(upbeat music) >> Good morning or good afternoon, or good evening depending on where you are, and welcome to this session: Reduce claims fraud with data. Very excited to have you all here. My name is Monique Hesseling and I'm Cloudera's managing director for the insurance vertical. First and foremost, we want to let you know that we know insurance. We have done it for a long time collectively, personally, I've done it for over 30 years. And, you know, as a proof of that, we want to let you know that we insure, we insure as well as we do data management work for the top global companies in the world, in north America, over property casualty, general insurance, health, and life and annuities. But besides that, we also take care of the data needs for some smaller insurance companies and specialty companies. So if you're not one of the huge glomar, conglomerates in the world, you are still perfectly fine with us. So why are we having this topic today? Really digital claims and digital claims management is accelerating. And that's based on a couple of things. First and foremost, customers are asking for it. Customers are used to doing their work more digitally over the last decennium or two. And secondly, with the last year or almost two, by now with the changes that we made in our work processes and the society at large around Covid, both regulators, as well as companies, have enabled digital processing and a digital journey to a degree that they've never done before. Now that had some really good impacts for claims handling. It did meant that customers were more satisfied. They felt they have more control over their processes in the claims, in the claims experience. It also reduced in a lot of cases, both in commercial lines, as well as in personal lines, the time periods that it took to settle on a claim. However, it, the more digital you go, it, it opened up more access points for fraudulence activities. So unfortunately we saw indicators of fraud, and fraud attempts, you know, creeping up over the last time period. So we thought it was a good moment to look at, you know, some use cases and some approaches insurers can take to manage that even better than they already are. And this is how we plan to do that. And this is how we see this in action. On the left side, you see progress of data analytics and data utilization around in this case, we're talking about claims fraud, but it's a generic picture. And really what it means is most companies that start with data efforts pretty much started around data warehousing and preliminary analytics and all around BI and reporting, which pretty much is understanding what we know, right? The data that we already have utilizing that to understand better what we know already. Now, when we move to the middle blue color, we get into different types of analytics. We get into exploratory data science, we get to predictions and we start getting in the space of describing what we can learn from what we know, but also start moving slowly into predicting. So first of all, learn and gather insights of what we already know, and then start augmenting with that with other data sets and other findings, so that we can start predicting for the future, what might happen. And that's the point where we get to AI, artificial intelligence and machine learning, which will help us predict which of our situations and claims are most likely to have a potential fraud or abuse scenario attached to it. So that's the path that insurers and other companies take in their data management and analytics environments. Now, if you look at the right side of this slide, you see data complexity per use cases in this case in fraud. So the bubbles represent the types of data that are being used, or the specific faces that we discussed on the left side. So for reporting, we used a DBA data policy verification, claims, files, staff data, that it tends to be heavily structured and already within the company itself. And when you go to the middle to the more descriptive basis, you start getting into unstructured data, you see a lot of unstructured text there, and we do a use case around that later. And this really enables us to better understand what the scenarios are that we're looking at and where the risks are around, in our example today, fraud, abuse and issues of resources. And then the more you go to the upper right corner, you see the outside of the baseball field, people refer to it, you see new unstructured data sources that are being used. You tend to see the more complex use cases. And we're looking at picture analysis, we're looking at voice analysis there. We're looking at geolocation. That's quite often the first one we look at. So this slide actually shows you the progress and the path in complexity and in utilization of data and analytical tool sets to manage data fraud, fraud use cases optimally. Now how we do that and how we look at that at Cloudera is actually not as complicated as this slide might want to, to, to give you an impression. So let's start at the left side, at the left side, you see the enterprise data, which is data that you as an organization have, or that you have access to. It doesn't have to be internal data, but quite often it is. Now that data goes into a data journey, right? It gets collected first. It gets manipulated and engineered so that people can do something with it. It gets stored something, you know, people need to have access to it. And then they get into analytical capabilities for insight gathering and utilization. Now, especially for insurance companies that all needs to be underpinned by a very, very strong security and governance environment. Because if not the most regulated industry in the world, insurance is awfully close. And if it's not the most regulated one, it's a close second. So it's critically important that insurers know where the data is, who has access to it, for what reason, what is being used for, so terms like lineage, transparency are crucial, crucially important for insurance. And we manage that in the shared data experience that goes over the whole Cloudera platform and every application, or tool, or experience, you use within Cloudera. And on the right side, you see the use cases that tend to be deployed around claims and claims fraud, claims fraud management. So over the last year or so, we've seen a lot of use cases around upcoding, people get one treatment or one fix on a car, but it gets coded as a more expensive one. That's a fraud scenario, right? We see also the more classical fraud things and we see anti-money laundering. So those are the types of use cases on the right side that we are supporting on the platform around claims fraud. And this is an example of how that actually looks like. Now, this is a one that it's actually a live one of a company that had claims that dealt with health situations and pain killers. So that obviously is relevant for health insurers, but you also see it in, in auto claims and car claims, right? You know, accidents. There are a lot of different claims scenarios, that have health risks associated with it. And what we did in this one is, we joined tables in a complex schema. So you have to look at the claimant, the physician, the hospital, all the providers that are involved, procedures that are being deployed medically, medicines has been utilized to uncover the full picture. Now that is a hard effort in itself, just for one claim at one scenario. But if you want to see if people are abusing, for example, painkillers in this scenario, you need to do that over every instant that this member, this claimant has, you know, with different doctors, with different hospitals, with different pharmacies or whatever, That classically it's a very complicated and complex the and costly data operations. So nowadays that tends to be done by graph databasing, right? So you put fraud rings within a graph database and walk the graph. And if you look at it here in that, you can see that in this case, that is a member that was shopping around for painkillers and went to different systems, and different providers to get multiple of the same big LR stat. You know, obviously we don't know what he or she did with it, but that's not the intent of the system. And that was actually a fraud and abuse case. So I want to share some customer success stories and recent AML and fraud use cases. And we have a couple of them and I'm not going to go in an awful lot of detail about them because we have some time to spend on one of them immediately after this. But one of them, for example, is voice analytics, which is a really interesting one. And on the baseball slide that I showed you earlier, that would be a right upper corner one. And what happened there is that an insurance company utilized the, the divorce records. They got from the customer service people, to try to predict which one were potentially fraudulence. And they did it in two ways. They look at actually the contents of what was being said. So they looked at certain words that were being used, certain trigger words, but they also were looking at tone of voice, pitch of voice, speed of talking. So they try to see trends there, and hear trends that would, that would ping them for a potential bad situation. Now, good and bad news of this proof of concept was, it's, we learned that it's very difficult, just because every human is different to get an indicator for bad behavior out of the pitch or the tone or the voice, you know, or those types of nonverbal communication in voice. But we did learn that it was easier to, to predict if a specific conversation needed to be transferred to somebody else based on emotion. You know, obviously as we all understand, life and health situations tend to come with emotions. Also, people either got very sad or they got very angry or, so the proof of concept didn't really get us to affirm understanding of potential fraudulence situation, but it did get us to a much better understanding of workflow around claims escalation in customer service, to route people, to the right person, depending on, you know, what they need, in that specific time. Another really interesting one, was around social media, geo open source, all sorts of data that we put together. And we linked to the second one that I listed on the slide here that was an on-prem deployment. And that was actually an analysis that regulators were asking for in a couple of countries for anti-money laundering scams, because there were some plots out there that networks of criminals would all buy low value policies, surrender them a couple of years later. And in that way, got criminal money into the regular amount of monetary system, whitewashed the money and this needed some very specific and very, very complex link analysis because there were fairly large networks of criminals that all needed to be tied together with the actions, with their policies to figure out where potential pin points were. And that also obviously included ecosystems, such as lawyers, administrative offices, all the other things. Now, but most, you know, exciting, I think that we see happening at the moment and we, we, you know, our partner, of analytics just went live with this with a large insurer, is that by looking at different types, that insurers already have, unstructured data, their claims notes, reports, claims filings, statements, voice records, augmented with information that they have access to, but that's not theirs. So it's just geo information obituary, social media, deployed on the cloud, and we can analyze claims much more effectively and efficiently, for fraud and litigation than ever before. And the first results over the last year or two, showcasing a significant decrease, significant decrease in claims expenses and, and an increase at the right moment of what a right amount in claims payments, which is obviously a good thing for insurers. Right? So having said all of that, I really would like to give Sri Ramaswamy, the CEO of Infinilytics, the opportunity to walk you through this use case, and actually show you how this looks like in real life. So Sri, here you go. >> So insurers often ask us this question, can AI help insurance companies, lower loss expenses, litigation, and help manage reserves better? We all know that insurance industry is majority, majority of it is unstructured data. Can AI analyze all of this historically, and look for patterns and trends to help workflows and improve process efficiencies. This is exactly why we brought together industry experts at Infinilytics to create the industry's very first pre-trained and pre-built insights engine called Charlee. Charlee basically summarizes all of the data, structured and unstructured. And when I say unstructured, I go back to what Monique, basically traded, you know, it is including documents, reports, third party, it reports and investigation, interviews, statements, claim notes included as well, and any third party enrichment that we can legally get our hands on, anything that helps the adjudicate, the claims better. That is all something that we can include as part of the analysis. And what Charlee does is takes all of this data and very neatly summarizes all of this, after the analysis into insights within a dashboard. Our proprietary natural language processing semantic models adds the explanation to our predictions and insights, which is the key element that makes all of our insights action. So let's just get into understanding what these steps are and how Charlie can help, you know, with the insights from the historical patterns in this case. So when the claim comes in, it comes with a lot of unstructured data and documents that the, the claims operations team have to utilize to adjudicate, to understand and adjudicate the claim in an efficient manner. You are looking at a lot of documents, correspondences reports, third party reports, and also statements that are recorded within the claim notes. What Charlee basically does is crunches all, all of this data, removes the noise from that and brings together five key elements, locations, texts, sentiments, entities, and timelines. In the next step. In the next step, we are basically utilizing Charlee's built-in proprietary natural language processing models to semantically understand and interpret all of that information and bring together those key elements into curated insights. And the way we do that is by building knowledge, graphs, and ontologies and dictionaries that can help understand the domain language and convert them into insights and predictions that we can display on the dashboard. And if you look at what is being presented in the dashboard, these are KPIs and metrics that are very interesting for a management staff or even the operations. So the management team can basically look at the dashboard and start with the summarized data and start to then dig deeper into each of the problematic areas and look at patterns at that point. And these patterns that we learn, from not only from what the system can provide, but also from the historic data, can help understand and uncover some of these patterns in the newer claims that are coming in. So important to learn from the historic learnings and apply those learnings in the new claims that are coming in. Let's just take a very quick example of what this is going to look like for a claims manager. So here the claims manager discovers from the summarized information that there are some problems in the claims that basically have an attorney involved. They have not even gone into litigation and they still are, you know, experiencing a very large average amount of claim loss when they compare to the benchmark. So this is where the manager wants to dig deeper and understand the patterns behind it from the historic data. And this has to look at the wealth of information that is sitting in the unstructured data. So Charlee basically pulls together all these topics, and summarizes these topics that are very specific to certain losses combined with entities and timelines and sentiments, and very quickly be able to show to the manager where the problematic areas are and what are those patterns leading to high, severe claims, whether it's litigation or whether it's just high, severe indemnity payments. And this is where the managers can adjust their workflows, based on what we can predict using those patterns that we have learned and predict the new claims. The operations team can also leverage Charlee's deep level insights, claim level insights, in the form of red flags, alerts and recommendations. They can also be trained using these recommendations, and the operations team can mitigate the claims much more effectively and proactively, using these kind of deep level insights that need to look at unstructured data. So at the, at the end, I would like to say that it is possible for us to achieve financial benefits, leveraging artificial intelligence platforms like Charlee and help the insurers learn from their historic data and being able to apply that to the new claims, to work, to adjust their workflows efficiently. >> Thank you very much Sri. That was very enlightening as always. And it's great to see that actually, some of the technology that we all work so hard on together, comes to fruition in, in cost savings and efficiencies and, and help insurers manage potential bad situations, such as claims fraud better, right? So to close this session out as a next step, we would really urge you to assess your available data sources and advanced or predictive fraud prevention capabilities, aligned with your digital initiatives to digital initiatives that we all embarked on, over the last year are creating a lot of new data that we can use to learn more. So that's a great thing. If you need to learn more, want to learn more about Cloudera and our insurance work and our insurance efforts call me, I'm very excited to talk about this forever. So if you want to give me a call or find a place to meet, when that's possible again, and schedule a meeting with us. And again, we love insurance. We'll gladly talk to you until SDC and parts of the United States, the cows come home about it. And we're done. I want to thank you all for attending this session, and hanging in there with us for about half an hour. And I hope you have a wonderful rest of the day.