Insurance Claims Are AI's Messiest Data Problem — Denys Linkov, Wisedocs

Okay, thank you for Okay, thank you for joining us for the joining us for the joining us for the last last workshop today. So, I'm workshop today. So, I'm Dennis. I'll talk Dennis. I'll talk Dennis. I'll talk a little bit about a little bit about a little bit about insurance claims data and insurance claims data and insurance claims data and how powerful it how powerful it how powerful it can be can be can be for for for enterprise enterprise enterprise AI programs. So let's AI programs. So let's AI programs. So let's begin. Who here begin. Who here begin. Who here works in insurance? works in insurance? works in insurance? Several people. Who here Several people. Who here Several people. Who here wants to work in wants to work in wants to work in insurance? Okay, insurance? Okay, insurance? Okay, interesting. Several people. interesting. Several people. interesting. Several people. OK. Insurance OK. Insurance is a very interesting is a very interesting is a very interesting sector of our sector of our sector of our economy, where people economy, where people economy, where people often get into often get into often get into by accident, but it by accident, but it by accident, but it ensures the functioning of ensures the functioning of ensures the functioning of such an important such an important such an important part of our part of our part of our ecosystem. We'll ecosystem. We'll ecosystem. We'll talk a little about talk a little about talk a little about insurance in general. insurance in general. insurance in general. And ultimately, we will And ultimately, we will And ultimately, we will focus this focus this focus this conversation on how to conversation on how to conversation on how to improve our lives in improve our lives in improve our lives in general. So let's general. So let's general. So let's focus on the focus on the focus on the medical requirements. We will medical requirements. We will medical requirements. We will focus on focus on focus on improving improving claims processing and claims processing and patient care outcomes, and I will patient care outcomes, and I will patient care outcomes, and I will keep keep keep coming back to coming back to coming back to that. I think that. I think that. I think that in many that in many that in many discussions about fintech discussions about fintech discussions about fintech and insurance, things and insurance, things and insurance, things sometimes become too sometimes become too business-oriented. That's why I always business-oriented. That's why I always want everyone to want everyone to want everyone to focus on focus on focus on this aspect. A little this aspect. A little this aspect. A little about myself. I am Dennis. I about myself. I am Dennis. I about myself. I am Dennis. I lead the AI ​​program at lead the AI ​​program at lead the AI ​​program at Wisdocs. We help Wisdocs. We help Wisdocs. We help companies process companies process companies process complex health complex health complex health insurance claims. I have been in insurance claims. I have been in insurance claims. I have been in this position for two this position for two this position for two years, and since 2024 we years, and since 2024 we years, and since 2024 we have seen a huge have seen a huge have seen a huge transformation.

transformation. You may have seen You may have seen news about news about news about health coverage health coverage health coverage and insurance and insurance and insurance in general. The entire in general. The entire in general. The entire transformation transformation transformation around GLP-1 in our around GLP-1 in our around GLP-1 in our economy and healthcare system economy and healthcare system has also changed how has also changed how insurance sets insurance sets insurance sets prices. We also see prices. We also see prices. We also see that that that spousal insurance and insurance coverage in general spousal insurance and insurance coverage in general continue continue to evolve due to to evolve due to to evolve due to rising costs and rising costs and rising costs and other factors other factors other factors changing the approach to changing the approach to changing the approach to insurance. So insurance. So insurance. So the report the report the report will consist of will consist of will consist of three main three main three main sections. The first is about the sections. The first is about the sections. The first is about the technological technological technological transformation we are experiencing transformation we are experiencing transformation we are experiencing . Second—I'll . Second—I'll . Second—I'll delve a little deeper into the delve a little deeper into the delve a little deeper into the insurance space, insurance space, insurance space, discussing some of the discussing some of the discussing some of the details, but it'll details, but it'll details, but it'll just be a brief introduction. just be a brief introduction. just be a brief introduction. And then I'll talk about how And then I'll talk about how And then I'll talk about how AI will help make AI will help make AI will help make insurance and the insurance and the insurance and the patient experience even patient experience even patient experience even better. Let's talk better. Let's talk better. Let's talk about technological about technological about technological transformation. transformation. transformation. We've probably all We've probably all We've probably all embarked on embarked on embarked on some form of digital some form of digital some form of digital transformation in our transformation in our transformation in our companies, whether companies, whether we work in smaller startups or large or large or large enterprises. enterprises. enterprises. The problem is that we The problem is that we The problem is that we never never never completed this completed this completed this transformation. AI transformation. AI transformation. AI has hit us in 2023 and has hit us in 2023 and has hit us in 2023 and 2024, but most 2024, but most 2024, but most companies haven't companies haven't companies haven't completed their migration to the completed their migration to the completed their migration to the cloud, started cloud, started cloud, started using using using big data to its full big data to its full big data to its full potential, and potential, and potential, and deployed a clean deployed a clean deployed a clean data lake, right data lake, right ? Who here has a ? Who here has a ? Who here has a data lake that you are data lake that you are data lake that you are proud of? OK.

proud of? OK. proud of? OK. Very, very few people. Very, very few people. Very, very few people. Is it even possible Is it even possible Is it even possible to create a data lake to create a data lake to create a data lake that you can be proud of, that you can be proud of, that you can be proud of, right? And on top of right? And on top of right? And on top of that, there's so much hype around AI right now that, there's so much hype around AI right now that, there's so much hype around AI right now , and , and , and we need to we need to we need to make sure we make sure we make sure we have the right have the right have the right prerequisites before we prerequisites before we prerequisites before we deploy real-world deploy real-world deploy real-world applications. So, in applications. So, in applications. So, in insurance and insurance and insurance and many other many other many other industries, paper was the industries, paper was the industries, paper was the main way of main way of main way of processing data, processing data, processing data, right? You received right? You received right? You received stacks of papers stacks of papers stacks of papers while processing a medical while processing a medical while processing a medical application. You application. You application. You were organizing them. You were organizing them. You were organizing them. You had these huge had these huge had these huge tables piled high with them. tables piled high with them. tables piled high with them. You used You used You used cards to divide cards to divide cards to divide them into specific sections. And them into specific sections. And them into specific sections. And this makes it very difficult to this makes it very difficult to this makes it very difficult to extract anything extract anything extract anything outside of the statement itself outside of the statement itself , because how can you , because how can you , because how can you cross-analyze the statements cross-analyze the statements cross-analyze the statements when everything is on paper, when everything is on paper, when everything is on paper, right? You can right? You can right? You can put everything into put everything into put everything into as many as many as many folders as you want, but it's very folders as you want, but it's very folders as you want, but it's very difficult. We still haven't difficult. We still haven't difficult. We still haven't left the paper age left the paper age . There are still many practices in the world . There are still many practices in the world . There are still many practices in the world that that that use paper use paper : whether in a : whether in a : whether in a doctor's office, in doctor's office, in doctor's office, in legal departments, legal departments, legal departments, or in insurance or in insurance or in insurance companies themselves. So we have companies themselves. So we have companies themselves. So we have n't given up on n't given up on n't given up on paper yet, but we're paper yet, but we're paper yet, but we're close to it. close to it. Speaking of Speaking of digitalization, we still haven't digitalization, we still haven't given up on given up on faxes. The fax market faxes. The fax market faxes. The fax market continues to grow.

continues to grow. continues to grow. It is worth over It is worth over It is worth over $7 billion $7 billion $7 billion worldwide, which is always worldwide, which is always worldwide, which is always surprising. I mentioned surprising. I mentioned surprising. I mentioned this in my talk this in my talk this in my talk at “An Engineer SF” at “An Engineer SF” at “An Engineer SF” last year, and last year, and last year, and nothing has changed. nothing has changed. nothing has changed. Fax is still Fax is still Fax is still used. used. used. We are currently We are currently We are currently integrating faxes into integrating faxes into integrating faxes into our digital systems, our digital systems, our digital systems, converting them into a converting them into a converting them into a fairly clear format, fairly clear format, fairly clear format, but this still does not but this still does not but this still does not solve the problem of solve the problem of solve the problem of information exchange in the information exchange in the information exchange in the healthcare system healthcare system healthcare system . If . If . If you think about you think about you think about the speed of the speed of technology adoption, in technology adoption, in 2008, only 8% used electronic 2008, only 8% used electronic 2008, only 8% used electronic health records (EHR) health records (EHR) . And only with the beginning of the . And only with the beginning of the COVID pandemic did we COVID pandemic did we COVID pandemic did we reach reach reach 80-90% in EHR implementation. 80-90% in EHR implementation. 80-90% in EHR implementation. And again, many And again, many And again, many hospitals still hospitals still hospitals still operate on paper. operate on paper. operate on paper. So we haven't fully So we haven't fully So we haven't fully passed the era of passed the era of passed the era of digitalization and digitalization and digitalization and digital digital digital transformation yet, and everyone transformation yet, and everyone transformation yet, and everyone already wants already wants already wants to use AI. to use AI. If you think about the If you think about the types of forms types of forms types of forms people receive people receive people receive when processing when processing when processing insurance claims, insurance claims, insurance claims, here are three sample forms here are three sample forms here are three sample forms from three major from three major from three major states for states for workers' compensation.

Workers' compensation is when Workers' compensation is when you get injured on the you get injured on the you get injured on the job and you file job and you file job and you file a claim, "Hey, I a claim, "Hey, I a claim, "Hey, I hurt my arm when I hurt my arm when I hurt my arm when I fell, or I was operating a fell, or I was operating a fell, or I was operating a forklift and forklift and forklift and something went wrong." And something went wrong." And something went wrong." And none of these shapes none of these shapes none of these shapes look the same, look the same, look the same, right? So in the 50 different right? So in the 50 different right? So in the 50 different states in the US, you states in the US, you states in the US, you typically have 50 typically have 50 typically have 50 different types of forms. different types of forms. different types of forms. They all require They all require They all require different information, different information, different information, they are not they are not they are not standardized. standardized. standardized. They are fairly They are fairly They are fairly standardized in standardized in standardized in the sense that each the sense that each the sense that each form looks form looks form looks the same, so you can the same, so you can the same, so you can achieve some achieve some achieve some certainty, but certainty, but certainty, but even then the data in even then the data in even then the data in them varies. them varies. them varies. So, even within the So, even within the So, even within the United States, let alone United States, let alone United States, let alone other countries around the world, we other countries around the world, we other countries around the world, we collect information collect information collect information differently, and those differently, and those differently, and those versions sometimes versions sometimes versions sometimes change. If change. If change. If we talk about what we talk about what we talk about what worries people in the worries people in the worries people in the insurance industry, it's the insurance industry, it's the loss ratio, loss ratio, right? That is, how many right? That is, how many right? That is, how many premiums do you collect and premiums do you collect and premiums do you collect and how much money do you how much money do you how much money do you pay out? I'll pay out? I'll pay out? I'll talk a little talk a little talk a little more about these numbers more about these numbers more about these numbers and their significance, but and their significance, but and their significance, but in general, this is what in general, this is what in general, this is what determines the financial determines the financial determines the financial performance of performance of performance of an organization, an organization, an organization, right? When you report right? When you report right? When you report quarterly or quarterly or quarterly or annual earnings, you annual earnings, you annual earnings, you focus on focus on your your loss ratio, loss ratio, how well you how well you how well you forecast risks, forecast risks, forecast risks, how you how you how you underwrite, how you underwrite, how you underwrite, how you fight fight fight fraud, and how you fraud, and how you fraud, and how you manage expenses.

manage expenses. manage expenses. So companies want to So companies want to So companies want to cut costs and cut costs and cut costs and make more make more make more money, right? This is money, right? This is money, right? This is just capitalism in the just capitalism in the just capitalism in the general sense general sense . And if we look . And if we look . And if we look at the statistics, at the statistics, at the statistics, we usually we usually we usually hover around a hover around a hover around a 50% 50% 50% loss ratio in terms of the amount of loss ratio in terms of the amount of loss ratio in terms of the amount of money that money that money that has to be has to be has to be paid out for paid out for paid out for insured events insured events , compared to the , compared to the , compared to the premiums received. premiums received. Now I'm adding all these Now I'm adding all these contextual contextual contextual details, and in my entire details, and in my entire details, and in my entire speech I only mentioned AI speech I only mentioned AI speech I only mentioned AI once. Again once. Again once. Again , it's very important , it's very important , it's very important to think about the to think about the to think about the problem we're problem we're problem we're trying to solve trying to solve and only choose AI and only choose AI and only choose AI when it's the when it's the when it's the right solution. right solution. To get the right To get the right AI solutions, we AI solutions, we AI solutions, we need to go need to go need to go through the right through the right through the right transformations, transformations, transformations, right? We need to right? We need to right? We need to have the right goals have the right goals have the right goals in mind. We in mind. We in mind. We need teams that need teams that need teams that can clearly can clearly can clearly articulate what exactly articulate what exactly articulate what exactly we want to do with AI. we want to do with AI. we want to do with AI. We can't just We can't just We can't just let people let people let people go on stage and go on stage and go on stage and shout "SH, SH, SH, shout "SH, SH, SH, " right? We " right? We " right? We need the right need the right need the right people to people to people to drive this drive this drive this transformation, and we transformation, and we transformation, and we need the right need the right need the right framework: governance, framework: governance, framework: governance, security, security, security, infrastructure, and infrastructure, and infrastructure, and data. data.

data. Now this is the question. Now this is the question. Now this is the question. For those working in the For those working in the For those working in the corporate corporate corporate sector: who among you could sector: who among you could sector: who among you could get a get a get a database or virtual database or virtual database or virtual machine within a week machine within a week machine within a week of requesting it? OK. of requesting it? OK. of requesting it? OK. Who couldn't? Okay, that's Who couldn't? Okay, that's Who couldn't? Okay, that's progress. Who among you progress. Who among you progress. Who among you could deploy could deploy could deploy an application to a an application to a an application to a production production production environment in a environment in a environment in a week? New week? New week? New app. OK. Who could app. OK. Who could app. OK. Who could n't? Okay. So, n't? Okay. So, n't? Okay. So, we still have a we still have a we still have a lot of work to do. Who lot of work to do. Who lot of work to do. Who can choose a new can choose a new can choose a new AI vendor in a AI vendor in a AI vendor in a week? Ah, here is week? Ah, here is week? Ah, here is the answer. No the answer. No the answer. No hands. Yes. So it still hands. Yes. So it still hands. Yes. So it still takes us time to takes us time to takes us time to go through go through go through all these different processes. all these different processes. all these different processes. They are certainly They are certainly They are certainly better than they were 10-20 better than they were 10-20 better than they were 10-20 years ago in terms of years ago in terms of years ago in terms of digital digital digital transformation, but transformation, but transformation, but many companies are still many companies are still many companies are still stuck on stuck on stuck on governance and governance and governance and security. And a lot of it has security. And a lot of it has security. And a lot of it has to do with the to do with the to do with the factors here. Which of factors here. Which of factors here. Which of you has had management you has had management you has had management make similar make similar make similar requests? Give me a requests? Give me a requests? Give me a data set that would be data set that would be data set that would be ideal. Yes? ideal. Yes? ideal. Yes? A few people, right? We we we have A few people, right? We we we have A few people, right? We we we have seen this seen this seen this before. Hey, can you before. Hey, can you before. Hey, can you quickly pull quickly pull quickly pull that report for me? that report for me? that report for me? It's much easier now with AI It's much easier now with AI It's much easier now with AI , but you have to , but you have to , but you have to assume that this assume that this assume that this data exists, data exists, data exists, right? So, what if we right? So, what if we right? So, what if we come up with a more come up with a more come up with a more complex query, right complex query, right ? Let's say we have an ? Let's say we have an ? Let's say we have an insurance claim insurance claim insurance claim for a specific patient for a specific patient for a specific patient and their condition, in this and their condition, in this and their condition, in this case an case an case an ankle injury, and we ankle injury, and we ankle injury, and we want to assign want to assign want to assign them to a doctor who has a them to a doctor who has a treatment success rate in the top 50%, treatment success rate in the top 50%, right? Within some right? Within some right? Within some recent time recent time recent time frame. This query is frame. This query is frame. This query is quite complex, but quite complex, but quite complex, but what if you could just what if you could just what if you could just write it for

write it for write it for your dataset and your dataset and your dataset and get the result, get the result, get the result, right? I would be impressed right? I would be impressed , right? Data is , right? Data is , right? Data is very, very complicated in a very, very complicated in a very, very complicated in a corporate corporate corporate environment. So, in order for environment. So, in order for environment. So, in order for us to get us to get us to get these kinds of results and these kinds of results and these kinds of results and insights, we insights, we insights, we need to need to need to structure our structure our structure our data systems in a clearly data systems in a clearly data systems in a clearly defined way. So defined way. So defined way. So let's talk a little let's talk a little let's talk a little about about about medical insurance medical insurance medical insurance claims. In general, I have claims. In general, I have claims. In general, I have grouped grouped grouped insurance based on the insurance based on the NAIC classification. We have property and accident insurance, accident insurance, life insurance, life insurance, life insurance, health and general health and general health and general medical insurance. medical insurance. medical insurance. I will talk about I will talk about I will talk about property property property and and and casualty insurance, casualty insurance, casualty insurance, as those are as those are as those are usually the usually the usually the most complex requirements. If most complex requirements. If most complex requirements. If you think about P&C, I you think about P&C, I you think about P&C, I mostly worked in the mostly worked in the accident insurance part. That is, accident insurance part. That is, auto insurance, auto insurance, auto insurance, various types of various types of various types of liability, liability, liability, general general general liability, liability, product liability, workers' compensation. There is also the workers' compensation. There is also the whole property side, whole property side, whole property side, which has its own set of which has its own set of which has its own set of complexities. But complexities. But complexities. But overall, it's a overall, it's a overall, it's a $1 trillion market, right $1 trillion market, right ? It is ? It is ? It is premium-based, and premium-based, and premium-based, and reinsurance has reinsurance has reinsurance has its own forms.

its own forms. its own forms. So if we talk So if we talk So if we talk about these blue blocks, about these blue blocks, about these blue blocks, many of them many of them many of them handle complex handle complex handle complex medical claims. Someone medical claims. Someone medical claims. Someone was injured due to a was injured due to a was injured due to a product defect, product defect, product defect, someone was in someone was in someone was in a car accident, a car accident, a car accident, someone was injured at someone was injured at someone was injured at work. Currently, reviewing work. Currently, reviewing work. Currently, reviewing an application and receiving an application and receiving an application and receiving a response from the patient a response from the patient a response from the patient or applicant takes or applicant takes or applicant takes a lot of time. At a lot of time. At a lot of time. At best, best, best, it takes a few it takes a few it takes a few weeks. In the worst weeks. In the worst weeks. In the worst case, it could case, it could case, it could drag on for drag on for drag on for several years. Depending several years. Depending several years. Depending on the complexity on the complexity on the complexity of the situation, the complexity of the of the situation, the complexity of the of the situation, the complexity of the regulatory regulatory regulatory environment, and the level of environment, and the level of environment, and the level of litigation in this litigation in this litigation in this area. And even after area. And even after area. And even after receiving medical receiving medical receiving medical services, the services, the services, the case can case can case can go on and go on and go on and on. So, on. So, on. So, it's a complex process, it's a complex process, it's a complex process, and I want us all to and I want us all to and I want us all to imagine how much more imagine how much more imagine how much more efficient the healthcare efficient the healthcare efficient the healthcare system could be system could be if we reduced this by a if we reduced this by a factor of ten, right? factor of ten, right? factor of ten, right? You were injured at You were injured at You were injured at work, you filed a claim, you work, you filed a claim, you work, you filed a claim, you went through the went through the went through the process, and everything became process, and everything became process, and everything became digital and digital and digital and streamlined. So why is streamlined. So why is streamlined. So why is it taking so long it taking so long it taking so long ? There are many ? There are many ? There are many stages of transferring stages of transferring stages of transferring cases even in the cases even in the cases even in the medical aspect, not to medical aspect, not to medical aspect, not to mention the mention the mention the legal side or legal side or legal side or directly directly directly insurance, right?

insurance, right? insurance, right? So, when you So, when you So, when you are injured in such a are injured in such a are injured in such a situation and situation and situation and need need need insurance insurance insurance compensation for a compensation for a compensation for a complex claim, you complex claim, you complex claim, you need to get a need to get a need to get a diagnosis and visit diagnosis and visit diagnosis and visit various specialists. various specialists. various specialists. Sometimes experts have Sometimes experts have Sometimes experts have different opinions, don't different opinions, don't different opinions, don't they? Then you need to they? Then you need to they? Then you need to make sure you make sure you make sure you have the proper have the proper have the proper coverage, whether it's coverage, whether it's coverage, whether it's primary care or primary care or primary care or one of the additional one of the additional one of the additional forms of forms of forms of liability. You liability. You liability. You undergo a feasibility review, undergo a feasibility review, where doctors must where doctors must agree among themselves on agree among themselves on agree among themselves on the necessity of the the necessity of the the necessity of the procedure. We have procedure. We have procedure. We have treatment and treatment and treatment and billing billing . You're trying . You're trying . You're trying to figure out what to figure out what to figure out what your score means, right your score means, right ? With an explanation of ? With an explanation of ? With an explanation of benefits and benefits and benefits and insurance coverage. insurance coverage. insurance coverage. You receive payments You receive payments You receive payments and continue and continue and continue treatment. There are still so treatment. There are still so treatment. There are still so many additional many additional many additional links that can links that can links that can be added, and crossroads, be added, and crossroads, be added, and crossroads, depending on the steps, depending on the steps, depending on the steps, but overall it is but overall it is but overall it is not a simple process at all. And not a simple process at all. And not a simple process at all. And along the way, a lot along the way, a lot along the way, a lot can go wrong. can go wrong. can go wrong. I just listed I just listed I just listed about 15 points about 15 points about 15 points that come to mind that come to mind , but it's related to , but it's related to , but it's related to human resources issues, human resources issues, human resources issues, insurance insurance insurance companies, where companies, where companies, where insurance agents are insurance agents are insurance agents are often fired often fired often fired or or or retired, and retired, and retired, and technology, where technology, where technology, where people are constantly people are constantly people are constantly processing the same processing the same processing the same data. This also has data. This also has data. This also has to do with to do with to do with situations where situations where situations where applicants are applicants are applicants are under stress, sometimes under stress, sometimes under stress, sometimes make make make mistakes, and act not mistakes, and act not mistakes, and act not entirely rationally, entirely rationally, entirely rationally, right? Because if you've right? Because if you've right? Because if you've been in a traumatic been in a traumatic been in a traumatic car accident, car accident, car accident, you're not going to be the you're not going to be the you're not going to be the most rational most rational most rational person. So, person. So, person. So, let's talk let's talk let's talk about how technology about how technology about how technology can help us.

can help us. Let's take a look at the Let's take a look at the process of filing a process of filing a workers' compensation claim. Has workers' compensation claim. Has anyone seen a workers' compensation lawsuit before anyone seen a workers' compensation lawsuit before ? ? A few people, okay A few people, okay . So, usually, when . So, usually, when . So, usually, when someone is injured, someone is injured, someone is injured, they should report they should report they should report it to the it to the it to the workplace and also workplace and also workplace and also file the appropriate file the appropriate file the appropriate paperwork. The three paperwork. The three paperwork. The three forms I showed— forms I showed— from Texas, New York, and from Texas, New York, and from Texas, New York, and California—are California—are California—are typical typical typical initial initial initial intake forms. And you intake forms. And you intake forms. And you go through this go through this go through this process, visiting process, visiting process, visiting different specialists: different specialists: different specialists: whether in the emergency whether in the emergency whether in the emergency department, or for department, or for department, or for additional additional additional treatment, etc. And your treatment, etc. And your treatment, etc. And your request is not always request is not always request is not always processed processed processed immediately until you immediately until you immediately until you submit documents submit documents submit documents later. Now that later. Now that later. Now that you're going through this you're going through this , there are different types of , there are different types of , there are different types of insurance claims. insurance claims. insurance claims. I've created some I've created some I've created some synthetic synthetic synthetic examples here, but they're examples here, but they're examples here, but they're pretty close to what pretty close to what pretty close to what you'd you'd you'd encounter in encounter in encounter in real-world practice. real-world practice. real-world practice. So you go So you go So you go through all these through all these through all these treatments, you get treatments, you get treatments, you get different procedures, you different procedures, you different procedures, you may be in may be in or out of the insurance company's network, you may or out of the insurance company's network, you may travel for travel for travel for it, and sometimes you it, and sometimes you it, and sometimes you get get get conflicting conflicting conflicting information. And as information. And as information. And as we go we go we go through each of these through each of these through each of these processes, all the processes, all the processes, all the documents look documents look documents look different. Every different. Every different. Every hospital has its hospital has its hospital has its own way of own way of own way of billing, billing, billing, right? Every doctor right? Every doctor right? Every doctor has their own way of has their own way of has their own way of writing notes writing notes writing notes about the progress of treatment. This is about the progress of treatment. This is about the progress of treatment. This is not consistent at all not consistent at all not consistent at all in in in any way shape or any way shape or any way shape or form. Therefore, the data is form. Therefore, the data is form. Therefore, the data is unstructured and unstructured and unstructured and often does not often does not often does not come in come in come in electronic form electronic form . It's a PDF, it's an . It's a PDF, it's an . It's a PDF, it's an email, it's a fax. So you email, it's a fax. So you email, it's a fax. So you have all these have all these have all these technological technological technological complexities complexities complexities superimposed on the superimposed on the underlying business complexities. And

underlying business complexities. And then you have something like then you have something like then you have something like this, right this, right this, right ? Someone does ? Someone does ? Someone does an assessment and there are three different an assessment and there are three different an assessment and there are three different results, right? results, right? results, right? For example, regarding For example, regarding the patient's range of motion. How are you the patient's range of motion. How are you going to resolve going to resolve going to resolve this to make a this to make a this to make a decision on the insurance decision on the insurance decision on the insurance claim? We have claim? We have claim? We have n't even put the other n't even put the other n't even put the other part on all of this yet. part on all of this yet. part on all of this yet. While these While these While these medical processes are ongoing, medical processes are ongoing, medical processes are ongoing, there is also the risk of there is also the risk of there is also the risk of lawsuits. So lawsuits. So , in insurance, , in insurance, , in insurance, usually the insurer usually the insurer usually the insurer is on the is on the is on the defendant's side, and there is also the defendant's side, and there is also the defendant's side, and there is also the plaintiff's side. The plaintiff's side. The plaintiff's side. The plaintiff's side plaintiff's side plaintiff's side is interested in is interested in is interested in processing the processing the processing the largest possible volume of cases and largest possible volume of cases and largest possible volume of cases and obtaining the obtaining the obtaining the largest possible amount largest possible amount largest possible amount of compensation. So you're of compensation. So you're of compensation. So you're effectively effectively effectively in a situation where there are two in a situation where there are two in a situation where there are two different different different incentive systems. So, on the incentive systems. So, on the incentive systems. So, on the plaintiff's side: act plaintiff's side: act plaintiff's side: act as quickly as possible, as quickly as possible, as quickly as possible, file as file as file as many claims as possible, many claims as possible, many claims as possible, put up put up put up huge billboards huge billboards huge billboards everywhere. We saw them everywhere. We saw them everywhere. We saw them when we flew here, when we flew here, when we flew here, I guess, right? They are I guess, right? They are I guess, right? They are everywhere. And the everywhere. And the everywhere. And the plaintiffs' side actually plaintiffs' side actually plaintiffs' side actually implemented AI implemented AI implemented AI much faster, much faster, much faster, compared to the compared to the compared to the insurance side insurance side insurance side or the defense side. or the defense side. or the defense side. Because you get Because you get Because you get a percentage based on a percentage based on a percentage based on how fast you how fast you how fast you work and how work and how work and how big your requirement is big your requirement is . The range of actual . The range of actual . The range of actual fee percentages fee percentages fee percentages for lawyers, depending for lawyers, depending for lawyers, depending on the jurisdiction, can on the jurisdiction, can on the jurisdiction, can range from 9% to 15% range from 9% to 15% in California for in California for workers' compensation. In other workers' compensation. In other jurisdictions, you jurisdictions, you jurisdictions, you get to the get to the get to the 1/3 or even 40 1/3 or even 40 % level. So, in addition to % level. So, in addition to % level. So, in addition to healthcare, insurance healthcare, insurance , technological , technological , technological challenges, and challenges, and challenges, and administration, administration, administration, you now have you now have you now have lawsuits, and insurance lawsuits, and insurance lawsuits, and insurance companies are putting companies are putting companies are putting a lot of money aside for

a lot of money aside for a lot of money aside for lawsuits. lawsuits. The problem here is The problem here is that if you haven't undergone that if you haven't undergone that if you haven't undergone proper digital proper digital proper digital transformation, you transformation, you transformation, you become especially become especially become especially vulnerable to vulnerable to vulnerable to attackers, attackers, attackers, right? So you may be right? So you may be right? So you may be sent repeated sent repeated sent repeated demands, you may be demands, you may be demands, you may be sent fraudulent sent fraudulent sent fraudulent demands, because it's demands, because it's demands, because it's even easier to do that now even easier to do that now even easier to do that now . So if . So if . So if you don't have the ability to you don't have the ability to you don't have the ability to handle handle handle these things effectively, there's these things effectively, there's more and more fraud coming into the system, more and more fraud coming into the system, right? This sounds like a right? This sounds like a right? This sounds like a situation where your situation where your situation where your website is easily website is easily website is easily vulnerable to a DDoS attack, vulnerable to a DDoS attack, vulnerable to a DDoS attack, right? People right? People right? People will continue to will continue to will continue to launch DDoS attacks launch DDoS attacks launch DDoS attacks on you. And we forget what's on you. And we forget what's most important when most important when we start solving we start solving we start solving all these sub-problems, do all these sub-problems, do all these sub-problems, do n't we? So, this is what n't we? So, this is what matters the most in terms of matters the most in terms of quality of quality of quality of patient care, reducing patient care, reducing patient care, reducing costs, ensuring a costs, ensuring a costs, ensuring a fast process and a fast process and a fast process and a good experience. So let's good experience. So let's good experience. So let's go through and go through and go through and solve some of these solve some of these solve some of these problems. I will talk problems. I will talk problems. I will talk about three different phases and about three different phases and about three different phases and how they can how they can how they can help. So, help. So, help. So, number one, number one, number one, let's talk about clean let's talk about clean let's talk about clean data, how can we data, how can we data, how can we achieve that. Number achieve that. Number achieve that. Number two, let's talk about two, let's talk about two, let's talk about document analysis and document analysis and document analysis and then discuss then discuss then discuss the factors between the lawsuits.

the factors between the lawsuits. Okay, clean data. Okay, clean data. So, has So, has So, has anyone here worked with anyone here worked with anyone here worked with claims data or claims data or claims data or any other any other any other insurance data before? insurance data before? insurance data before? Several people. Good. Several people. Good. Several people. Good. How many duplicate How many duplicate How many duplicate records have you records have you records have you processed or processed or processed or seen? Many? Yes, seen? Many? Yes, seen? Many? Yes, millions, right? millions, right? millions, right? Depending on the Depending on the Depending on the jurisdiction you jurisdiction you jurisdiction you work with, it could work with, it could work with, it could be up to 50% or even be up to 50% or even be up to 50% or even more, right? So there are more, right? So there are more, right? So there are so many duplicates coming in so many duplicates coming in so many duplicates coming in , and sometimes , and sometimes , and sometimes it's intentional, right it's intentional, right ? Sometimes, to process ? Sometimes, to process ? Sometimes, to process a request, you a request, you a request, you have to constantly have to constantly have to constantly “pull” people, right? “pull” people, right? “pull” people, right? Follow my Follow my Follow my request, follow request, follow request, follow my request, okay? my request, okay? my request, okay? Because people are Because people are Because people are slow, you slow, you slow, you get a kind of get a kind of get a kind of cascading cascading cascading failure effect that we failure effect that we failure effect that we see in technology. see in technology. see in technology. Mixed up records with Mixed up records with Mixed up records with other patients, other patients, other patients, right? You have Bob Joe and right? You have Bob Joe and right? You have Bob Joe and Bob Joe Jr., Bob Joe Jr., Bob Joe Jr., right? Such medical right? Such medical right? Such medical records are often records are often records are often mixed up with each other. mixed up with each other. mixed up with each other. This violates HIPAA. This This violates HIPAA. This This violates HIPAA. This entails a lot of entails a lot of entails a lot of regulatory risks regulatory risks , but the system and , but the system and , but the system and digital digital digital transformation are transformation are transformation are not not not adapted to this. And adapted to this. And adapted to this. And decisions decisions decisions can actually be made can actually be made can actually be made when when when other people's medical records other people's medical records other people's medical records are mixed together. Is are mixed together. Is are mixed together. Is n't that right? So this is an n't that right? So this is an n't that right? So this is an especially sensitive especially sensitive especially sensitive moment when it comes moment when it comes moment when it comes to a complex application to a complex application to a complex application where something could be where something could be where something could be missed. Someone's missed. Someone's missed. Someone's record is accidentally record is accidentally record is accidentally added by a doctor or added by a doctor or added by a doctor or administrator, and administrator, and administrator, and suddenly the entire application suddenly the entire application suddenly the entire application collapses. Then we collapses. Then we collapses. Then we have the more have the more have the more expected problems, expected problems, expected problems, like poorly like poorly like poorly scanned or scanned or scanned or undigitized data undigitized data . Isn't that right? You have . Isn't that right? You have . Isn't that right? You have handwriting that is handwriting that is handwriting that is impossible to decipher, impossible to decipher, impossible to decipher, especially a doctor's especially a doctor's especially a doctor's handwriting, right? Some handwriting, right? Some handwriting, right? Some documents come documents come documents come on paper. So we

on paper. So we on paper. So we keep getting keep getting keep getting examples of this kind of examples of this kind of examples of this kind of poor quality data, and it poor quality data, and it poor quality data, and it happens for many happens for many happens for many reasons, right? reasons, right? reasons, right? Some of this Some of this Some of this happens because there's happens because there's a lot of copy and a lot of copy and paste in the medical system, right? paste in the medical system, right? paste in the medical system, right? Doctors or nurses, Doctors or nurses, Doctors or nurses, when updating medical when updating medical when updating medical records, do this records, do this records, do this all the time, right? all the time, right? all the time, right? They copy and They copy and They copy and paste data between paste data between paste data between patients. They patients. They patients. They do this by copying do this by copying do this by copying records between records between records between visits. visits. visits. You may You may You may have seen it yourself, but have seen it yourself, but have seen it yourself, but it happens it happens it happens often in this field. Electronic often in this field. Electronic often in this field. Electronic health records are health records are health records are not really not really not really set up set up set up consistently. So has consistently. So has consistently. So has anyone here ever seen an anyone here ever seen an electronic electronic electronic health record? Your health record? Your health record? Your own medical own medical own medical card, right? Has anyone here card, right? Has anyone here card, right? Has anyone here switched to a different switched to a different switched to a different doctor or network? Yes. doctor or network? Yes. doctor or network? Yes. Do they look the Do they look the Do they look the same? No, they don't same? No, they don't same? No, they don't look the same. look the same. look the same. I'm from Toronto, Canada, and I'm from Toronto, Canada, and I'm from Toronto, Canada, and many of our many of our many of our hospitals hospitals hospitals do a new do a new do a new implementation every time they implementation every time they implementation every time they set up an EMR. And set up an EMR. And set up an EMR. And this often happens both this often happens both this often happens both through initial through initial through initial processes within processes within processes within the hospital and within the the hospital and within the the hospital and within the network. There are very network. There are very network. There are very interesting dynamics as to interesting dynamics as to interesting dynamics as to why they are why they are why they are set up the way they are. set up the way they are. set up the way they are. And if you want And if you want And if you want to talk about to talk about to talk about data purity, people data purity, people data purity, people do some really weird do some really weird do some really weird things in ECM with things in ECM with things in ECM with nested fields.

nested fields. So even though we have So even though we have electronic medical electronic medical electronic medical records and systems, records and systems, records and systems, they're all different, they're they're all different, they're they're all different, they're structured strangely, structured strangely, structured strangely, so you have to so you have to so you have to convert them. And convert them. And convert them. And often the data is difficult often the data is difficult often the data is difficult to extract, so in the end to extract, so in the end to extract, so in the end people just choose to people just choose to people just choose to send faxes or send faxes or send faxes or PDFs. So PDFs. So PDFs. So ensuring ensuring ensuring data purity is not an easy data purity is not an easy data purity is not an easy task. AI can task. AI can task. AI can help us, digital help us, digital help us, digital transformation can transformation can transformation can help us, but there are help us, but there are help us, but there are many factors that many factors that many factors that affect this at the affect this at the affect this at the initial stage. This is an initial stage. This is an initial stage. This is an example of example of example of illustrative data. illustrative data. illustrative data. Duplicate entries. Duplicate entries. Duplicate entries. The real data The real data The real data looks quite looks quite looks quite similar, with a lot of similar, with a lot of similar, with a lot of duplicates duplicates duplicates in certain networks. And it's in certain networks. And it's in certain networks. And it's really annoying if you really annoying if you really annoying if you get these get these get these duplicate entries, duplicate entries, duplicate entries, right? So we have right? So we have right? So we have a number of clients a number of clients a number of clients asking, "Can asking, "Can asking, "Can this organization this organization this organization stop sending stop sending stop sending us so many us so many us so many duplicates?" And if duplicates?" And if duplicates?" And if you don't have robust you don't have robust you don't have robust systems in place to detect systems in place to detect systems in place to detect this, you keep this, you keep this, you keep getting the same getting the same getting the same low-quality data. low-quality data. Therefore, it is very important Therefore, it is very important to have technical to have technical to have technical solutions and solutions and solutions and detection systems to detection systems to detection systems to guarantee the purity of guarantee the purity of guarantee the purity of your own data.

your own data. Now that you have this Now that you have this data, there are data, there are data, there are mechanisms that mechanisms that mechanisms that allow you to actually allow you to actually allow you to actually fix the situation, fix the situation, fix the situation, right? For right? For right? For regulatory or regulatory or regulatory or other other other interpersonal interpersonal interpersonal reasons, right? But reasons, right? But reasons, right? But you still have you still have you still have this this this cluttered data as a cluttered data as a cluttered data as a starting point. And when starting point. And when starting point. And when you can detect you can detect you can detect things like things like things like duplicates, which again duplicates, which again duplicates, which again seems like a seems like a seems like a solved problem, solved problem, solved problem, you no longer incentivize you no longer incentivize you no longer incentivize submitting the same submitting the same submitting the same claims multiple times with claims multiple times with claims multiple times with different dates or different dates or different dates or other people other people other people having to having to having to correct their data. correct their data. The other part, besides The other part, besides the purity of the data, the purity of the data, the purity of the data, is that these is that these is that these claims have a very claims have a very claims have a very large volume. When I first large volume. When I first large volume. When I first started started started working and saw an 8GB working and saw an 8GB working and saw an 8GB PDF file in our database PDF file in our database , it , it , it was very strange. So, was very strange. So, was very strange. So, these claims contain these claims contain these claims contain everything that is everything that is everything that is possible. They contain possible. They contain possible. They contain images, right? images, right? images, right? If you think about If you think about If you think about car accidents, car accidents, car accidents, for example? There are for example? There are for example? There are images of the site of the images of the site of the images of the site of the injury, collisions, etc. injury, collisions, etc. There are X-ray reports, There are X-ray reports, ultrasounds, and everything ultrasounds, and everything else. And sometimes it else. And sometimes it else. And sometimes it just causes just causes just causes crashes in many crashes in many PDF viewing programs. If PDF viewing programs. If you think about some of the you think about some of the you think about some of the older computers older computers older computers you might have, you might have, you might have, trying to open an trying to open an trying to open an 8GB PDF would be quite a 8GB PDF would be quite a 8GB PDF would be quite a challenge. And challenge. And challenge. And even for even for even for technological systems technological systems , right? We , right? We , right? We talk talk talk about context windows all the time.

about context windows all the time. about context windows all the time. How are you going to How are you going to How are you going to fit the context of an 8- fit the context of an 8- gigabyte PDF file gigabyte PDF file gigabyte PDF file into a context window? into a context window? into a context window? It's not as simple as It's not as simple as It's not as simple as saying, "Hey, Claude, saying, "Hey, Claude, saying, "Hey, Claude, take a look at my 10- take a look at my 10- page PDF and give page PDF and give page PDF and give me an answer," or " me an answer," or " Hey, ChatGPT or Gemini," Hey, ChatGPT or Gemini," Hey, ChatGPT or Gemini," right? Not so simple. right? Not so simple. right? Not so simple. In addition, the actual In addition, the actual In addition, the actual processing of the case processing of the case processing of the case requires a significant requires a significant requires a significant amount of time. So, amount of time. So, amount of time. So, these are either technical these are either technical these are either technical challenges or business challenges or business challenges that challenges that challenges that lead to very lead to very lead to very complex technical complex technical complex technical tasks. Now, if tasks. Now, if tasks. Now, if we work further and we work further and we work further and add more and add more and add more and more complexity here, more complexity here, more complexity here, many of these many of these many of these claims also claims also claims also have have have inconsistencies. inconsistencies. inconsistencies. For example, you For example, you For example, you might have notes might have notes might have notes about the course of the disease from an about the course of the disease from an about the course of the disease from an early early early patient report that say, patient report that say, patient report that say, "Here is the X-ray that was "Here is the X-ray that was "Here is the X-ray that was scheduled for such and scheduled for such and such a date." You receive a such a date." You receive a such a date." You receive a package of documents from package of documents from package of documents from the applicant, but this the applicant, but this the applicant, but this X-ray is not there. X-ray is not there. X-ray is not there. Truth? And it's very Truth? And it's very Truth? And it's very easy to miss easy to miss easy to miss unless you have some unless you have some unless you have some way of highlighting it. way of highlighting it. way of highlighting it. Um, and without that, Um, and without that, Um, and without that, again, the whole again, the whole again, the whole trajectory of the case trajectory of the case trajectory of the case could change: whether could change: whether could change: whether proper treatment was provided proper treatment was provided proper treatment was provided , whether , whether , whether the person actually sought the person actually sought the person actually sought help, and how that help, and how that help, and how that can be reconciled with the can be reconciled with the can be reconciled with the bills. So it's a bills. So it's a bills. So it's a very common very common very common situation where, due to situation where, due to situation where, due to data clutter, you data clutter, you data clutter, you have to keep have to keep have to keep asking, "Hey, asking, "Hey, asking, "Hey, can you can you can you send me this send me this send me this report?" "Oh, we forgot to report?" "Oh, we forgot to report?" "Oh, we forgot to send the fax." "Oops, send the fax." "Oops, send the fax." "Oops, the fax didn't work." " the fax didn't work." " Oh, we changed the Oh, we changed the Oh, we changed the internal system." Is that internal system." Is that internal system." Is that true? Therefore, true? Therefore, true? Therefore, the organization must the organization must the organization must have appropriate have appropriate have appropriate lines of defense in place to lines of defense in place to lines of defense in place to detect such cases in a timely manner detect such cases in a timely manner detect such cases in a timely manner . And an insurance . And an insurance . And an insurance expert can expert can expert can simultaneously simultaneously simultaneously process 200 process 200 process 200 insurance cases.

insurance cases. insurance cases. So he won't be able to So he won't be able to So he won't be able to keep everything in his head, keep everything in his head, keep everything in his head, remembering, "Oh, yeah, remembering, "Oh, yeah, remembering, "Oh, yeah, I need to check if I need to check if I need to check if this X-ray is there or not." Is that this X-ray is there or not." Is that this X-ray is there or not." Is that true? You can true? You can true? You can take notes and take notes and take notes and continue working, but continue working, but continue working, but the information is still the information is still the information is still lost. This is where lost. This is where lost. This is where technology can technology can technology can greatly help in greatly help in greatly help in solving such solving such solving such problems. Then problems. Then conflicting diagnoses arise . This also . This also . This also happens quite happens quite happens quite often. So who here often. So who here often. So who here sought a second sought a second sought a second opinion on treatment? opinion on treatment? A few people, right? Um, A few people, right? Um, sometimes we look to her sometimes we look to her sometimes we look to her to get a different to get a different to get a different perspective on a situation. perspective on a situation. perspective on a situation. Sometimes you get Sometimes you get Sometimes you get completely different completely different completely different answers, right? And the more answers, right? And the more answers, right? And the more complicated the matter, the more complicated the matter, the more complicated the matter, the more likely it is that likely it is that likely it is that some some some disagreements will arise. So in disagreements will arise. So in disagreements will arise. So in this case, when you this case, when you this case, when you are reviewing a medical are reviewing a medical are reviewing a medical claim and see a claim and see a claim and see a conflicting diagnosis conflicting diagnosis , or want to clarify , or want to clarify , or want to clarify the amount of benefits, you the amount of benefits, you the amount of benefits, you refer the case refer the case refer the case to an Independent Medical to an Independent Medical to an Independent Medical Evaluation (IME). Then Evaluation (IME). Then Evaluation (IME). Then another doctor another doctor another doctor reviews your reviews your reviews your case. And then you case. And then you case. And then you have to manage have to manage have to manage these doctors, these doctors, these doctors, provider networks, provider networks, provider networks, and the experts themselves.

and the experts themselves. and the experts themselves. So, we are So, we are So, we are piling up piling up piling up the complexity again. More the complexity again. More the complexity again. More people in the ecosystem, people in the ecosystem, people in the ecosystem, more problems. more problems. more problems. Artificial intelligence Artificial intelligence Artificial intelligence can be very can be very can be very useful now, and I think it would be useful now, and I think it would be useful now, and I think it would be strange to strange to strange to speak in 2026 speak in 2026 speak in 2026 without mentioning the without mentioning the without mentioning the evaluation of results. evaluation of results. evaluation of results. But we really But we really But we really used AI used AI used AI to try to to try to to try to solve these issues: solve these issues: solve these issues: conflicting diagnoses conflicting diagnoses , missing records, and , missing records, and , missing records, and more. So more. So more. So we created a benchmark we created a benchmark we created a benchmark called MLCR called MLCR called MLCR using using using artificial intelligence to artificial intelligence to artificial intelligence to measure measure measure the effectiveness of AI in the effectiveness of AI in the effectiveness of AI in answering these answering these answering these questions. So, questions. So, questions. So, let's go to our let's go to our let's go to our blog that we were blog that we were blog that we were talking about, and I'll talking about, and I'll talking about, and I'll scroll down to scroll down to scroll down to show you what show you what show you what this benchmark looks like. We have this benchmark looks like. We have this benchmark looks like. We have released six released six released six different levels of different levels of different levels of questions related questions related questions related to medical to medical to medical requirements. Some of them are requirements. Some of them are requirements. Some of them are very simple, very simple, very simple, like when did like when did like when did this injury happen and this injury happen and this injury happen and what was it like, right? what was it like, right? what was it like, right? Then they get Then they get Then they get a little more complicated. a little more complicated. a little more complicated. How many operations How many operations How many operations were performed? What is were performed? What is were performed? What is this this this patient's capacity? And eventually patient's capacity? And eventually patient's capacity? And eventually we move on to we move on to we move on to increasingly complex increasingly complex increasingly complex issues that are issues that are issues that are even harder to solve. And often even harder to solve. And often even harder to solve. And often when you ask when you ask when you ask a question as an a question as an a question as an insurance adjuster, insurance adjuster, insurance adjuster, doctor, or doctor, or doctor, or administrator, you don't administrator, you don't administrator, you don't have the full have the full have the full context of what's context of what's context of what's going on. This is where going on. This is where going on. This is where we see we see we see hallucinations, where hallucinations, where hallucinations, where people assume that people assume that people assume that something is in the medical something is in the medical something is in the medical record, when record, when record, when in fact it is in fact it is in fact it is not there. So, let's not there. So, let's dig a little deeper into the dig a little deeper into the results. This is results. This is results. This is available on the available on the available on the Artificial Analysis website. If you Artificial Analysis website. If you Artificial Analysis website. If you want to take a look, we've want to take a look, we've want to take a look, we've released an released an released an open- open- open- source dataset for the first source dataset for the first source dataset for the first three levels of three levels of three levels of questions that we

questions that we questions that we have, the have, the have, the simpler ones. And we have a simpler ones. And we have a simpler ones. And we have a private set that private set that private set that we host with we host with we host with Artificial Analysis to Artificial Analysis to limit benchmarking to some extent . So it's very interesting . So it's very interesting . So it's very interesting to watch to watch to watch how the models work. how the models work. how the models work. We actually We actually We actually measure this along measure this along measure this along three different axes. three different axes. three different axes. Let me Let me Let me go back to that go back to that go back to that slide. We measure slide. We measure slide. We measure it by brevity. it by brevity. it by brevity. How concise How concise How concise is the answer? We is the answer? We is the answer? We compare this with compare this with compare this with the answer the answer the answer confirmed by a confirmed by a confirmed by a human expert. human expert. human expert. Then we move on Then we move on Then we move on to accuracy. Is to accuracy. Is to accuracy. Is the answer correct the answer correct ? And the last thing is ? And the last thing is ? And the last thing is completeness. Has completeness. Has completeness. Has all all all relevant relevant relevant information in information in information in the question been answered? So, the question been answered? So, the question been answered? So, we come back here. The we come back here. The we come back here. The frontier of technology frontier of technology frontier of technology continues to continues to continues to advance very advance very advance very rapidly, which is why rapidly, which is why rapidly, which is why evaluation is so evaluation is so evaluation is so important. When we important. When we important. When we released the released the released the scoring system, the highest scoring system, the highest scoring system, the highest accuracy we accuracy we accuracy we saw on our saw on our saw on our reasoning levels reasoning levels reasoning levels was 20%. After was 20%. After was 20%. After a few months, we already a few months, we already a few months, we already have 75% at very have 75% at very have 75% at very high levels of high levels of high levels of reasoning. Yes? reasoning. Yes? reasoning. Yes? Models are evolving Models are evolving Models are evolving very quickly, but that's very quickly, but that's very quickly, but that's great because it great because it great because it means we can means we can means we can start to implement start to implement start to implement a range of more complex a range of more complex a range of more complex issues. And we have issues. And we have issues. And we have more confidence that more confidence that more confidence that someone has to someone has to someone has to review it in the review it in the review it in the system. So, in our system. So, in our system. So, in our system, there is always a system, there is always a system, there is always a person person person involved in the process. This is involved in the process. This is involved in the process. This is really important. You really important. You really important. You also start also start also start to get these to get these to get these Pareto frontier cost curves Pareto frontier cost curves Pareto frontier cost curves regarding the cost of regarding the cost of regarding the cost of actually asking and actually asking and actually asking and getting the proper getting the proper getting the proper answer. And that has also answer. And that has also answer. And that has also been improving been improving been improving over the past over the past over the past few months. And you few months. And you few months. And you also get also get also get some trade-offs.

some trade-offs. some trade-offs. So, one of the interesting So, one of the interesting So, one of the interesting things when we things when we things when we ran these tests on ran these tests on ran these tests on document analysis document analysis document analysis was that the was that the was that the GPT models were GPT models were GPT models were great in terms of great in terms of great in terms of accuracy, but they were accuracy, but they were accuracy, but they were missing some missing some missing some key pieces of key pieces of key pieces of information. The information. The information. The Anthropic models were quite Anthropic models were quite Anthropic models were quite good, and they are good, and they are good, and they are much more complete in terms of much more complete in terms of much more complete in terms of the amount of information. the amount of information. the amount of information. But they also have But they also have But they also have slightly lower accuracy slightly lower accuracy slightly lower accuracy here. here. here. Can I ask Can I ask Can I ask a question? a question? a question? Of course. Of course. Of course. Accuracy, how do you Accuracy, how do you Accuracy, how do you measure it measure it measure it compared to the human compared to the human compared to the human factor? factor? factor? Yes, compared to the Yes, compared to the Yes, compared to the human response, and human response, and human response, and then we do a then we do a then we do a review by three review by three review by three judges. So yes. So judges. So yes. So , we have accuracy here. , we have accuracy here. , we have accuracy here. You see what the You see what the You see what the curve looks like and curve looks like and curve looks like and we have a completeness that is we have a completeness that is we have a completeness that is still not resolved in still not resolved in still not resolved in most most model providers. And yes, regarding the model providers. And yes, regarding the previous previous previous question, this is question, this is question, this is based on a based on a based on a human-verified human-verified human-verified answer and a answer and a answer and a calibrated calibrated calibrated set of judges. In set of judges. In set of judges. In our blog we our blog we our blog we describe the methods describe the methods describe the methods we we we use here, and we have use here, and we have use here, and we have also released a also released a also released a follow-up on follow-up on follow-up on how we conducted this how we conducted this how we conducted this calibration. So calibration. So calibration. So while we while we while we use AI use AI use AI to structure to structure to structure our data and our data and our data and extract important extract important extract important insights, we want to insights, we want to insights, we want to make sure that make sure that make sure that every step of that every step of that every step of that process is well process is well process is well calibrated, calibrated, calibrated, because what if your because what if your because what if your data is wrong at data is wrong at data is wrong at the beginning and you the beginning and you the beginning and you start start start using it using it using it further? You didn't further? You didn't further? You didn't get anything. So again get anything. So again get anything. So again , it's very important , it's very important , it's very important as you go through the application as you go through the application as you go through the application process to process to process to calibrate your calibrate your calibrate your judges to create judges to create judges to create good scores. I good scores. I good scores. I think this is a think this is a think this is a recurring theme in recurring theme in recurring theme in Ed Most's speeches. And you Ed Most's speeches. And you Ed Most's speeches. And you want to make sure want to make sure want to make sure you have experts you have experts you have experts who can actually who can actually who can actually verify that verify that verify that information, so you information, so you

information, so you work across work across work across teams to teams to teams to achieve that. Okay achieve that. Okay . So, that's within the scope of . So, that's within the scope of . So, that's within the scope of one statement. We one statement. We one statement. We know that with a certain know that with a certain know that with a certain degree of confidence degree of confidence degree of confidence we can extract we can extract we can extract certain types of insights. certain types of insights. certain types of insights. We can flag We can flag We can flag conflicting diagnoses conflicting diagnoses , we can remove , we can remove , we can remove duplicates, we can duplicates, we can duplicates, we can search for missing search for missing search for missing records. Let's move on to records. Let's move on to records. Let's move on to something more interesting, something more interesting, something more interesting, namely comparing the namely comparing the namely comparing the statements with each other. Isn't that right statements with each other. Isn't that right statements with each other. Isn't that right ? So, if you are an ? So, if you are an ? So, if you are an average insurance average insurance average insurance company, you might company, you might company, you might collect $1 billion in collect $1 billion in collect $1 billion in insurance insurance insurance premiums across a portfolio of premiums across a portfolio of premiums across a portfolio of different lines of different lines of different lines of business. And you can business. And you can business. And you can pay out, say, pay out, say, pay out, say, 20,000 claims a year or 20,000 claims a year or 20,000 claims a year or even more, even more, even more, depending on depending on depending on what line of what line of what line of business you're working with. business you're working with. business you're working with. Now you want to better Now you want to better Now you want to better understand what exactly understand what exactly understand what exactly affects your affects your loss ratio. What exactly are loss ratio. What exactly are the demands that the demands that the demands that people are making? In which people are making? In which people are making? In which geographical areas? geographical areas? geographical areas? What are the types of injuries? and which What are the types of injuries? and which What are the types of injuries? and which doctors are involved doctors are involved doctors are involved in the process. This is where I in the process. This is where I in the process. This is where I think it gets think it gets think it gets really interesting: really interesting: really interesting: once you've extracted this once you've extracted this once you've extracted this data from the case, you can data from the case, you can data from the case, you can start building start building start building interesting visualizations, interesting visualizations, interesting visualizations, regressions, and regressions, and regressions, and analytics. Yes? So the analytics. Yes? So the analytics. Yes? So the two questions two questions two questions we were asked to we were asked to we were asked to answer answer answer are about how are about how are about how much people much people much people travel within the travel within the travel within the network, right?

network, right? network, right? Let's say the patient is Let's say the patient is Let's say the patient is in San Francisco or in San Francisco or in San Francisco or New York, right? Are New York, right? Are New York, right? Are they receiving they receiving they receiving treatment in their treatment in their treatment in their area? And is this the area? And is this the area? And is this the person who should person who should person who should provide treatment? provide treatment? provide treatment? We also have networks of We also have networks of We also have networks of who actually who actually who actually collaborates with whom. This is a collaborates with whom. This is a collaborates with whom. This is a slightly edited slightly edited slightly edited data, a subset from data, a subset from data, a subset from California, where you can California, where you can California, where you can see which see which see which lawyers work lawyers work lawyers work together, which doctors, together, which doctors, together, which doctors, chiropractors, chiropractors, chiropractors, physical therapists, and physical therapists, and physical therapists, and what the what the what the actual working actual working actual working relationships and history look like within the relationships and history look like within the relationships and history look like within the case itself. case itself. This is the level of advanced This is the level of advanced analytics that can be analytics that can be analytics that can be obtained by obtained by obtained by extracting data from extracting data from extracting data from your cases. You also your cases. You also your cases. You also move on to move on to move on to interesting medical interesting medical interesting medical findings, findings, findings, right? So, what conditions right? So, what conditions right? So, what conditions occur simultaneously? occur simultaneously? occur simultaneously? For example, if For example, if For example, if someone has a back injury, someone has a back injury, someone has a back injury, right? You may have a right? You may have a right? You may have a number of different number of different number of different diagnoses that can be diagnoses that can be diagnoses that can be represented by represented by represented by ICD codes that coexist ICD codes that coexist ICD codes that coexist and see why it is and see why it is and see why it is present in present in present in one doctor's report but not one doctor's report but not one doctor's report but not another. And in this another. And in this another. And in this case, again, case, again, case, again, you only get these you only get these you only get these capabilities capabilities capabilities when you can when you can when you can solve the solve the solve the data mining problem in data mining problem in data mining problem in one case and then one case and then one case and then generalize that to generalize that to generalize that to your entire your entire your entire insurance portfolio.

insurance portfolio. Now let's move on to the Now let's move on to the financial part, shall we financial part, shall we financial part, shall we ? So, let's take a ? So, let's take a ? So, let's take a small subsegment small subsegment small subsegment of two of two of two physical therapy sessions: different physical therapy sessions: different physical therapy sessions: different doctors charge different doctors charge different doctors charge different amounts, right? We've amounts, right? We've amounts, right? We've probably probably probably seen this before. And sometimes that's seen this before. And sometimes that's seen this before. And sometimes that's okay. If you are a okay. If you are a okay. If you are a better doctor, you better doctor, you better doctor, you should be entitled to should be entitled to should be entitled to charge more, right? charge more, right? charge more, right? If you can If you can If you can help someone help someone help someone recover as recover as recover as quickly as possible in five quickly as possible in five quickly as possible in five sessions instead of sessions instead of sessions instead of ten, you should be ten, you should be ten, you should be entitled to charge entitled to charge entitled to charge more, right? So, uh, more, right? So, uh, more, right? So, uh, but it's really important to but it's really important to but it's really important to understand costs and understand costs and understand costs and focus on focus on focus on minimizing them, because minimizing them, because minimizing them, because once you have once you have once you have cost inflation in cost inflation in cost inflation in different parts of the different parts of the different parts of the healthcare system, healthcare system, healthcare system, it keeps it keeps it keeps going up, right? going up, right? going up, right? Insurance premiums are Insurance premiums are Insurance premiums are going up, uh, the fees they going up, uh, the fees they going up, uh, the fees they charge us charge us charge us are going up, uh, and are going up, uh, and are going up, uh, and we're getting less for we're getting less for we're getting less for our money. So, our money. So, our money. So, within this analysis, within this analysis, within this analysis, can you see can you see can you see if there is a common factor if there is a common factor if there is a common factor in the different insurance in the different insurance in the different insurance claims? What exactly is claims? What exactly is claims? What exactly is the amount of insurance the amount of insurance the amount of insurance coverage you should coverage you should coverage you should provide? How exactly provide? How exactly provide? How exactly should you assess these should you assess these should you assess these costs for costs for costs for insurance claims. insurance claims. insurance claims. And you can And you can And you can predict how things predict how things predict how things will change over will change over will change over time, right? time, right? time, right? For example, For example, For example, there was a there was a there was a 10% year-over-year increase for 10% year-over-year increase for 10% year-over-year increase for this type of treatment this type of treatment this type of treatment at a certain doctor, at a certain doctor, at a certain doctor, right? So, will right? So, will right? So, will this happen this happen this happen across the entire across the entire across the entire jurisdiction? Will jurisdiction? Will jurisdiction? Will this happen across the this happen across the this happen across the industry as a whole? Then industry as a whole? Then industry as a whole? Then you can start you can start you can start doing doing doing counterfactual counterfactual counterfactual analysis or changes in analysis or changes in analysis or changes in analysis to say: analysis to say: analysis to say: if we change our if we change our if we change our network, if we network, if we network, if we point people in a point people in a point people in a certain direction, certain direction, certain direction, if we change the if we change the if we change the way we way we way we bill, right? What bill, right? What bill, right? What potential savings potential savings potential savings can be achieved?

can be achieved? can be achieved? And what procedures should And what procedures should And what procedures should people undergo people undergo people undergo according to according to according to existing standards of existing standards of existing standards of medical medical medical care? care? care? There was an article recently about whether There was an article recently about whether There was an article recently about whether anyone had seen the article anyone had seen the article anyone had seen the article about overbilling about overbilling about overbilling by $1 billion in by $1 billion in by $1 billion in additional additional additional charges? Did you see charges? Did you see charges? Did you see this? So these are examples this? So these are examples this? So these are examples of how if you can't of how if you can't handle data effectively, people handle data effectively, people will find ways to will find ways to will find ways to submit additional submit additional submit additional records or engage in records or engage in records or engage in less transparent less transparent less transparent activities. So, all of activities. So, all of activities. So, all of this means that when this means that when this means that when you have good data, you you have good data, you you have good data, you can start can start can start making those kinds of making those kinds of making those kinds of decisions. And once you decisions. And once you decisions. And once you have this solution, this have this solution, this have this solution, this information about information about information about the requirements, you can the requirements, you can the requirements, you can start integrating start integrating start integrating it into the broader system, it into the broader system, it into the broader system, right? You can start right? You can start right? You can start using this in using this in using this in other areas of your other areas of your other areas of your business: if you business: if you business: if you know the trends in know the trends in know the trends in auto liability or auto liability or workers' compensation, are workers' compensation, are we seeing the we seeing the we seeing the same in general same in general liability? liability? If you know that If you know that If you know that your underwriting model your underwriting model is missing certain is missing certain trends or lacks trends or lacks trends or lacks sufficient sufficient sufficient explainability, explainability, explainability, can we make can we make can we make the model more understandable the model more understandable the model more understandable based on what we based on what we based on what we see? How do we see? How do we see? How do we create the same create the same create the same magical magical magical data table that we would like to data table that we would like to data table that we would like to see in our see in our see in our data warehouse? So, data warehouse? So, data warehouse? So, the more data you have the more data you have the more data you have , the more complete the , the more complete the , the more complete the picture you can picture you can picture you can create and the better create and the better create and the better you can evaluate and you can evaluate and you can evaluate and price the price the price the product for the product for the product for the end consumer.

end consumer. And this is the information that is And this is the information that is needed for needed for needed for AI applications to work effectively AI applications to work effectively , whether in , whether in , whether in underwriting, underwriting, fraud prevention, or fraud prevention, or claims processing. claims processing. So, if you imagine the So, if you imagine the ideal ideal ideal scenario, right? We scenario, right? We scenario, right? We want to achieve want to achieve want to achieve broader broader broader cost reductions. So, if we cost reductions. So, if we cost reductions. So, if we want to reduce want to reduce want to reduce costs for everyone, costs for everyone, costs for everyone, right? Taking into right? Taking into right? Taking into account the experience of account the experience of account the experience of patients. If we patients. If we patients. If we want to reduce the time it takes to want to reduce the time it takes to want to reduce the time it takes to process an insurance process an insurance process an insurance claim. If we want to claim. If we want to claim. If we want to improve the speed of improve the speed of improve the speed of people's return to people's return to people's return to work and the quality of work and the quality of work and the quality of their treatment, their treatment, their treatment, we need to have we need to have we need to have the right data and the the right data and the the right data and the right digital right digital right digital transformation. So this is a transformation. So this is a transformation. So this is a clear example of the clear example of the clear example of the work we have also work we have also work we have also done at WiseDocs. We done at WiseDocs. We done at WiseDocs. We also saw also saw also saw significant significant significant cost savings. Therefore, cost savings. Therefore, cost savings. Therefore, going through this going through this going through this digital digital digital transformation process and transformation process and transformation process and having the having the having the right data is right data is right data is very important. This very important. This very important. This concludes our concludes our concludes our seminar. I wanted to seminar. I wanted to seminar. I wanted to leave enough leave enough leave enough time for questions, time for questions, time for questions, as this is a fairly as this is a fairly as this is a fairly large topic. Thank you for your large topic. Thank you for your large topic. Thank you for your attention, I will be attention, I will be attention, I will be happy happy happy to answer your to answer your to answer your questions. Yes, I questions. Yes, I questions. Yes, I will share the will share the will share the blog post. Yes, the question is blog post. Yes, the question is blog post. Yes, the question is here. Yes, the question here. Yes, the question here. Yes, the question was about how we was about how we was about how we structure the data structure the data structure the data to generate all to generate all to generate all this information. This this information. This this information. This happens through happens through happens through several different levels.

several different levels. several different levels. I will also share I will also share I will also share some of my some of my some of my speeches on this topic. speeches on this topic. speeches on this topic. The basic level is the The basic level is the The basic level is the use of OCR and use of OCR and use of OCR and images to images to images to extract information extract information extract information from basic from basic from basic documents. documents. documents. A kind of standard A kind of standard A kind of standard pre-processing. pre-processing. pre-processing. Preprocessing PDF Preprocessing PDF files is also files is also files is also not an easy task, not an easy task, not an easy task, if anyone has worked with it if anyone has worked with it if anyone has worked with it . Those . Those . Those 8 gigabyte files contain 8 gigabyte files contain 8 gigabyte files contain some strange examples, some strange examples, some strange examples, one in a million. So, one in a million. So, one in a million. So, when you have data, when you have data, when you have data, text, and images, text, and images, text, and images, we apply a number of we apply a number of we apply a number of models, like NLP models, like NLP models and models and models and data mining models. We data mining models. We data mining models. We sometimes use sometimes use sometimes use VLM for structuring VLM for structuring , although the standard , although the standard entity extraction model works entity extraction model works fine. We develop fine. We develop fine. We develop various classifiers various classifiers various classifiers on top of them. Then on top of them. Then on top of them. Then we have a we have a we have a summarization layer that summarization layer that summarization layer that summarizes the content of the summarizes the content of the summarizes the content of the documents. documents. documents. The person then The person then The person then checks the information checks the information to make sure to make sure to make sure it is correct. And it is correct. And it is correct. And finally, we finally, we finally, we run additional run additional run additional levels of analysis. levels of analysis. levels of analysis. So we have a model So we have a model So we have a model of abstraction that of abstraction that of abstraction that concentrates the density of concentrates the density of concentrates the density of information as we information as we information as we move up, but move up, but move up, but it is tested it is tested it is tested at different stages by at different stages by at different stages by different models and different models and different models and with the participation of people. This with the participation of people. This with the participation of people. This means that when you means that when you means that when you move up to the move up to the move up to the most valuable most valuable most valuable elements, that's when elements, that's when elements, that's when AI becomes effective. And these are the AI becomes effective. And these are the AI becomes effective. And these are the MLCR type questions MLCR type questions MLCR type questions that may have a that may have a that may have a correct answer. The correct answer. The correct answer. The MLCR MLCR MLCR uses uses uses clean data to clean data to clean data to ask these questions.

ask these questions. ask these questions. Real data is not Real data is not Real data is not clean, so clean, so clean, so you need to go through you need to go through you need to go through several stages to several stages to several stages to clean it and achieve a clean it and achieve a clean it and achieve a result. Yes, result. Yes, result. Yes, usually our usually our usually our clients, larger clients, larger clients, larger clients, have several clients, have several clients, have several business lines business lines business lines within them, and they have within them, and they have within them, and they have different rules for different rules for different rules for data exchange between different data exchange between different data exchange between different teams. Other teams. Other teams. Other clients work clients work clients work in only one in only one in only one line of business. So line of business. So line of business. So they can they can they can only work in only work in only work in auto insurance, or auto insurance, or auto insurance, or only in only in only in disability insurance, disability insurance, disability insurance, or only in workers' or only in workers' compensation insurance compensation insurance . But we . But we . But we take general take general take general trends to trends to trends to try to identify try to identify try to identify and label them. So, and label them. So, and label them. So, the question arises the question arises the question arises within the same office, within the same office, within the same office, right? How do you right? How do you right? How do you discover some of these discover some of these discover some of these insights or, say insights or, say , geographical , geographical , geographical features? Then features? Then features? Then you have a level above, which you have a level above, which you have a level above, which may be a broader may be a broader may be a broader company, but company, but company, but insurance companies insurance companies insurance companies also sometimes have a also sometimes have a also sometimes have a set of set of set of individual businesses within their structure. And individual businesses within their structure. And individual businesses within their structure. And how do you actually how do you actually how do you actually transfer this to a transfer this to a transfer this to a centralized centralized centralized data store? data store? data store? Again, every Again, every Again, every company is a little company is a little company is a little different, but we different, but we different, but we move at the move at the move at the speed speed speed our our our partners want. We live by partners want. We live by partners want. We live by requirements. So we requirements. So we requirements. So we focus on focus on focus on insurance claims.

insurance claims. insurance claims. Claims are our Claims are our Claims are our main bread, main bread, main bread, especially complex especially complex especially complex insurance claims. This is what insurance claims. This is what our entire artificial our entire artificial intelligence pipeline is focused on. But as intelligence pipeline is focused on. But as intelligence pipeline is focused on. But as we we we evolve and evolve and evolve and move beyond simpler move beyond simpler move beyond simpler tasks, like tasks, like tasks, like taking a requirement, taking a requirement, taking a requirement, processing it, and processing it, and processing it, and extracting extracting extracting information, you information, you information, you start to see start to see start to see deeper deeper deeper aspects that aspects that aspects that help help help organizations organizations organizations make decisions at the make decisions at the make decisions at the portfolio level. Yes. portfolio level. Yes. portfolio level. Yes. Yes, that's a good Yes, that's a good Yes, that's a good question. So question. So question. So the question is, how do we the question is, how do we the question is, how do we engage someone engage someone engage someone when there are so many of when there are so many of when there are so many of these problems? these problems? these problems? We usually work We usually work We usually work with those for whom the with those for whom the with those for whom the most important thing is most important thing is readiness for change, readiness for change, readiness for change, right? We've all right? We've all right? We've all probably worked with probably worked with probably worked with teams or teams or teams or organizations that organizations that organizations that want change and those that don't want change and those that don't want change and those that don't . . . It's a very, very different It's a very, very different It's a very, very different experience. So, experience. So, experience. So, assuming this is an assuming this is an assuming this is an organization that is organization that is organization that is looking for change, we looking for change, we looking for change, we usually usually usually work through a few work through a few work through a few sample files in sample files in sample files in different formats. different formats. different formats. Usually people Usually people Usually people send us something send us something send us something they are working on they are working on they are working on now or something they are now or something they are now or something they are trying trying trying to test against to test against to test against historical data. And historical data. And historical data. And then we go then we go then we go through a through a through a review process: what did you review process: what did you review process: what did you remove, what did remove, what did remove, what did our system remove, how do they our system remove, how do they our system remove, how do they align? Does align? Does align? Does this match this match this match the information you the information you the information you want to see? Do want to see? Do want to see? Do you have different types of you have different types of you have different types of standards? So if standards? So if standards? So if we go back, we go back, we go back, for example, to the different for example, to the different for example, to the different types of documents here, types of documents here, types of documents here, let's go back let's go back let's go back to the intake form.

to the intake form. to the intake form. This form may contain This form may contain This form may contain different dates. You different dates. You different dates. You need the date of the injury need the date of the injury need the date of the injury or the date the report or the date the report or the date the report was filed, was filed, was filed, right? So it's a matter of right? So it's a matter of right? So it's a matter of aligning business aligning business goals. Let's go through the goals. Let's go through the goals. Let's go through the process for a few process for a few process for a few standard cases. standard cases. standard cases. If everything works well If everything works well , we start a , we start a , we start a larger pilot larger pilot larger pilot project, and then project, and then project, and then move on to the move on to the move on to the interesting stage of interesting stage of interesting stage of integrating with integrating with integrating with suppliers, suppliers, suppliers, going beyond going beyond going beyond simple simple manual document uploads, to manual document uploads, to API integrations and API integrations and API integrations and working with all the working with all the working with all the complexities. So as long as complexities. So as long as complexities. So as long as there is a clear business there is a clear business outcome and goal for the outcome and goal for the outcome and goal for the transformation, it is a transformation, it is a transformation, it is a fairly straightforward fairly straightforward fairly straightforward process. And again process. And again , as we go through , as we go through , as we go through this journey, sometimes we this journey, sometimes we this journey, sometimes we don't have ready-made solutions, don't have ready-made solutions, don't have ready-made solutions, sometimes we do, and sometimes we sometimes we do, and sometimes we sometimes we do, and sometimes we need multiple need multiple need multiple integrations with integrations with integrations with internal internal internal systems. So, yes. systems. So, yes. systems. So, yes. So, let's start with the So, let's start with the So, let's start with the first part of first part of first part of the question: how do we the question: how do we the question: how do we comply with comply with comply with regulatory standards and regulatory standards and regulatory standards and how do we collaborate with how do we collaborate with how do we collaborate with partners? It partners? It partners? It depends on depends on depends on how fast how fast how fast people move. And it people move. And it people move. And it also depends on also depends on also depends on how how how quickly quickly quickly regulatory requirements change. regulatory requirements change. regulatory requirements change. The system we have now The system we have now The system we have now set up always set up always set up always requires requires requires human participation to make a human participation to make a human participation to make a decision. We provide decision. We provide decision. We provide this data, for example, in the this data, for example, in the this data, for example, in the case of conflicting case of conflicting case of conflicting diagnoses: we'll diagnoses: we'll diagnoses: we'll show: "Hey, on show: "Hey, on show: "Hey, on page two and on page two and on page two and on page twenty page twenty page twenty the diagnoses the diagnoses the diagnoses are different, you are different, you are different, you should look at that should look at that should look at that , right?" So as , right?" So as , right?" So as long as the person is involved long as the person is involved long as the person is involved in the process, I would say in the process, I would say in the process, I would say that it removes about that it removes about that it removes about 70% of the regulatory 70% of the regulatory 70% of the regulatory issues that issues that issues that people have. Regulation is people have. Regulation is people have. Regulation is also constantly also constantly also constantly catching up with catching up with catching up with reality, so there is reality, so there is reality, so there is a lot of material a lot of material a lot of material to read, such to read, such as the NAIC reports I as the NAIC reports I as the NAIC reports I mentioned; This mentioned; This mentioned; This organization publishes organization publishes organization publishes many different reports.

many different reports. many different reports. There are relevant There are relevant There are relevant recommendations regarding recommendations regarding recommendations regarding this. We also this. We also this. We also communicate with communicate with communicate with internal internal internal management departments management departments management departments and regulatory and regulatory and regulatory teams. And if teams. And if teams. And if they want to see they want to see they want to see some information, we'll some information, we'll some information, we'll share it with them, share it with them, share it with them, like how like how like how our AI pipeline works, right? our AI pipeline works, right? our AI pipeline works, right? This is the type of data This is the type of data This is the type of data used. used. used. Ultimately, as long as you don't Ultimately, as long as you don't Ultimately, as long as you don't create a "black create a "black create a "black box," as long as everything box," as long as everything box," as long as everything can be verified and can be verified and can be verified and the person remains in the the person remains in the the person remains in the control loop, control loop, control loop, it solves a lot of it solves a lot of it solves a lot of issues. In addition to the issues. In addition to the issues. In addition to the regulatory regulatory regulatory aspect, there is a whole aspect, there is a whole aspect, there is a whole layer of business decisions, layer of business decisions, layer of business decisions, such as how such as how such as how to approach " to approach " nuclear verdicts." “ nuclear verdicts." “ Nuclear verdicts” are Nuclear verdicts” are Nuclear verdicts” are very large very large very large payouts or claims of payouts or claims of payouts or claims of $10 million, $10 million, $10 million, or “thermonuclear” ones of $ or “thermonuclear” ones of $ or “thermonuclear” ones of $ 100 million, which 100 million, which 100 million, which significantly significantly significantly redistribute redistribute redistribute business reserves. So business reserves. So business reserves. So for these cases, for these cases, for these cases, people usually people usually people usually come with a certain come with a certain come with a certain approach to how approach to how approach to how they want to work with them they want to work with them they want to work with them . And within the . And within the . And within the platform, you can platform, you can platform, you can say, “I’ll say, “I’ll say, “I’ll process my process my process my most complex claims most complex claims most complex claims first.” Or I'll first.” Or I'll first.” Or I'll take the one that take the one that take the one that was served first. was served first. was served first. Um, and these are all kind of Um, and these are all kind of Um, and these are all kind of business decisions business decisions business decisions that people can that people can that people can make. Um, the second make. Um, the second make. Um, the second part of the question is about the part of the question is about the part of the question is about the internal goals of internal goals of internal goals of people's organizations. Um people's organizations. Um , I mean, we , I mean, we , I mean, we want to work with want to work with want to work with partners who partners who partners who share the same share the same share the same approach to improving approach to improving patient care, patient care, reducing delays in reducing delays in reducing delays in claims processing, and claims processing, and claims processing, and making the best making the best making the best decisions. Often our decisions. Often our decisions. Often our clients don't even clients don't even clients don't even think too much think too much think too much about the cost of about the cost of about the cost of medical medical medical care itself, but about care itself, but about care itself, but about everything around it, everything around it, everything around it, right? So if you're a right? So if you're a right? So if you're a big insurance big insurance big insurance company, you're actually company, you're actually company, you're actually spending hundreds of spending hundreds of spending hundreds of millions of dollars on millions of dollars on millions of dollars on legal fees, legal fees, legal fees, right? This is something that's

right? This is something that's right? This is something that's not even related to the not even related to the not even related to the patient experience, and patient experience, and patient experience, and there's a lot that can be there's a lot that can be there's a lot that can be optimized and optimized and optimized and accelerated there, right? accelerated there, right? accelerated there, right? Um, there are Um, there are Um, there are regulatory issues regulatory issues , as in certain , as in certain , as in certain jurisdictions, like jurisdictions, like jurisdictions, like California, cases California, cases California, cases often end up in often end up in often end up in court. Um, you need to court. Um, you need to court. Um, you need to figure out how to work with that figure out how to work with that figure out how to work with that . Um, there are a whole . Um, there are a whole . Um, there are a whole bunch of intermediaries bunch of intermediaries bunch of intermediaries who are also involved in the who are also involved in the who are also involved in the process and who are also process and who are also process and who are also working inefficiently working inefficiently . So, depending on . So, depending on . So, depending on the client, everyone has a the client, everyone has a the client, everyone has a different goal. Um, but in different goal. Um, but in different goal. Um, but in our conversations we our conversations we our conversations we always come to the point always come to the point always come to the point that there is an obvious that there is an obvious that there is an obvious problem that problem that problem that needs to be solved with needs to be solved with needs to be solved with technology technology . Let's solve this . Let's solve this . Let's solve this first. Um, and again first. Um, and again first. Um, and again , insights , insights , insights start to surface start to surface start to surface later when they say, “ later when they say, “ Oh, I didn’t know that about Oh, I didn’t know that about Oh, I didn’t know that about my business.” Let's my business.” Let's my business.” Let's decide this later. decide this later. decide this later. Yes, the question is Yes, the question is Yes, the question is about the about the about the decision-making API in Jev. decision-making API in Jev. decision-making API in Jev. Um, as part of that, I Um, as part of that, I Um, as part of that, I don't know if anyone don't know if anyone don't know if anyone has seen the hard copies of the has seen the hard copies of the has seen the hard copies of the AI ​​news that are AI ​​news that are AI ​​news that are handed out at the handed out at the handed out at the conference. There's an conference. There's an conference. There's an article I'm writing for article I'm writing for article I'm writing for that. Um, so we've been that. Um, so we've been that. Um, so we've been using using using classification models for a long time. classification models for a long time. classification models for a long time. I usually I usually I usually had a strong opinion had a strong opinion had a strong opinion on this. on this. on this. She hasn't come out yet today. She hasn't come out yet today. She hasn't come out yet today. She will appear in the She will appear in the She will appear in the next issue. next issue. next issue. Um, I think it's very Um, I think it's very Um, I think it's very useful, but an LLM isn't the useful, but an LLM isn't the useful, but an LLM isn't the right solution right solution right solution for everything, right?

for everything, right? for everything, right? So, when we So, when we So, when we talked about this, right? talked about this, right? talked about this, right? This AI segment This AI segment This AI segment can also be divided, can also be divided, can also be divided, right? Classification, right? Classification, right? Classification, extractive models, extractive models, extractive models, compared to more compared to more compared to more complex complex complex decision-making models. So decision-making models. So decision-making models. So for us, cost is for us, cost is for us, cost is really important. That's really important. That's really important. That's why we also why we also why we also trained our own trained our own trained our own models. Accuracy is very models. Accuracy is very models. Accuracy is very important, as is the issue of important, as is the issue of important, as is the issue of hallucinations. So I hallucinations. So I hallucinations. So I think Jev wasn't that think Jev wasn't that think Jev wasn't that surprised that it surprised that it surprised that it came up. What was more came up. What was more came up. What was more surprising was how surprising was how surprising was how long it took, because I long it took, because I long it took, because I think the phenomenon of think the phenomenon of think the phenomenon of using LLM to using LLM to using LLM to solve any solve any solve any problem due to the problem due to the problem due to the maximum maximum maximum number of tokens number of tokens number of tokens went on for a very long time went on for a very long time went on for a very long time until we started until we started thinking a lot more about the costs thinking a lot more about the costs this year. So, once this year. So, once this year. So, once Javis is up and running, Javis is up and running, Javis is up and running, you can you can you can test it out. We test it out. We test it out. We created our own created our own created our own models, keeping the models, keeping the models, keeping the same same same paradigm in mind. Therefore. paradigm in mind. Therefore. paradigm in mind. Therefore. Yes, I will answer the Yes, I will answer the Yes, I will answer the second part. As second part. As second part. As an organization, we haven't an organization, we haven't an organization, we haven't changed the entire industry changed the entire industry changed the entire industry to the extent we to the extent we to the extent we would like, right? Therefore, we would like, right? Therefore, we would like, right? Therefore, we cannot say that we cannot say that we cannot say that we have reduced the have reduced the have reduced the application processing time by an application processing time by an application processing time by an average of 5 days average of 5 days average of 5 days for each patient. for each patient. for each patient. We would really like to We would really like to We would really like to say that. But we are say that. But we are say that. But we are not in that not in that not in that position yet. For position yet. For position yet. For individual organizations, individual organizations, individual organizations, we have case we have case we have case studies on the website.

studies on the website. studies on the website. I quickly added a I quickly added a I quickly added a few of these as well. There are quite few of these as well. There are quite few of these as well. There are quite significant results in significant results in significant results in business areas business areas business areas regarding regarding regarding cost reduction and speed. cost reduction and speed. cost reduction and speed. So you can So you can So you can check them out, check them out, check them out, and we can and we can and we can chat chat chat later if you later if you later if you need more need more need more specific context. specific context. The first notification The first notification of injury or damage of injury or damage of injury or damage may depend on may depend on may depend on the subdomain. Yes, the subdomain. Yes, the subdomain. Yes, depending on depending on depending on the jurisdiction, there are the jurisdiction, there are the jurisdiction, there are also requirements for also requirements for also requirements for execution time when execution time when execution time when processing them. For example, in processing them. For example, in processing them. For example, in Texas, you only have a Texas, you only have a Texas, you only have a few hours to few hours to few hours to process your application to process your application to process your application to receive a response. In receive a response. In receive a response. In other states, it may other states, it may other states, it may be 7 or 21 days to be 7 or 21 days to be 7 or 21 days to receive a response. receive a response. receive a response. The situation The situation The situation becomes more complicated if becomes more complicated if becomes more complicated if you are an employee who you are an employee who you are an employee who has worked in a factory for, has worked in a factory for, has worked in a factory for, say, 30 years and you are say, 30 years and you are say, 30 years and you are 60 years old, because you have a 60 years old, because you have a 60 years old, because you have a large medical large medical large medical card, right? There is card, right? There is card, right? There is a lot of information here, as a lot of information here, as a lot of information here, as in the original report in the original report , but for a full , but for a full , but for a full analysis of the statement analysis of the statement analysis of the statement you need to go back you need to go back you need to go back and try to and try to and try to get it. Perhaps get it. Perhaps get it. Perhaps your last name your last name your last name changed due to changed due to changed due to marriage. Maybe marriage. Maybe marriage. Maybe you changed cities or you changed cities or you changed cities or medical medical medical providers, right?

providers, right? providers, right? So, you need to So, you need to So, you need to restore this entire restore this entire restore this entire complete history. And I complete history. And I complete history. And I believe that, unfortunately, in believe that, unfortunately, in believe that, unfortunately, in many court many court many court cases, information is cases, information is cases, information is intentionally intentionally intentionally omitted because omitted because omitted because many many many organizations have not organizations have not organizations have not undergone digital undergone digital undergone digital transformation and transformation and transformation and cannot process cannot process cannot process cases cases cases quickly enough. If that were quickly enough. If that were quickly enough. If that were all it took, it would be all it took, it would be all it took, it would be much faster, much faster, much faster, but unfortunately or more but unfortunately or more but unfortunately or more realistically speaking realistically speaking , there is a much , there is a much , there is a much broader backstory.