Get a demo
Select the type of demo you'd like to experience with Indico Data.
Select the type of demo you'd like to experience with Indico Data.
Watch Indico Data CEO Tom Wilde step in as co-host alongside Michelle Gouveia, VP at Sandbox Insurtech Ventures, in season 2 episode 6 of Unstructured Unlocked with Silvia Signoretti of Swiss InsurTech Hub and Naveen Dhar of Microsoft.
HOST:
Hello everyone. Thank you for joining us today for our webinar from Expertise to ai, bridging the knowledge gap and insurance underwriting today with our guests from Indico Data Swiss InsureTech hub and Microsoft. Before we get started, we’re just going to go into a few housekeeping items. This webinar will be recorded and it’ll be sent out within 24 hours of the webinar. Everyone’s going to be muted except for our speakers today, but please submit questions along the way throughout the webinar and we will address them at the end. We’re going to allot a good 1520 minutes to answer any questions that folks might have, and then if we don’t get to your question or we want to connect offline, we’ll share again our information with a copy of the recording so we can get in touch after. And without further ado, I’m going to hand it over to our terrific speakers to introduce themselves.
Tom Wilde:
Great.
Silvia Signoretti:
Should I go first? Please introduce. Thank you very much for inviting me to this webinar. My name is ti, I’m the president and co-founder of the Swiss InsureTech Hub. This is an profit association creating an ecosystem in network of insurance professional in Switzerland, so I’m very happy to be here and contribute to this discussion.
Tom Wilde:
Great. Welcome, Naveen.
Naveen Dhar:
Hi, I’m Naveen d I’m a senior director at Microsoft. I’m part of the Microsoft Industry team. We are focused on insurance transformation, helping companies through insurance transformation. Most of the team, including me, has an insurance background. I work with insurance companies carriers specifically before joining Microsoft, so I look forward to the discussion.
Tom Wilde:
Excellent. Welcome, and I’m Tom Wild, your host Today I’m the CEO of Indico data. Indico data has developed an AI first submission and claims intake solution helping major insurers with complex unstructured data to structured data challenge. I’m very excited to have Sylvia and Navin here today to talk about a very current topic in the industry, which really relates to the intersection of talent acquisition and trying to grow and service the carrier’s business around underwriting claims and policy servicing. A couple statistics to start us off with here to anchor the conversation, 71% of property casualty insurers expect to increase their staff in the next 12 months. So the insurance industry is quite healthy right now, by and large, growing rapidly. There’s new demands around the world to manage new kinds of risks, and that means more workload around all of the three key functions that insurers face, underwriting claims and policy servicing.
The challenge. On the other hand, while most insurers are planning to grow over the next 15 years, the Chamber of commerce estimates that 50% of the insurance workforce will be at retirement age. Insurance has long been, especially when you think about underwriting more of an apprenticeship type business. And so the institutional knowledge and skill that exists in the underwriting pool of talent is difficult to replace, difficult to onboard new folks, and this is a lot of pressure coming from these two ends of the spectrum, a desire to grow the staff, but a retiring staff potentially. And 25% of all insurance companies said recruiting and hiring has become more difficult. And that’s also part of, I think a global trend in all industries, but specifically in insurance where specific skills, experience and knowledge are required to successfully do the job. This is an acute problem. That’s the table setting for what we’re going to discuss today and specifically how this intersects with at the same time the rapid growth of artificial intelligence and how artificial intelligence can augment the work that carriers do and perhaps alleviate some of this stress by extending the capabilities of the team that they have and making overall the companies more efficient and in the process perhaps reduce overall expense ratios and improve loss ratios.
So that’s what we’re going to discuss here today. Talia, if you could take us to the next slide. Thank you. So how can we leverage artificial intelligence to augment and address some of these major issues here to drive efficiency within the insurance sector? So Sylvia, maybe I’ll start with you. You have a specific view of one of the most important insurance markets in the world, the Swiss insurance market, obviously a center of a lot of the reinsurance activities in the world. What have you heard from the membership in your organization around this question of talent acquisition specifically?
Silvia Signoretti:
I think maybe even to take a bigger picture, not only, I mean whatever in the network I’ve been creating, the challenge of talent acquisition insurance generally is not a new topic. It is something that’s also in my former employer, Zurich Insurance been tackling and discussing at board level. But probably you’re right now it is becoming a more clear issue as we see from demographic statistics, quite a large portion of lets say technical functions that employees will retire. And of course companies have been looking at process automation. There’s an opportunity to really create efficiency, try to create roles that are more focused to the more technical aspects and less on the administrative aspect. But probably we haven’t seen enough of those, let’s say super dramatic changes in the industry. But we still see a lot of venture techs of course, bringing quite a lot of new ideas to revisit these classic processes in insurance to really help in bring the industry forward. So I think the interest that is there is just that there’s still challenges to overcome now to really embed the solution deeper.
Tom Wilde:
Do you think this notion of underwriting as an apprenticeship, is that accurate? Is that a fair characterization of how underwriters think about the work they do?
Silvia Signoretti:
Well, not sure. What I think is a challenge to get the people into these type of apprenticeship to really get into this role is also how they would be interested into insurance as an industry per se. So I think what is also changing in today’s demographic or this generation that is supposed to supply the bigger portion of the workforce in the future, there may be less and less interest of insurance as a potential employer. And probably also because now they are touched or their feelings are close to their need of looking at insurance as the protection for people have the tendency to not own houses or not own cars. So they get less of these, call it a customer type of experience to think of insurance as a potential employer. But I think from an insurance company, there is a lot that can be done because it’s one of these, I would say from my own experience as I haven’t even thought of getting to work in insurance 20 years ago.
At the end, it is an industry that can offer a lot. There is an amazing great job for every type of creative technical type of profile to really enjoy quite a diverse type of career. But I think it really comes from the board HR function to really help create a more compelling stories of what insurance can deal. And then of course we have to go back to labor operations function and say, okay, what can we do more to link to push the ball a little bit further to embed technological solution to improve processes and really look at and the writing functions or claims function in a way that we have to bring together the technical skills and limited skills together to this gap, this key gap.
Tom Wilde:
These two stats on the screen suggest that the underwriting departments certainly believe that embracing new technology will help address some of this. Naveen, I know that Microsoft has been spending quite a bit of time investing in the employee experience. Your new product, Veeva is centered on sort of helping the digital experience of the employee. When you think about insurance specifically and the growth of things like chat, GPT and the offerings around Azure, OpenAI, what have you seen in terms of how younger talent is looking at those technologies as part of their daily job?
Naveen Dhar:
Yeah, so Tom, let me take it a little step back. You talked about apprenticeship, and I’m going to elaborate a little bit for what I’m seeing and then connect it to your next question. In the US we have the baby boomers going to retire. So 20, 30 by 2030, we have a huge talent shortage. The insurance companies are not only looking to hire new employees because there’s decades of experience there. And traditionally in the old days, I think apprenticeship model is a very interesting word In some sense it resonates with me and there’s always been this getting to know the intricacies and you have to work with an underwriter to learn how do you manage the knowledge transfer? And I think Silvia talked about the needs of the younger generation and the insurance companies are looking to see how they can effectively use technology as a way to help with the risk and maintain task to help the underwriter.
It’s not to replace the underwriter, but it is to help the underwriter and that productivity gain comes through. So whether it is through tools that we are going to deploy, whether we use Viva chart GPT and build custom copilots, but the intention is that process because if you look at the world you have in the emerging worlds in the Asia side, the latam side, the business is growing rapidly. You need more underwriters on the US side, you have people retiring and so the need for talent is always there. And as you get fresh blood to Trivia’s point, I think we still have to work on how to attract people, but how are you going to train them and how can you make it compelling enough that they see that technology could play an important role? And I think that’s the issue that insurance is working through today.
Tom Wilde:
Yep. Great. Sylvia, in terms of that knowledge transfer question, how can technology help move from sort of individual knowledge to institutional knowledge and how much of that is enough and how much is too much? Where’s the line draw between trying to create an institutional asset around the underwriting experience versus training new underwriting leaders?
Silvia Signoretti:
Yeah, I think again, when we look at what insurance company are trying to achieve from a vision perspective, I believe no one is thinking that we’ll have a fully automated insurance company. So we are still looking at bringing together the skilled human person with a, well, let’s say embraced technology. So it’s a little bit of looking at the situation, let’s say, or dimension like culture in an organization, not organizational, operational, technical setup. And of course the training piece to keep the population, the employee population continuously call it evolving in their skills in light of the technology evolution. So I think it is really what usually happens in insurance companies that we see that the data management is pretty compartmentalized, siloed in certain areas and the reason need to bring it closer to maybe the underwriting claims community. And so just using the simple or the well-known principle of of data, really making sure there is quite a stronger knowledge, a stronger data skills within the underwriting community itself to really decide on to making them, owning their data and also using them in the way they feel they think they need to do it to really address their business questions.
So this is what Veeva has to happen. So working on dimension of culture, organizational, technical level, and of course continuous learning.
Tom Wilde:
When we look at things like Azure Open ai, we’ve kind of moved on the spectrum from the early days of artificial intelligence where predictive type use cases were quite feasible. We could use data and do statistical regression and do things like risk analysis and pricing using kind of predictive type technologies. Then we moved into extractive type capabilities. How do we turn documents into data? How can we interpret customer or how can we extract key attributes from customer feedback? Now we’ve moved into a world of interpretive capabilities and that has pros and cons. So as you talk to insurers and they suddenly realize there are these new interpretive type capabilities that can contribute to helping make the team more, what are you seeing as their both enthusiasm and their concerns when they look at this?
Naveen Dhar:
Yeah, let me take an example, right? So let’s take a underwriting process for let’s say high network underwriting or maybe reinsurance underwriting. Effectively what happens is you get documents and the documents could be financial documents, it could be medical documents, it could be an ID to identify who it is. Traditionally what we had was we used to have a shared center that would actually take those documents, read them and identify and summarize what those documents were before it went on to the underwriter who had a summary to work from. But that process of getting data sometimes and identifying what data is missing, used to go back and forth multiple times because guess what? The underwriter gets it and says, well, I’m missing one document. The process starts all over again. And so you went back and forth. And then that is not only using up the underwriter’s time but is also consuming other resources in the insurance industry.
And the insurance industry is facing a talent shortage. So the conversation has been now with these, as you called it extractive technologies, could I help make this process a little easier, a little more productive? Can I use those technologies to read these structured unstructured semi-structured documents, extract data, maybe run them against a certain rules to say you’re missing these elements before the underwriter sees it or maybe even get to the level of summarizing it for the underwriter. And so now what has happened is the productivity goes up because the moment the agent submits the document, I could have AI tell you that you’re missing three documents and that process is shortened. The underwriter now starts getting more complete set of documents, so he’s not wasting his time going to half and realizing I’m missing four more, start the process. So effectively what it’s doing is making the underwriter productive, changing the game. And if you go back to the last question, we talked, the new talent sees this as a way to leverage AI to make themselves more productive.
Tom Wilde:
So one solve for this is not necessarily to have to keep hiring and hiring and hiring, but to make the existing team more productive and generate more throughput in that manner.
Naveen Dhar:
And I think the challenge is that there is a talent shortage. It’s not like we can hire as many as we want. Look at the experience our younger generation has. They want to do as they do with instant consumer experiences that they have, so that talent pool is interested in leveraging technology and not doing mundane tasks. They want to be able to be more productive, leverage technology the right way. So it’s a win-win for both sides. It helps the underwriter become more productive. We can get more talent and make it more lucrative for them to enter these professions. I think it’s a win-win for both sides.
Tom Wilde:
Talia, next slide.
Silvia Signoretti:
Maybe in the meantime, what I think is interesting of what Nev said is this clear understanding of what the process has to be from end-to-end. And this is something where sometimes companies still they want to address the end-to-end analysis of the process and making sure there is a clear call it list of controls in every key steps to really make these processes super efficient. And I think it’s a very, very first call it the activities that the company has really invest in the process analysis before then harnessing all the potential of data next to it. No, I
Tom Wilde:
Like that. I think really being outcome focused is really critical. I think there’s a bit of at times of mistaken belief that you don’t need to strictly define successful outcomes when using ai. And in my experience, that hasn’t been the case, right? AI at the end of the day is an approach or a tool that can help you achieve an outcome, but it should not be used just for the sake of using it. And I think that’s a really important point, Sylvia, is to find what outcome will contribute to your success. One of the things that I’ve heard over and over again in the underwriting world is that most underwriting processes, only about 30% of the submissions are quoted. And that’s partly because of this topic we’re talking about. There simply isn’t enough human throughput, human bandwidth to address the full river of submissions. Is that something, Sylvia, that you’ve seen in then alternative Naveen as well? Is that true? And do the insurers want to be able to quote a hundred percent or are they happy with 30%? Are they getting the right 30%, et cetera? Those kinds of questions?
Silvia Signoretti:
Yeah, I admit they don’t have these direct insights, but of course when performance of a function is assessed, they look at these KPIs and they go back and definitely analyze, okay, what is the root cause of not getting even for the end of the day, it’s always about growing, so increase the top line. But you can really do that if you harness all the potential business opportunity out there. So yes, definitely. I mean this is a concern. I mean, how then all these call it assess KPIs are turned into action to make a case on our real dose solution technology solution really do help in turn the business result upwards is something that I would probably have to see myself, but this is part of the performance management system companies do definitely put in place.
Tom Wilde:
And I think you and I have talked about this in the past, this sort of challenge of accessing the total river of submissions, but as the slide depicts here, how to do that safely, securely, scalable, how to make sure that you are looking at the right kind of risks and spending time on the right kind of submissions. How has this manifested in your conversations with your customers and partners?
Naveen Dhar:
I’ve had this conversation I think with some reinsurance companies. I think the challenge that they posed as this are deposed, as they said, we are getting a lot of submissions, but we have staff only to go through 70% of those to provide a quote. And they said, well, we have two options. Either we change our current process or we figure out a way to identify which of those submissions are more likely to get us more revenue and are more likely to get us win as business so we can prioritize the work that our underwriters are doing. So the final quote would still be through an underwriter. It would still be human in the loop code, but is there a way to use AI to prioritize what is it that the underwriter works on? Which I thought was a very interesting concept because in underwriting in the old days, when we have a human look through these documents, nobody asks, which line did you get this information from?
Or where did you go? We assume that the human has looked through the document, identified the right information and identify the risks and put it into the risk system. But as we start going through and leveraging ai, the ask is that I need to have an audit trail. I need to be able to go back and know where was decision made, what were the parameters that went into making the decision. And so I think that’s the piece that’s going on where we are trying to figure out how can I make the underwriter productive? Where can I automate and what can I do? I’m still not at a place where I can automate completely. It’s human in the loop is definitely required, but how can I make the underwriter use the technology as an aid, as an assistant to make him more productive?
Silvia Signoretti:
Maybe the point you’re making here, Tom, is how your technology can help to demonstrate two benefits of ai, one on how we improve the quality of underwriting, the 70% of the submission are taken so that you really had the real time risk assessment. So then you then tackle the additional 30% of the submission to really grow the top line because I think this is what is needed. Bring a demonstration of the real benefit on along the value chain of underwriting and see what are the missing opportunities.
Tom Wilde:
And solving this has a very direct benefit of increasing gross written premiums, but doing it along the risk frontier that the insurer desires and can tolerate. I think one of the fascinating things that I experienced when we started working on this problem, AI is still very much a data in decision out sort of chain. And if you put bad data in, you get bad decisions out. And when we would sit down with a team of underwriters and ask them, and I think Naveen, you were kind of poking at this, ask them to explain how the underwriting decisions get made. What was pretty interesting is there was kind of a fairly wide range of how each underwriter in a specific insurer thought about the problem and how they could define the problem. And by define, I mean provide sort of training examples. Because AI needs training, right?
Regardless of whether it’s a very large language model or a small language model or machine learning model, you still have to give it examples that it can learn from. And I think that was a very clarifying moment. I remember this one particular customer where we were sitting around the table and they realized, wow, there’s more variability in how we’re making these decisions than we thought. Even though it was a room full of people with 20 years of experience, that’s not necessarily a comfortable place to be, right? But when you start to try to codify and augment some of these processes, it forces those discussions and I think ultimately leads to a better outcome because now you’ve had a chance to say, well, how do we really want to make these decisions? If we know that, then we can apply it to a much bigger set of submissions and get the results that we’re looking for. Yeah, for sure.
In terms of Naveen, when you talk to insurers about the use of ai, and again, staying with our safe, secure, scalable theme for this slide, how have they thought about bias and how they’re training AI to make these decisions? And I think a lot of times bias gets construed as the social bias that we all recognize as certainly issues to address in any kind of ai, but within an insurance company it’s usually less about that, more about sort of decision bias in one direction or the other. You have to be careful not to codify bad decisions as much as you do good decisions.
Naveen Dhar:
And I think I want to take a step there, a little step back because I think there’s two parts to this conversation, right? One is the documents coming into underwriting that is today manually somebody summarizes and goes through, and then there is the actual underwriting decision. I think the documents coming in, I think you called it extracting data. AI
Tom Wilde:
Is extractive and then interpretive now with J,
Naveen Dhar:
I think the extractive one is more open for automation. We are at a stage where we are willing to do that. A lot of insurance companies are doing it. We have so many cases that are going through as an aid to the underwriter, the actual decisioning. If you were looking at life insurance, when the data is discreet, it’s not through LLM. I’m getting it. I’ve got specific example, and I’m not going through a LLM model to extract data there. Of course, automation definitely is working on the interpretive side as well, or any other on the PNC side or commercial side. But when you’re talking about things that are based off of extractive data coming through LLM data coming through, then the decisioning piece, I believe right now we are not yet at a point where that’s going to be automated. I think there’s a lot of questions that we’ve talked about by us and other stuff, but I think that today it’s still going to be the underwriter. That’s why I say underwriter is in the loop. The decisioning part is done by the underwriter. It’s the aid to the underwriter that’s being automated.
Tom Wilde:
Yeah. Sylvia, you and I got to spend some time in Zurich last week at the event that you hosted a wonderful event well attended, and really interesting conversations. What were some of the takeaways you had in terms of how carriers should get started with trying to apply AI to some of these challenges? What did you take away from the discussions in Zurich there?
Silvia Signoretti:
Yeah, I think there were a couple of, let’s say ingredients in the question. One is definitely looking at the, let’s say, solid partnership to really rely on, well, let’s say prepared institution or companies that can really help insurance company in navigating what are the potential aspects, the potential positive outcome, the technology can bring in processes. And it also, of course, we spend also quite a lot of time of course, really looking at the use cases and making sure that how those use cases are built on the right, let’s say infrastructure with the right set of application and making them available. So some people talked about the many number of use cases available and how they can further extrapolate on them, but then they are one of the biggest challenge everybody noticed made these use cases available and implemented is definitely the element of governance and security and safety.
And I think this is also in most mature markets, the area where a lot work still needs to be done and where I see companies talking about AI governance framework for monitoring, creating the tools to monitor what these MLL models are and how they could help. But the question is always is this, do we have enough control, preventive control, or post-implementation controls that really help provide enough, call it let’s say safety in the way those models are deployed. And so those were still the biggest concern, and I don’t think there is probably the answer that satisfies everybody’s needs in particularly in the legal departments. But those were the fundamental questions we addressed.
Tom Wilde:
Yeah. Nave, maybe last comment and we’ll take some questions. How about from Microsoft’s perspective, how have you been coaching customers around this walk, run, sprint sort of metaphor when it comes to adopting, especially some of the more modern gen AI applications in these spaces? I know that certainly internal knowledge management has been the first place, a lot of people have started lower risk and kind of immediate value, but how have you sort of seen people start to move those technologies out to the customer experience?
Naveen Dhar:
Normally when I talk to the customers, I talk about it in three different dimensions. One is how do you use the customer experience as a way to reduce or increase your revenue or reduce your expenses? So the example may be I have an accident and the customer and the way we can get the customer to interact, which reduces my call center costs and other. So that’s the piece of where AI could help. The second piece is around the processes. So if you look at our claims process, underwriting process, insurance processes in general across the board, there are many areas where we have opportunities to automate. Sometimes full automation, sometimes automation with a human in the loop. That helps. And the third piece then is about leveraging things that exist that can help. So for example, our copilots, we have the capability to help. So example, in a claims world, I’m sending out a letter to a customer, can I add empathy?
Can AI built in for all the existing products? I can change the way I write it. So I think the piece that we are talking today is squarely in the center piece. We’re just talking about our processes and the insurance perspective. Where can we leverage AI to help make it more productive, make the employees more productive, whether it is the underwriting or claim. So that’s been just the conversation we have been having, and there seems to be a growing demand from insurers in trying to go through all aspects of the process to see how they can leverage AI and automation and helping productivity.
Tom Wilde:
Great. Alright, well we will move to the q and A portion of the conversation here for the next five to 10 minutes and take questions. Talia, if you’ve got a question queued there, I have one on my side here as well, but go ahead and let us know if you’ve got one there in front of you.
HOST:
Yeah. So we have one here, AI and regulation around not quoting declining submissions. Is there regulation risk around this?
Tom Wilde:
So just to interpret the question, is there regulatory risk around deciding not to quote a piece of insurance? Is that the gist of the question? Yeah,
HOST:
I think the question is, is there regulatory risk around not quoting AKA declining a submission as it relates to utilizing AI to do so?
Tom Wilde:
Well, it’s a great topic in general. So maybe let’s poke at regulatory aspects of AI for a second. I know the state of Connecticut, I think was the first DOI in the United States to issue their requirements around the use of AI and insurance. Maybe I’ll start with you, Naveen. What have you seen there from customers in terms of how to address what the regulators, who in general regulators are one to two years behind where a market is, they try to keep up with the technology, but what’s happening in the regulatory space?
Naveen Dhar:
I think the key there is traceability. As long as you can document and trace your decision, then you’re going to be okay. And which is why I said that’s why when I talked about, I split it and said, what can you do for the personal lines where you have discreet data versus where you get unstructured documents through LLM, right? So where you have discrete data, you can trace back and I think LLM will get there. We’ll have more and more comfort level with that automation. But where you have discrete data, where you have traceability, I think it works fine. And I think that’s the key auditability and traceability from a legal perspective,
Tom Wilde:
Sylvia, in Europe, the data privacy regulations are as strict as anywhere in the world. I know in Zurk we did have some conversations around this specifically with some perspective from the compliance and legal folks. What’s the temperature around gen AI in the European markets that you’ve gathered? I think you’re muted.
Silvia Signoretti:
Interesting. No, I think there is quite a huge focus on the impact of these regulations. So as I mentioned before, the legal and compliance department are all over the place, putting a lot of effort and also hiring the really specific skills to really be on the business partner side with the technical function to make sure, and also the operations function to make sure there is enough control, enough understanding on how those models and the output of those models may have a reputational impact, of course on the company. So this is a topic that will continue to stay as this is an area that is in continuous call it also in evolution per se, because we see more and more of those, let’s say, I don’t want to say intrusive tech regulation, but limiting from an advancement or internal advancement in thermal technological advancement perspective perspective, they could be quite limiting.
Naveen Dhar:
Yeah,
Tom Wilde:
Go ahead. Were going to say So
Naveen Dhar:
No, I was going to add, I think, Tom, it would be good also to get you to talk about, because Indico has your view in terms of how you guys are leveraging, especially the piece that I talked about from a document perspective. And it’d be interesting to see how you guys, especially for this quoting question, what are you hearing from the marketplace as well?
Tom Wilde:
Yeah, I think that what we think about now with the arrival of AI to apply to these kinds of things is you need to think about the decisions that the company makes and let’s call underwriting a decision, right? It’s a decision to quote the risk or to bind the risk. Think about that in terms of a supply chain. And that’s really what, when you talk about traceability, when people talk about physical supply chains, they talk about traceability. So your decision supply chain is quite similar. So if you think about the whole continuum, what documents were used, what data came from those documents, what interpretations of that data were used and maintaining what we like to call that chain of custody throughout the whole process, you can always work backwards to figure out what the source was, and you can always work forwards to see what the decision resulted.
And if you can do that, that’s true. By the way, whether you’re using AI or a human-based decision process, that traceability, that explainability I think is really important to make sure your processes are getting better all the time to make sure they’re accurate and robust, that the enterprise policies are being applied. And then in turn, being able to respond robustly to regulatory bodies to make sure that you’re doing it in a fair and consistent and transparent way. I think the first wave of AI was a lot of black boxes. People would show up with models that they would bring to banks or insurance companies and say, Hey, this is our proprietary risk model. Don’t worry about what’s inside the black box. It just works. And that’s no longer going to work. And so AI presents some challenges around traceability and explainability that you have to carefully think through when building that out. But yeah, I like to think about it like the supply chain because you can apply concepts like traceability to it.
Naveen Dhar:
Yeah, I like the analogy. I’m going to actually steal the analogy. That’s awesome. Okay. Okay.
HOST:
Okay. The next question. Yeah. Have you observed any significant differences between the type of company in the insurance sector, insurer, reinsurer, broker, et cetera, and their interest in implementing ai?
Tom Wilde:
That’s a good question. Sylvia, you want to take that one to start? Are there differences or can this be universally applied?
Silvia Signoretti:
I think this could be universally applied from a process standpoint of view, but I suppose you are the expert. Maybe you have a differentiated example, if any, but this could be my take.
Tom Wilde:
I mean, maybe I’d love to hear how you’ve heard this. I think there’s some connection between the volume and complexity of the tasks and how useful AI can be. I think for very low volume tasks, the challenge and expense in trying to apply automation to them may not deliver ROI. So there’s a sweet spot of volume and complexity where AI really delivers. What do you think?
Naveen Dhar:
I think you’re spot on. I think, and also to Sylvia’s point, every one of them uses documents. Every one of them reads to read documents. But I think, Tom, you’re spot on. Just because you can do it does not mean you have to do it unless it makes business sense.
Tom Wilde:
Yeah. It goes back a little bit to Sylvia’s point earlier in the conversation about defining what success looks like, right? And what is the ROI that you’re trying to capture? Because you may find that there’s a sweet spot there of investment and return, and there’s a set of use cases or tasks that fall below that line in the set that are well above that line. I think it’s important to consider that when trying to apply these technologies and automation to any of these tasks. And that does vary back to the question, Talia opposed, I think that does vary by the type of insurance that’s being underwritten. I think that in personal lines like auto and home mobile has had probably a bigger transformative effect than AI because they are able to sort of dictate to the consumer, you have to use our mobile app. And as consumers we sort of like that because it’s convenient. So if you think about how you interact with your life insurance or auto insurer, it’s probably today 90% through a mobile app. Whereas in commercial lines, no one has enough market power to sort of dictate one single mobile app or even a set of forms. Despite many, many attempts at this, it’s still a kind of Tower of Babel problem. You have many, many inputs, many, many outputs and a challenge to sort of normalize that so you can make good decisions from that data.
HOST:
Other questions? One more? Yeah. Can the speakers highlight problems encountered when working with unstructured data and how those problems have been overcome? Are there themes in the challenges that we see?
Tom Wilde:
Sure. You want to go first? We can go around the table.
Naveen Dhar:
Yeah. I think the challenge I think happens to be there is a set of things. So for example, you get lot of documents and then the first step has to classify those documents. Am I going through a financial document? Am I going through a health document? Am I going through a ID document? And then we need a structured versus unstructured. If it’s structured, what are you using? Because today, if I have something which is says social security number where it says SSN on the other form, how do I know it’s the same thing? Or if I have a form which says SSN at the top of the form, but the next form has SSN at the bottom, do I have to rebuild my tools to read it? And so the whole idea that we have gone to gen ai, where we can use that then brings up the question is where is the best piece?
Where do I, because if I use the old technology or not old, well, it’s working so fast as something three months old is also old. So if I was to use structured documents and I work with it, it’s cheaper. But if I use unstructured document or I’m using generative ai, it’s going to be a little more expensive. So where do I draw the line? How do I build the best mesh of structured unstructured with the technology to get the best returns for the dollar I spend? I think that is the main crux that companies have to deal with.
Tom Wilde:
And I think this problem, if you really kind of peel the onion is very complex. You are dealing often with multiple document formats, right? Everything from native formats like Word or Excel to scan documents, PDF to JPEGs, people take photos of documents and send those in. You have images and photos, videos now that are becoming part of the claims landscape for sure. You have handwriting, you have languages. Almost all the carriers that we work with work in multiple countries. And the bundle that comes in as part of the underwriting submission can have any different numbers and variances of those things that have to be captured into data. So it is a deeply complex challenge, especially in commercial lines to do efficiently. It is one of the reasons why good carriers might only write 30, 40, 50% of the submissions they get. Because historically, the complexity of trying to unpack all of those documents and all of those formats has just proven to be we’re going to throw bodies at it. That’s how we’re going to try to do it.
Silvia Signoretti:
I think you mentioned element of the large variety of type of data, the quality, the accuracy, and then what if then some of this data is altered and what type of additional call it controls, safety controls you need to put in place to also monitor that aspect. So probably this is another challenge because I think you mentioned the images that sometimes get uploaded even for claims management. So there is quite an additional level of accuracy that needs to be put in place to manage these. I think
Tom Wilde:
On the accuracy front, and I think Nave, you touched on it earlier in the conversation, your ability to apply what is now characterized as human in the loop is vital. This is not going to be something that gets automated anytime soon. It really is more of a, Microsoft calls it a copilot. We sometimes refer to it as a bionic arm, but that human, the loop piece is vital. This is all about making the team more efficient, more effective than applying a robot to do this work, which is just not on the near term horizon. That especially in commercial lines, which is very complex. Good. Well, I appreciate the conversation. I want to thank you for attending. Sylvia, thank you so much for your perspective. It’s wonderful to talk to you again. I want to thank the audience for spending 45 minutes with us here today. So without further questions, we’ll sign off. As Talia mentioned, we’ll make this available in replay for all of the registered attendees. Thanks again.
Silvia Signoretti:
Thank you all.
Check out the full Unstructured Unlocked podcast on your favorite platform, including:
