenges would be actually be around how quick we can implement this, how which are the integration points that we would look we would do actually with them. And actually what we are actually covering with these particular functionalities that we are explaining to them.
Mike
That’s fascinating because actually what’s interesting here, Tanja, is speed and integrations are two topics that we saw in the other verticals. When you’re working with insurance executives and leaders, this idea of speed, this idea of trying to integrate and the famous words of legacy systems comes up. What do you see? Now there are some other hot topics you see at the moment.
Tanja
Insurance companies are traditionally very heavy with legacy systems because of mergers in the past, or you know, or the data is a very high-level data-driven industry. So, they always were, and they are still, struggling with how to connect those legacy systems, how to bring to a client on one side just one view, or how to bring to an employee just one view. This is a constant, constant search for that. In that sense. So yes, legacy systems are always a token integration to them.
Mike
It’s an interesting tension because on one hand, you’ve got all the legacy infrastructure that’s robust and tested and then the pressure to provide these great, unified, integrated customer experiences and keep up with these frontier models and technologies. I can imagine that that creates like a pressure cooker situation where you’ve got mission critical business in the legacy system and the need to keep moving fast to keep up with competitive pressures. Is that something that you kind of feel when you have these conversations?
Tanja
It’s gonna depend on who you talk to. Because if you talk to a business person, he will just not care. Like, I just need this. I want a seal that looks good, this feels good, you know. Tanja, we have these 1994 systems. I don’t care about what happens to them. I want to have a new stuff. So I think it shifted now that business sees this beautiful, great, modernized front ends and wants to use them, wants to have them. And it’s pressuring back to IT just do something about it, right? They really go into fix it almost.
Mike
So it’s almost like your job is a bridge builder between the CTO and the CEO.
Tanja
Absolutely, absolutely. Because now the CEO will go to them. Looking at historically, IT was very powerful, obviously, or maybe still is, but his powerfulness says we can’t, it’s too much. It has these arguments, and now a CEO has a big better knowledge about it, educated more, understanding what are you saying to me now, these things, you know, the neighbor is doing this faster, where are you? So the pressure is coming from them to, let’s say, to IT levels, and they need to keep up with that. But I do think it doesn’t really help the IT sphere because they still have to manage this, right? They don’t need to certainly solve it, but they need to manage it. So they know and they want to bring those solutions to the CEO, but they still know we have this legacy system. What are we going to do? Do we just you know, plug it out? Do we replace them? Do we rip them? Do we so those decisions are in their spa sphere. So they are we are now in that part of the conversations right now. Are we going to replace everything that we did? Then you have a lot of egos coming, like but I built it for fifteen years, you know? So but that’s I I think that time needs to move towards it and we need to see where this is going to go to.
When an agent joins the insurance process
Mike
Well, this is sets us up beautifully for some of the questions that I’ve got prepared for you. And the first thing I would like to do is really talk about how things really work in practice. And so, thinking about this move where you have legacy systems on one side, agentic systems on the other, and trying to build bridges between them. Here is my first question. So we’re at the right time to ask this question. What actually happens when an AI agent becomes part of the insurance process. And Miruna, I would love your thoughts here because what we’re really talking about is perhaps previously there’s been to the side a little chat, ask a question, get an answer, maybe get a summary back. Okay, so there’s almost chapter one of AI. But now that’s coming embedded into the workflow, into the processes. It’s doing more than just a summary and an answer. It’s actually getting in the action. It’s perhaps even doing reasoning, doing decision making. So what actually happens for an insurance company when AI starts going into the workflow, what do you see?
Miruna
First of all, it’s making their life easier and of course making the work more efficient, more quicker, let’s say, because one thing it’s important here in the insurance industry is to keep your clients. So when you are working faster and of course greater, you keep your clients.
Mike
So when you say keeping your clients, you’re touching upon like the retention of your in this case, your customers’ customers so they can retain their customers better. Okay. This is really interesting. So you’re saying you might deploy an agentic solution on top of your legacy system. And that has a very internal focus. It’s like engineers, product managers, all the people that’s involved in getting those systems working together. But what’s quite interesting is the first thing that you say is, well, this helps them be more competitive on the outside in the marketplace. So I wonder, how far are we with that? Is it still first movers that are getting the advantages? Or is everybody there? Where are we with insurance getting to those more agentic workflows and offering more value to their customers and retaining more customers? Is it early days or are we further on?
Miruna
I think we are in the early days, [unclear]. If you think like this. ’Cause we had a lot of discussions with the customers and they are still not one hundred percent sure that they would need AI. And this is our job actually here to convince them that they need AI in inside their current core systems.
Mike
Perfect, because Tanja, what we heard is like that we know there’s an outcome of retaining customers which is going to drive revenue and drive growth. What’s the bridge? We talked about this before. What’s the bridge that you need to build for them so that they can see investing in these new agentic systems? And unlocking that commitment so that they can obtain that advantage. What’s the missing piece that you have to help them see and connect?
Tanja
What I and if we if you go and even compare it to a banking, insurance industry for a client is very complicated. Meaning, you know, you have the money on the bank account, you need to get that money, loan it, whatever. So, bank stuff is kind of interesting and it has everyday interaction with the client. So, it’s much easier to sort of automatize that in a sense. So, like give them every day some information. Well, insurance is different because interactions are not all the time. So, you just sort of need to once in 10 years, you might want to buy life insurance, but maybe the stakes.
Mike
Of that interaction are really high because there might be a claim, a distribution, all sorts of conversations.
Tanja
I mean, and it’s a lot connected to, you know, how are you deciding to buy insurance, for example. If you want to decide to buy insurance, you need to know so much. So you tend to put it on a person. Like an agent in front of you. And we haven’t seen really, really great disruption with the fact: like, I’ll give you everything online and then you choose. Because it’s not so deterministic. You can’t choose this deterministically. And now, with the AI, when you put AI there and the questions come: what about my friends? What about my family? Something that you talk to, you know, to Mike on the other side when he’s selling you insurance. You know, you kind of are getting those questions and answers when you have a conversational, let’s say, agent to to put in those questions. But the thing is which we are wondering, is the client want to ask those questions? Are they willing to give this this data, right? Is it something that they need we need to be concerned about? Or insurance agents will still come in play. This is where insurance companies still don’t know what’s going to do.
Claims, emotion and the human in the loop
Mike
Well, this is interesting. So, right now, what you’re sort of mapping out for us is AI and agentic solutions agents working in the process forces a reconsideration of when the human in the loop.
Tanja
Yes. I’ll take an example of claim, right? In the claims world, it’s a lot of digital paths already. But the digital path is just collecting the data. It doesn’t give anything back to a person. It just goes, what’s your name? What’s your policy number? Give me this document, give me that. It’s easy to do that. It doesn’t give anything, you know, in a sense, like, oh, your accident was very heavy, at least like the conversation, because the emotion in insurance is the highest in claims. So you would need to be able to communicate emotions back. And we now know a bit that our LLMs tend to come into that space. And you can create such an emotional space or safety for a person in claims.
Mike
For example, it’s really interesting because we kind of started this conversation around connecting the dots between investing in agents, unlocking better customer experience and retention. And then all of a sudden, what was like a technical and an ROI discussion now becomes like when is a human in the loop? What’s the empathy we need to have for customers? Because it’s a high involvement interaction. Do you find that that is something that’s well understood in the leading insurance companies? Or is everybody sort of realizing this at the moment? Like what’s that?
Tanja
I think insurance companies are trying to get to the space and figure it out, but I do believe this is very connected in the thinking process, let’s say, of insurance company. I just I will give an example when we are because we are on claims. Because insurance companies so data driven and so would like to collect this data as soon as possible, funny enough, the first question in the claim is what’s your insurance policy? What it should be is what happened? Yeah. You know, in a sense. But because you need to get the data as fast as possible, because if I don’t have the data, automation doesn’t work. I can’t really pull it off. But in common sense of AI, you know, because things in AI are happening at the same time, and we have these conversations a lot with clients on the other side, you don’t decide one after another, and then the human comes in the loop. You get everything in the face immediately because AI agents just do it together.
Mike
Yes.
Tanja
Right? So you kind of agents knows everything and it can play with it. So it can start asking this. They already know the insurance policy. Don’t mind him. So it’s doing these questions.
Mike
What a fascinating observation. So if I’m to understand it, for any executive in insurance right now, when they are considering the internal technical decisions about investments in infrastructure, it sounds to me, Miruna, like they’re really going to be finding out that it’s a question of redesigning human in the loop redesigning interactions with their customers, right?
Tanja
Yep. What Miruna I think what I would I would just like to insinuate to to explain, remember when we were having a demonstration was full redesign of underwriting with that lady, they were talking about what did she say to us? I mean, that I think is super, yeah.
Miruna
This looks amazing. That was her first impression when she sees she’s in that interface because they currently work with a lot of, as I’ve mentioned, with a lot of systems, you know, and they have to do this in the first system, do that in the second system. And why not having all of them in one orchestrated system in just a single application?
Mike
And so, that orchestration that you’re talking about, working across different data and legacy repositories and pulling that together, that was wow. And what do you see when they then redesign their customer experience? You started with this is a way to retain your customers. So what does best practice look like when those agents do a great job?
Miruna
So the best practice would actually be to beside the back offices applications, to have also some front-end applications for the end customers that would also keep their interest in that insurance company.
Mike
So if you were to look I’m just thinking here claims underwriting distribution, where’s the biggest win in your opinion?
Miruna
The end customer yard may be on claims would be because there you need to to have a friendly interface to put all your view of all of your inter information to to get a faster result regarding your you know accident or what happened, the the incident that happened. The underwriters the underwriting part is more to the back office persons, you know, not necessarily to the end customer.
Mike
But I imagine what you’re presenting here is a bit of a win win. So if I was a customer and I was working with my insurance company on a claim, and it can go faster with less of those moments where you submit things and then two weeks later they say, sorry, that was the wrong document, or we’re missing this, or we need another one. That if that can go quicker, I am happy as a customer. But if there’s less handling internally, if the systems are automating some of those basic processes, that’s actually reducing the workload for the insurer, right?
Miruna
Exactly, yeah.
Mike
Okay, so that kind of closes the loop on the business case that you started with, doesn’t it?
Miruna
Exactly, yeah. So having having everything from the beginning gathered in the single application, so you can easily reach the customer and of course faster to to to inform him about the the conclusion
Why a claim still needs so many hands
Mike
So it’s fascinating for me in in asking this question about what bringing agents and AI into the process looks like it is it is really exciting there’s a lot of opportunity but there is a lot of work that needs to be done to do it correctly. Because it’s not as simple as throw it in a model, throw in a prompt, and everything is sorted. This is really enterprise level, mission critical. So it’s really interesting to hear that there’s really a lot of integration and human in the loop design moments that can make or break the system. I think it’s now time to get into some of the problems and challenges inside of insurance. Now it’s really interesting that you pointed out, Tanja, that this is a high-emotional moment. It feels to me like if I was thinking about a checking account or a mortgage in the banking sector, like a claim feels a bit more like a mortgage moment. It’s a bit more emotional, it’s a bit more sensitive. And it really brings up the following question that I want to put to you in really getting to the problems that insurers face. Why does an insurance claim still need a person to touch so many parts of the process before a decision is made?
Tanja
So insurance is a heavily regulated environment. And all these regulations that people kind of have to remember now today, so when this is done manually, you need to know, oh, I need to send this to legal, or oh, I need to send this to somebody else. I need an approval above that threshold. The threshold can be made by a company or by a regulator, depends on which part of the world we are working on. And humans in the loop, or the second, how is it like four eyes views, for example, these are very, very critical in insurance processes. And they will always ask for it. So insurance company will always ask for it. But the fun part is that there’s nothing to do with technology because you can go very straight through with technology. So you need to artificially stop it where an insurance company wants to stop it. Through the years of, let’s say, digitalization of insurance, insurance is looking through for straight through process, meaning how can I get to a claim and finish it without a human touch. That is doable. I’m not saying this is not something you can’t be like, you could say for a claim below 100 euros, I have certain checks, deterministic checks, and I can just go through it.
Miruna
But for the low value claims.
Tanja
Yeah, correct. Very low value claims. But there’s still possibility to have a fraud on top or something is happening. But you can’t check that deterministically. You can’t say if it’s a fraud, then put it there. How do I know it’s a fraud? Where do I check? What checks do I have? Do I put in one thousand line or million line code so that I can have a straight through process? You don’t do that. While with AI, when you plug in AI, it’s going to be smarter in checking those frauds and putting it up to the human review. So you’re going to have more straight-through process because AI is going to pop up only the ones that is going to understand and know this is a possible fraud.
Mike
So here’s my question. This brings us to the moment of: how do you decide where the automation stops and the human is reintroduced? Because these agentic solutions present an opportunity to redesign that process, right? What do you see as best practice when you can get that straight through? And what’s the best practice for knowing that moment like we should have human in the loop at this point.
Tanja
It’s just old-fashioned rules.
Mike
Okay, so how do we know those rules and how are they changing now?
Tanja
I mean, you are right. If you’re talking about also changes, they change a lot. I will not say every day, but with new legislation, things are changing. The thing is, when those rules change in, let’s say, old setup without AI, this needs to take weeks, times, weeks, months to recode them. If it’s automated, for example, if automation is straight-through process, something changes, it locks. It just doesn’t, you know, for a while this will be wrong. While with AI, having AI agent has this knowledge base that is bumping into and constantly checking. So, what are my rules? What are my rules? When you change the rules, he’s gonna find the new rules. So, he has an immediate learning curve. While a person you need to teach him, he’ll forget, there’s a lot of errors. When is the human in the loop is going to bounce into the rule book. But with AI, is not you we’re not going to have an error between knowing and the execution of the process.
One screen, any document
Mike
So I want you to think now about projects that FlowX.AI is doing, is executing and running with insurance companies. What do you see in the teams at insurers that are doing this well? Describe for me like what best practice looks like in how they work and how they think.
Tanja
I believe that most of our clients are interested in this one. One screen almost, we would say, right?
Mike
Sort of a unification.
Tanja
The unification. And we me and Miruna were sitting with the c clients the other like a long time when they were kind of explaining how they do it now, right? I click here and then I copy paste there and then I go to the typos and right. A lot of re-keying, but the thing is, right? The human is like, there’s not a problem, I know how to do it. It was so for them, then I do this and then I do that. And we are both like, how are you even going into the, you know, how are you finishing anything? And he says, I remember it, I know how to do it. And then the the manager said, yes, but you had 16 weeks of onboarding. So 16 weeks a person needs to learn, and then I don’t know how many times the person needs to observe how they are doing it. And it’s all digitalized. There’s no paper or manual. It’s just so many legacy systems, in a sense, in that way.
Mike
So are you saying, maybe Miruna, a question for you? Do you think then the insurers that are going to win, that are going to get this speed and unification advantage, are those the ones that truly understand their processes, the logic, the business rules? Are those the ones that can adapt and move fastest?
Miruna
Absolutely. So if they understand that we can do d this with the agentic solutions very easily and put them in place very quickly, they will understand that the the the result will will come easier for them and they will not have to have these checkpoints to the other systems every time.
Tanja
Mike, if I may, what I think is an important difference before before and now is that managers or CEOs of the companies were always used to think: okay, if I’m gonna go to a project, I need to have a very structured data. It’s gonna take months for me to prepare all the documents. I don’t know where they are. I don’t have a process, how it looks like, because in order to automate, you need to have exact steps because automation is A to B. There’s nothing else in between. So now with AI, when we tell them, just bring us documents, whatever you have. You have rules, you have terms and conditions that are changed in every policy. It’s refreshing for them. It’s refreshing. So I don’t have to have all this on my whatever, SharePoint or something. No, yeah, just let’s just see it. The more you have, the better. Is it it’s not structured, there’s no is it PDF, is it Excel? Is it Word? Is it somebody’s email? Doesn’t matter for AI anymore. The structure is not matters anymore.
Open reasoning, exact answers
Mike
This is a beautiful insight. So I just want to kind of go back one step to make sure I’m following both of you. So there’s this great ambition for straight through unified experiences front and back end. But actually the key to the thinking behind that is to move from like a very closed, fixed, traditional way of thinking about requirements and the architectures and moving into an open, being more curious, more real-time, adaptive way of thinking.
Tanja
You are correct. The thing that we need to be very cautious is accuracy of it because AI is open. So it has it has a behavior. So it’s going to give you a reasoning that is a bit open between A and B. And insurance companies will have to go will have to think or calibrate between automation. So I would not call and describe it an old-fashioned way, it’s automation, right? The automation still exists. So they will have to kind of go: where am I going to bring an AI agent that has a bit of a, you know, thinking, maybe thinking, and where it has to be to the point accurate. So you don’t want to have. Maybe I’ll give you twenty five thousand euros or you know, to for your claim. What do you think? It has to have exact way to get to your twenty five hundred, four hundred seventy-fourths.
Mike
Okay, so we can enjoy all the benefits of a unified, straight through process, a more open way of working. There you go. But accuracy.
Tanja
Yes.
Mike
So tell me, when you see an insurance company thinking through and executing accuracy correctly with agents and agentic thinking, what does that look like? How do I know it when I see it?
Tanja
You have a perfect example from a company, from an insurance company, that was trying to bring this, let’s say, space in decision making, right? Because if they would close it, the decision would be if I say A my decision is B, but they said I want to open it a bit. I want to give advice to advisor in order for this advisor to find a B in it. Right? So they said AI should bring this open space, but the human at the end should find the B in it. So that’s where the expert comes in place, the human in the loop comes in place, so that they decide what in that se section between A and B is a decision for them. And they test it. So what they need and what an insurance company kind of needs. They need constant A to B. That’s what this is the constant that they are searching. And this company tested our platform. They put in 1,000 cases that were the same, the same documents, and they got A to B for them, the same for each and every one of them.
Mike
So that’s where you can be confident because you’re getting the accuracy.
Tanja
Accuracy in from that, yes.
Mike
So confidence and doing the work in order to get that with those permutations and that testing is a big part of it. And then what about judgment and decision making? What does it look like when the match to the A and B, as you said, wasn’t clear. How do you kind of get that to the humans in a consistent way as well?
Tanja
I mean, it doesn’t, it’s not no different than today. So today you had to read 47 pages, now you read a paragraph. So your decision between 47 pages, which is A and B of yesterday, to a paragraph, it’s just easier to make this decision. It’s just much more knowledgeable. You feel you are more educated in that sense, because the AI agent already did the homework of reading. It gave you the abstract and says this is it, this is how you should decide. And from that, you take a decision from it.
Mike
And FlowX.AI is actually working in production live with real customers doing these things right now, today. Yes, yes, yes. Okay, so we are not promising some magical future in many years’ time. This is today.
Tanja
Yes, this is today.
Mike
Okay, but this is very exciting because we have this idea of this unified straight-through experience. We have the ability to embrace the openness of agents, but have the fallback, the safety, the trust of knowing that we’ve got accuracy, we’ve got confidence and good decisions and judgments being made every single day.
Tanja
Yes. And I think Miruna, to the point of confidence score of how this is displayed, maybe you can tell a bit about that confidence score.
Miruna
Yeah, yeah. So we’re not just giving a confidence score. We are also explaining how it reached that confidence score, what it took, what a what rules have been actually checked to get that result and why you got that result. So, for example, we have let’s say a threshold and the the confidence score is under that threshold. For example, let’s say a sixty percent confidence score. It’s exposing to the in the interface, to the to the assessor, why you have that score, not just put it in there and makes a black box, glass box.
Mike
Yeah, yeah. But it it seems to me quickly and in real time presenting insight and context to employees to help them guide through that process seems like a huge win, particularly if they look at the number and they’re like, they can double-click on that and they can go deeper. So they can be confident, they can be accurate. Okay, this is fantastic. So, just to kind of frame what we’ve done together so far, I think what we’ve seen is, wow. We all want that speed. We want that level of integration in order to deliver more value, retain the customers as we started with. But we are dealing with a subject matter that is very emotional, and we need humans in the loop. Bit different to a day-to-day checking account at the bank. And what’s interesting is this took us on a path to unifying that straight-through process. But in order to do that, we have to embrace the openness of agents and AI. But we also need our backup plans to make sure it’s 100% accurate.
Tanja
Good summary.
Myth or reality: AI decisions are a black box
Mike
All right, we got one more to go. Okay. And it’s it’s very apt that you mentioned this idea between black boxes and glass boxes. And I think the thing with insurance, because of the emotional state, the high stakes in the transactions, so it’s very different to logistics, which is incredibly high daily volume, lots of small microtransactions. We’re talking about things that are really top of mind for the end user, and there’s obviously risk here at play for the insurer. And so, when we think about this third myth or reality, of course, what we’ve been setting up is a world where AI and the agents are moving from being something like give me an answer, and we’re picking it up and we’re putting it into the action, putting it into the workflow. And on top of this, we’re bringing it into decision making. This is obviously there’s a lot of opportunity, but there’s also a lot of risk here. So let’s think: are we talking about myth or reality? Here we go: myth. If AI is involved in an insurance decision, you’ve created a total black box. Is this a myth or is this the reality? Think about where we are today. Maybe, Tanja, do you think if we actually put AI in the decision making, is this a bit of a black box that there are a lot of unknowns? Could this all go a little bit? Could there be hallucinations?
Tanja
I think we are too early to put it in real decision making. And on the field, we haven’t seen a client yet that would just put it in the decision making. Is still we still know from the home usage that do we trust it or not. So we don’t understand technology where we as FlowX.AI are kind of really narrow it to the point. So they still need to see that. So there’s some some time needs to fly. That’s one reasoning, I think. And the second reasoning is that also a regulator or legislation is not really or is treating these kind of cases differently. So if there is decision making involved, that AI is doing it, it has to be super transparent, it has to be very visible to a client. At the end of the day, there should not be any decision making instead of somebody else. It can be a decision I will move from one flow to another, but I will invest that money. That’s not anymore a topic, right? So it there are different kinds of decisions that an agents are doing in in that sense. But I think it’s legislation is kind of holding its time. We have to know this is one year old topic. I mean, we should not be truly jumping into, you know, fully autonomous something happening. And it’s not a technology problem again. This can be done, but it is how we gonna govern, what’s gonna be the rules. And I think from the perspective of technology and where FlowX.AI is, this is super important to follow so that we have modularly set our infrastructure that it can follow those rules.
Mike
So we seem to have when we want to reach perhaps the highest form of automation and agentic workflow, we’re saying, okay, that involves going right up into decision making. Not just you know, OCR inputting docs or reading the handwriting. We’re talking about the decision making. And there’s obviously a lot of reasoning and thinking going on in that moment.
Tanja
It’s very behavioral, Mike. We should not go into the human behavior yet, I think. You know, decision making.
Mike
That’s very interesting, right? So, as we approach it, where are we now in the proximity towards that North Star of decision making? And what are some of the next steps on the roadmap that you can see, for example, for the next 12 months? What do you imagine? Because you said it earlier, like we weren’t talking about this two years ago, and we’re talking about it now. We didn’t imagine. Just recently, people could not have imagined two years ago that these highly complex mathematical problems could be solved by AI. And now they are. So in making the progress with agents in insurance, where are we now? And what could you see being possible in the next twelve months?
Tanja
I think it’s all connected to the risk appetite in a sense. So when you have five years ago, you were talking about digitalization and you already had a threshold when you said just go for it, right? Just one hundred euro let’s do that and everybody can we’ll take a risk that this is wrong, but we can do it. And I think in this space of AI, we will come to the same. I think eventually it the person is going to like, why am I doing yes all the time? Let’s figure it out. Where is the threshold where AI is safe enough to do it? And it’s always going to be a threshold where no way AI is going to take any decision. So it’s about complexity and risk appetite from insurance companies to understand what happens if I’m wrong there. What’s my fault? Do I have to pay to anyone? Is somebody going to come to our space doing it? So it’s all about. About that, what is bigger and higher risk?
The next twelve months
Mike
And in the next 12 months, if there was one new step to take with agents in insurance, what would you like that to be? If there was one thing to create even more value for your customers, what do you hope can be put on the table for them in the next 12 months?
Tanja
12. Tomorrow. Your time flies so fast. I don’t know, Mike. I mean, it’s just so many things that I can’t pick one really that I would say this is about the agenda. I just hope and wish people would get educated about this more. We still get questions: you’re gonna just throw in my data or. You know, I think it’s a very nice comparison that Serban sometimes makes, you know, and he says, in a sense, when telephone was invented, do you think the world stayed in the line? You know, in a sense, do we think what LLM is going to be doing is still a black box from the perspective of us natural people, but not from the people that are actually working with it?
Mike
So what you’re saying to me is we need to better understand governance, resilience, the overall what Serban often calls institutional AI. We need to understand and map that in order to get the most out of AI in a safe manner, in a reliable, mission critical, enterprise ready manner?
Tanja
I think it’s super important, and we all feel the difference between last year at the same time, no, not nine months back, and conversations we had with clients, questions they were asking, you know, fear they had versus today where questions are totally different. And they understand a bit better. They tried something. They feel where they can fail. It’s just the speed of that is unimaginable. Nine months ago, when I said to them AI, oh, that’s not something our company will be doing, well, go to that company. They have strategic committees, thinking about whatever is going to come down the road. The speed is super fast. And yes, I don’t think there is a company thinking not thinking about it.
Mike
Miruna, I want to ask you if I look at some big areas within insurance, claims, underwriting, distributions. What area of insurance do you think could make the most gains over the next twelve months? If you look at all the things happening in insurance, do you have one that you’d bet on?
Miruna
I would bet probably on claims. I I mean I love these these modules of claim and also the functionalities that are the agentic functionalities that are under the the claim solutions. Because here, and I’m making again the reference to the end customer, their client, the customer’s clients, to have the satisfaction of actually having the best solution for them and be more efficient in this kind of processes.
Mike
Well, I think what’s so nice about that idea is that we talked earlier that the emotions for the customer’s customer is high. There’s a lot of context and data around understanding the claim itself and what actually happened. Serban was telling me about being able to interpret handwritten inputs, these sorts of things, which is great.
Tanja
Pictures.
Mike
Pictures of the scene, handwriting. We talked about a number of these things. So it’s great to see that there’s so much opportunity there because there’s nothing worse than, you know, particularly if you just imagine if your car is damaged and you can’t get to work and it takes weeks and weeks just to finally get your documents accepted on the submission of the claim. If we can collapse that, that’s a win for the customer. But as I said, it’s just one of the areas that agents can play a role in helping insurers drive their growth and retain their customers. So it must be pretty exciting for you to think about this next 12 months on the opportunity you have to bring this new paradigm and to help them put it to work. If you could teach or help insurance companies learn some hot topics for AI, we talked with Tanja. She was talking a lot about like learning and understanding is key. What topic would you love insurers to know better, to know deeper when it comes to agentic in the mission-critical workflows?
Miruna
I would say that the process of actually doing the eligibility, how this eligibility part is actually done in a claim, because you need to do a lot. Currently in the insurance companies, you do you need to do a lot of cross checks between a lot of documents, a lot of systems that we are we have mentioned. And this eligibility part is the most, let’s say, time consuming part, you know? And yes, this would be actually one part that I would like to to to teach them that is very, very let’s say necessary for them to to understand and pick up in their new solutions.
Tanja
Mike, I have also one thing that I think would be great to come across to everyone is that insurance companies are very ins you know, thinking we are too complex for the world to understand because they are, or you know, in that sense. But the fun part is, or the funny thing is, the AI loves complexity.
Mike
It does.
Tanja
You know? So, the underwriting process, for example, right, which is you can’t understand it from the street in a sense. It’s it’s it’s a super complex. Even the people working there are very scared, so it is isn’t it’s not everyone can do it in a in that sense. So they always thought you can’t automatize it. So digitalization just skipped them. They still get tons of documents. Yeah, they are digital and PDFs, but they had to read them, they had to have this knowledge around them. And now when AI can actually you spend so much time prepping, reading, understanding. Even in the claims and when you said when this eligibility check comes to a place, right, it’s similar here. So you need to understand why would you insure that, what pricing, and clients are waiting for that. And insurance company is nervous because it takes so much time to decide what I’m going to price you. So this is coming back. Am I going to stay or not? Can you give me now? Like, I need 10 minutes to get the price to re-quote. But the insurance company couldn’t do it. And now I think the message across is if it’s complex, it’s really brilliant for AI. And we see this that underwriting case, our very popular case, let’s say, is a use case is something that insurance companies are just loving because they finally understand the complexities being solved.
Mike
Yes. So it’s really interesting, like the appeal of agents to logistics was about volume. But what’s really interesting is you’re painting a picture: it’s the complexity of each transaction, the emotional stakes of every transaction. And what’s really wonderful about what I’ve learned from both of you today is that if we take this internal deployment of agents, that we can have a dramatic effect on the external customer experience and drive growth for these insurance companies at an enterprise scale, which is very exciting, isn’t it? Yes. Well, it is. Thank you. I really appreciate you sharing your ideas. Tanja, Miruna, thanks very much.
Tanja
Thank you.
Miruna
Thank you so much. Thank you. Thank you.