The implications of AI in underwriting

Lee Ann Thigpen of Robins Kaplan examines how AI is affecting underwriting in the property & casualty space.

Transcription:

Transcripts are generated using a combination of speech recognition software and human transcribers, and may contain errors. Please check the corresponding audio for the authoritative record.

Patricia Harman (00:04):
Hello and welcome to the Dig in podcast. I’m your host Patti Harman, editor-in-chief of Digital Insurance. Artificial intelligence is rapidly becoming part of our daily lives, whether we want to use it or not. Apparently after decades as a writer, AI thinks it can make much better word choices than I can. So passengers on an airport shuttle recently had a front row seat to me arguing with AI as I had to rewrite an email three times because it kept changing what I’d written to something that was incorrect. They were mildly amused when I explained why I was talking to someone that they couldn’t see. Despite AI being a tool that is still under development, it does have the potential to help us manage large amounts of data, process more mundane tasks, and in some cases, improve communication. Joining me today to discuss AI’s impact on P&C insurance and areas like underwriting, claims, customer service, and fraud detection, is Lee Ann Thigpen, a partner in the Minneapolis office of Robins Kaplan. Lee Ann is an experienced litigator representing both insurers and reinsurers in construction and energy related industries. Thank you so much for joining us today, Lee Ann.

Lee Ann Thigpen (01:26):
Thank you, Patti. I’m happy to be with you.

Patricia Harman (01:29):
So insurers have access to a lot more data and as they try to utilize it and make more accurate underwriting decisions, AI really is playing an important role. What are some of the ways that carriers are using AI as part of the underwriting process? And then are there certain lines where you’re seeing that maybe adoption is just a little bit easier than some others?

Lee Ann Thigpen (01:54):
Sure, sure. So for our listeners, which are probably pretty sophisticated, but just to kind of start at square one, underwriting is the process that an insurer would go through where they analyze the risk. If you think about insurance as essentially a risk game, the insurer is taking a bet that the fire’s probably not going to happen or the car crash is probably not going to happen, and the insured, you, who’s purchasing this policy, is also hoping that’s not going to happen, but that you are going to have a policy in place if the worst does happen. And so it’s essentially a bet and the insurer is analyzing that risk and trying to determine what is that worth. And so if you think about what AI does, which is taking a lot of data and cohesively being able to analyze it and create algorithms from it and actually get smarter because of it, it’s really exciting and amazing how it is impacting the insurance industry.

(03:01):
So when we think about this specific topic of underwriting, and that is the insurer analyzing the risk. So one of those is data analysis. So AI algorithms can take and analyze vast amounts of data from lots of different sources, including demographic information, claims history, credit scores, medical records, even social media activity. And by taking this data, the insurers can then have a better understanding of the risk that they would be signing on for. And what does that mean? It means they’ll have a better way of pricing that risk. We’ll talk about pricing in a minute. Hopefully, AI is going to help have insurers. Hopefully, it reduces the cost of premiums because they can get a really accurate account of what they’re insuring. And also fraud prevention. Again, we’ll talk about that. So another way that AI is being used now in the underwriting sphere is in predictive modeling.

(04:17):
So insurers already, even before AI, would have models that they would look at, they might have models that they get from NOAA because of hurricane modeling and what they could expect to see with a CAT 2 or a CAT 3 hurricane that came through. But AI, I mean the world sort of opens up. They can create models, predictive modeling on any number of areas that would assess the likelihood of future events, whether you’re talking about earthquakes, forest fires, whatever the event might be, even accidents, illnesses. I mean, it’s sort of limitless and the models take into account risk factors, and that helps the insurer assess the probability and severity of a potential loss. Again, it’s this risk game. And then one other area, I mean there’s lots of areas, but one other area that you see AI really coming into the underwriting arena is in risk segmentation.

(05:25):
So what does that mean? It allows an insurer to segment their risk pool more effectively. Why do we like that? Because then if you can be segmented into a smaller pool, that means the insurer can more easily define you, more easily define the risk, and that should result in a more accurate, more streamlined premium because they can better define the risk if the risk can be segmented. And so like I said, that helps them tailor not only underwriting criteria but their pricing and then also coverage options for their insureds so that if they go, well, we’ve segmented this group and we can tell that this type of loss happens very often with single family dwellings in this Tri-County area, we can offer them this endorsement that will help potentially cover that, we can sell it, that kind of thing. As for areas that it is being adopted more readily, as you might imagine it is being adopted more in what I would call the personal lines areas, so homeowners and car. And why is that? Everybody needs home insurance. Everybody needs auto insurance. So there’s just a bigger pool and it makes sense for the insurers to begin taking that data, which is a lot, and beginning to put it into models to affect pricing strategies and risk pools and that kind of thing.

Patricia Harman (07:10):
Thanks. What a great overview and such a great way to kind of lay out the different uses for AI at this point. Do you think that this is going to help expedite the underwriting process then in some instances or even, and I think you may have touched on this, in assessing new risks?

Lee Ann Thigpen (07:27):
Absolutely. In some instances it has resulted in what is being termed automated underwriting, AI-powered. Automated underwriting platforms essentially automate the underwriting process for standard or low risk applications, such as somebody who hasn’t had a wreck in five years, but they need auto coverage in this particular area. It speeds up the decision making on whether the insurer’s going to take that risk and it reduces the need for human intervention. So it’s something that can just be run and spit out a premium and a potential policy and it moves it much faster. So an automated underwriting type of thing. And what that does is it frees up underwriters to be able to focus on more complex cases that would require more human intervention, that kind of thing. Another thing that AI is helping with is what’s called a real-time risk assessment. So as opposed to a human underwriter taking a bunch of information on board, having to run a bunch of either models or read up on a bunch of stuff and then get back with somebody, AI, this real-time risk assessment enables insurers to continuously monitor, update risk in real-time based on changing circumstances such as market conditions, regulations, customer behavior.

(09:04):
And so then the underwriting strategies are dynamic and it allows the insurers to mitigate emerging risks proactively as they come in because AI is taking it on board and populating it within all these various algorithms.

Patricia Harman (09:24):
Okay. You spoke a minute ago about segmenting coverage. How does AI help carriers personalize the coverage when it comes to pricing risks then? Or even selecting what kinds of coverage to offer?

Lee Ann Thigpen (09:41):
So it enables insurers to offer more personalized underwriting decisions and pricing based on individual risk factors, individual preferences, individual behaviors. In theory, this should enhance the customer experience, improve customer satisfaction and retention for the insurers and for the insureds who might want to stay with the same insurance company year by year.

Patricia Harman (10:10):
Alright, yes, that makes a lot of sense. Another area where we’re seeing more adoption of AI is in the claims space. And it’s been interesting because it kind of depends on who I speak with, and I have found that there can kind of be mixed reviews about its use in the claims space. So from your perspective, how is AI being used as part of the claims process and are customers adapting well to its use in this area?

Lee Ann Thigpen (10:37):
So yes, and if you have ever dealt with a customer service chat bot either with an airline or some other company, that’s the use of AI in customer service. So you’ve already dealt with it a little bit, but there are lots of other ways that insurers use AI in the customer service claim space. I shouldn’t say customer service, I should say claim space, although it does affect the customers. So one is automated document processing. If you think about it, a lot of people are familiar with the term OCR. Basically it’s a process that has been around for 20-plus years where a computer can, Adobe offers it now where they read a document and you can word search it for whatever you’re looking for. Well, that same technology, OCR, can be used to automatically extract relevant information from documents such as claim forms, invoices. If you think about a property loss and you’re submitting invoices that can be scraped from the document and put into the claim itself through AI, police reports, medical records, all those can be used with that automated document processing and it helps reduce manual data entry.

(12:03):
You don’t need a human to do it. Computers can do it and accelerates the processing of the claim because you’re not waiting on a person to do it. The computer did it. Another way that AI is used in the claim space is predictive analytics. So AI models can analyze vast amounts of data to predict claim outcomes such as the likelihood of a claim being approved or denied the expected cost of the claim. Whether a claim like that should result in an immediate advanced payment of a certain amount. This helps make insurers make more informed decisions. It helps insurers be happier with the claims process. So if you think about, for example, a house fire and an AI could look, let’s say at a house fire in the southeastern United States, three bedroom house, three bathrooms in this particular area. Based on that, we know that it’s almost always going to be at least a $50,000 loss.

(13:10):
And so that should result in at the very least, an initial claim payment to the insured of $5,000. So in something like that, then the claims person can immediately send out a check, the insured will have that in hand, and there’s some processing that already happens as opposed to the claim being made by a human having to talk and maybe a week or two later getting a check issued. So those are things that AI can help with in the claim space that like a customer might not see firsthand, they might have chat bot and I know, we’re going to talk about chatbots in a little bit, but that AI is helping to revolutionize and speed up the claims process.

Patricia Harman (13:53):
And there isn’t anything wrong with expediting the process at all because it helps both the policyholder as well as the insurer. Absolutely. It just makes so much sense. In that instance, there’ve been some concerns that the increased adoption of technology will make it easier for bad actors to perpetrate insurance fraud. Is AI able to identify fraudulent claims and maybe mitigate their impact at all?

Lee Ann Thigpen (14:22):
Absolutely. It’s again, another kind of amazing area where AI can really help with detecting fraud. And so some people might be listening and going, well, I mean, why do I care about that? As an insured, I’m not going to commit fraud, but why does it matter to me? Well, here’s why it matters is that there is an unbelievable amount of data shows, I mean, I think it’s like in the billions per year of insurance fraud that happens. Well, if AI can reduce some of that insurance fraud, then it reduces the costs for everyone, reduces the cost of premiums, reduces everything. And so we all should want it not only for the moral thing because its right, but also because it should reduce the cost for everyone. So here’s a few ways that AI is in this fraud detection space and is helping with detecting fraud. One is called anomaly detection.

(15:21):
So AI, and again, I think this is fascinating. AI algorithms analyze these huge amounts of historical claims to identify patterns and then anomalies. And the anomalies are what indicate potential fraud. And so if you think about that compared to say a 35-year-old grizzled insurance claims adjuster veteran, yeah, they’ve seen a lot, but it takes 35 years for them to get there. Whereas AI’s got it day one on the job and can see anomalies that maybe even that somebody with that much experience wouldn’t see. Again, predictive modeling, just like you would build predictive modeling with underwriting, same with fraud. It enables insurers to build predictive models that assess the likelihood of a claim being fraudulent or just the amount of the claim being fraudulent. And it looks at claim histories, behavioral patterns, the type of loss, all those things. And so it helps the insurer allocate resources and whether they need to look at the claim more closely, that kind of thing. Again, pattern recognition. The AI can recognize patterns and techniques used in fraudulent claims such as staged accidents, inflated medical bills, false documentation. And because AI is continually learning from new data, it can stay on top of the new schemes and can adapt to evolving fraudulent schemes and improved detection accuracy quicker than perhaps a human would because they might not be aware of it or hear about it until it actually happens to ’em.

(17:15):
I mean, image and video analysis, it’s amazing. AI, while it can create fake pictures, can also detect when pictures have been doctored or faked. And so that’s kind of fascinating in itself. And then I think this is also interesting and really should be helpful for all of us. AI facilitates collaboration between insurers, law enforcement agencies and other stakeholders in the fight against insurance fraud. So as we sit here now, insurers may share information with a broker about how much a prior claim was, but with respect to some of the modeling that’s done, particularly in underwriting, that is proprietary and usually is not shared at all. If you think about data being able to be shared in AI that is helpful, should be helpful to everybody, to all insurers. That’s a very exciting thing again, so that it hopefully reduces people’s premiums, reduces the overall cost of doing business for everybody and for all players, insurers, law enforcement agencies and other stakeholders to be able to identify fraud and fraudulent trends and then share those best practices and coordinate with each other to combat fraud more effectively.

Patricia Harman (18:45):
It really makes a lot of sense, especially when you look at it across the entire insurance ecosystem. And then the fact that AI, like you were saying, can detect those anomalies and it can say, well, you had a claim here in California, had this one, it was the exact same claim. Look at the difference. And that to me is just so fascinating and a good use of AI at this point. I was at a conference earlier this year and I was really surprised and encouraged to see that companies are using AI to help with their risk mitigation efforts. And what are some of the ways companies are using AI to proactively identify risks? And does it seem to work maybe more effectively for certain types? And I’m thinking in terms of whether it’s commercial versus residential. Sometimes you’ll hear people say, oh, well we can do this on the personal line side because there’s so many claims like that, but commercial claims tend to be a little bit more complex. So I was just wondering what you were seeing from your perspective?

Lee Ann Thigpen (19:49):
Yes. So risk assessment algorithms can analyze these huge amounts of data to assess the risks associated with not only insuring individuals, but on the commercial side what the properties are or specific kinds of businesses. And by taking data from various sources, including historical data and external risk factors, AI systems can identify high risk entities, help insurers take proactive measures, hopefully to potentially mitigate losses. I would say it probably is that you’re seeing AI more in the personal lines area because of the greater number of risks that are out there. Everybody has to have it. However, the commercial lines and particularly in the energy area, I think it is very exciting and I think you’re going to see a huge expansion of using AI because if you think about the technological abilities that we have with respect to drones, with respect to infrared type scans, if you think about a refinery that has a certain kind of piping system metal, if you are able to use a drone to go in and take pictures and then maybe an infrared scan to, or even a robot to go over the pipes and test the thickness of things, and then all that gets dumped into the AI model based on how old the plant is and what is it that it’s producing and what is its historical risk and what losses has it had in the past, and what kind of maintenance does this plant do because all those things go into the underwriting assessment and feed that into AI.

(21:53):
You’re going to see these same predictive models in the energy sector and the refinery, which would be refining turbines, all that kind of thing. And then in the commercial, just if you think about a commercial building ethos or the kind of roofs that you have, all that sort of thing, it it’s coming and it’s really exciting to think about how it will impact every facet of insurance coverage from underwriting to premium, to claims processing to the payment of claims. So it is really just beginning, but it’s going to explode and it’s going to provide such a useful tool for the insurance industry and for insurance I think as well.

Patricia Harman (22:49):
Listening to you talk, I think of all the people who say, oh, insurance is just dull and boring and we’re not on the cutting edge of technology and hearing you explain this, it’s like there is so much going on in this space and just so much potential. So for me as an insurance nerd, it’s really exciting.

Lee Ann Thigpen (23:09):
Same. It is very exciting. And I think we’re going to talk about this in a little bit, but I think that is sort of the message to our compatriots and colleagues out there in the insurance industry, whether they work for insurers or are a vendor that services the insurance industry, which is what I would consider myself as an insurance coverage lawyer, but engineers, claims adjusters, independent claims adjusters, we should be excited about this and embrace it and not be afraid of it. We cannot be Luddites. I mean, it is coming, it’s here. And so we need to educate ourselves and be on the forefront of it because it’s coming. And like I said, it’s already here in many respects. So let’s get on board with it and figure out the ways we can use it to not only improve our company or our clients work, but also our own jobs. How do we make them better and provide better service to our customers or our clients?

Patricia Harman (24:12):
Very true. One of the other areas where we’re seeing some adoption of AI is in the customer experience area, which I think is really interesting, whether it’s the first notice of loss or the chat bots to get immediate answers. I was trying to get something yesterday and the chat bot popped up and I was like, okay, this is what I want. And they’re like, use me instead of trying to call. And it’s just very interesting to see how technology is adjusting and making us as consumers adjust too. So are there areas within the property space where you’re seeing a real adoption of AI, and if so, how is it affecting that customer experience?

Lee Ann Thigpen (24:54):
I have seen it with respect to the first notice of claim, the uploading of claims documents, and I haven’t really seen it. And maybe there are some insurers out there that are doing it with chatbots, particularly with respect to claims, although that’s going to come because I would think that as with most things, there are frequently asked questions that a chat bot can answer, but I have seen it with respect to notification of claims, uploading documents and making sure that the claim information is in. That has definitely been adopted, and I think it makes it so much easier than having to run copies or scan things in and hope they get it or really back in the old days mailing a bunch of stuff in and hoping somebody got it or looked at it. And so it’s coming with the rest of it. But certainly with respect to the notification of claims and provision of claims documents for sure.

Patricia Harman (26:00):
What are some of the mistakes you’ve seen as companies are trying to add AI into their processes? Is there anything you’re like, oh, I would’ve maybe rethought that or just things that companies should be aware of because we want to rush to adopt everything, but we also have to take the time to assess what risks could that incur for us as well.

Lee Ann Thigpen (26:24):
And I would say like we were just talking about, it’s still a little bit early specifically in the claims process, but I would say the one if I had a complaint, not really a complaint, but a suggestion is to make sure, and this is talking to our insurers as an insured or a customer of an insurance company, is to make sure that you road test very well whatever product or service you’re putting out there that is an AI service. Yes, it should, yes, insurance company, it should help you with doing things faster and not having to hire as many humans and that sort of thing, but make sure that you have road tested it in a variety of different ways with a variety of different humans trying to access it. Because even though you may think it’s intuitive, it’s not necessarily intuitive or somebody has a question that is not answered in your template that’s up there.

(27:29):
And I’ve seen that a couple of times and also experienced it myself, not necessarily in the insurance context, but with chat bots or these bots that you’re supposed to interact with. And it doesn’t answer the question that I have, which is it seems to be a frequently asked question and then I can’t get to either a human or another avenue to try to resolve the problem. So that would be my biggest takeaway is to road test it with a lot of different people and don’t just rely on the IT nerd who puts it together for you, use real people to road test it and tweak it.

Patricia Harman (28:09):
I would agree with that too. I know there are times that I’m just like, I just want a human, and they’re like, I don’t understand what you’re saying. A human, a real live person, please. Yes, exactly. And I’m thinking, I don’t think I have a heavy accent, but maybe your AI just doesn’t understand what I’m saying.

Lee Ann Thigpen (28:30):
Right. Or for whatever reason, it just does not provide an answer to that topic, and there’s no way to get out of the chatbot or no other avenue to get your answer.

Patricia Harman (28:43):
Very true. So what are the possibilities about the use of AI in the insurance industry that really excite you the most? I mean, we’ve covered a lot, but I was wondering if there’s anything in particular that you’ve been watching or you’re like, I think this is so cool.

Lee Ann Thigpen (28:57):
So a couple of the things that I think are really cool is the ability for insurers to take data that is usage based and factor that into an algorithm that would then inform on premiums for a policy. So an example of that would be in the auto insurance space, AI allows for usage-based insurance models where premiums are based on actual driving behavior rather than traditional metrics. And this data-driven approach helps in creating fairer pricing structures for people who drive the speed limit and don’t run red lights or whatever. And it would also encourage safer driving habits. And if you think about it, most of us have smart cars now, and so it would allow for data to be uploaded and used in the creation of an algorithm that would affect premium pricing. Same is true with health insurance now. Okay. So AI being used to analyze data from wearable devices and health apps to offer personalized health advice, predict health outcomes, optimize treatment plans, but also would help insurers potentially manage costs and improve outcomes.

(30:19):
So I mean, it’s crazy to think that your Fitbit could potentially inform a premium, but we saw the advent of some of these things with wellness visits so that insurers could get a feel for where people were. So this is just the AI version and the 15, 20-year version later. Another would be behavioral insights and engagement. So this would be where they take information from cookies and from social media activity and analyzing behavior and interactions to gain insights into preferences and trends. And so it would probably more tailor to insurers marketing their products, but also developing products for customers that don’t exist but need to exist. I hear people now talking about pet insurance all the time. Well, I mean, 15 years ago nobody even knew what pet insurance really was, and now it’s a real thing. That’s a product that was developed based on talking to people, observing people spending more time and money on their pets, that kind of thing. So it’s exciting and interesting.

Patricia Harman (31:39):
It really is. I was talking to one company and they were talking about developing technology to help mitigate workers’ comp injuries. I mean, think about that. If you’re working somewhere in a factory or a warehouse and there are repetitive movements that you have to make, it can help alleviate those or say, Hey, if you shift your body just a little bit, this makes a huge difference. So from that perspective, I think it’s really fascinating too.

Lee Ann Thigpen (32:06):
Absolutely.

Patricia Harman (32:07):
So we’ve covered a lot in the last 30 minutes. Is there anything that I haven’t asked you that you think is important for our audience to know? I think

Lee Ann Thigpen (32:15):
That, again, people do not need to be afraid of AI. It is coming and so we need to embrace it. You hear concerns about privacy and data security bias potentially where you go, well, I’m the outlier. You’re putting me in this little segment, but I’m the outlier or job displacement. I hear about this a lot, is that people are concerned that the automation is going to lead to job displacement and that they’ll be out of a job. But what we need to consider is that we can monitor for bias and fix technology, add things into the algorithm to correct for that, implement strategies on data protection. I’m saying this as we are getting news that every social security number in the country has been hacked, but continuing to push for data privacy regulations and then human oversight to make sure that we are not using AI, but combining it with human expertise that ensures that complex or nuanced situations are handled appropriately and given the time and care that they need, and that there’s a balance maintained between technology and human judgment.

Patricia Harman (33:38):
Well, thank you so much, Lee Ann, for sharing your insights with our audience. Thank you for listening to the Dig In podcast. I produced this episode with audio production by Adnan Khan. A special thanks this week to Lee Ann Thigpen of Robins Kaplan for joining us. Please rate us, review us, and subscribe to our content at www.digin.com/subscribe From Digital Insurance, I’m Patti Harman, and thank you for listening.