Laura Smith: Hey, everyone. Good morning. Thank you for attending. Today's webinar combating claim fraud reduce false positives with AI llm and graph databases.
Laura Smith: Our speakers today are Scott Heath, Vp. Of fraud and analytics with experimental, and Michael Moore, senior director strategy and innovation with Neo for Jay before we get started. I wanted to go over a few housekeeping items. There's A. QA. Icon at the bottom of your Zoom Meeting. If you have any questions during the webinar. Please leave them in that QA. Box that'll pop up when you click the button.
Laura Smith: and then we can go over that at the end of the session we'll have some Q. And A,
Laura Smith: we are recording today's session. I'll be sending a follow up email this week with the recording. And if you have any questions or issues during the webinar, please. Message the host. Thank you.
Scott Heath: Thanks, everybody. My name is Scott, and I have really exciting information that Michael and I are going to share with you today. What I'm going to do is probably going to drop out my video so that the presentation that I show will be more clear. But, Mike, do you want to say Hi.
Michael Moore: hey, folks? Thank you for taking the time to join us today. We're really excited to be
Michael Moore: presenting with one of our leading partners at Neo for Jxparrow, who have a ton of really great solutions and accelerators to help you get started with graph and graph based solutions
Scott Heath: super. So without further ado, we'll go ahead and jump in.
Scott Heath: We're gonna walk through kind of the high level. And then I'm gonna talk a little bit about some of the concepts. And then Michael is gonna bring us home what what we're gonna see today is a couple of vignettes of of what life would look like with this kind of technology. And then we'll talk specifically about why Neo, for Jay is so much different and is such a game changer. So before we get started.
Scott Heath: let's talk a little bit about sort of the backdrop of of what's going on. So in a lot of the situations, nobody has to tell us that
Scott Heath: claim fraud is in many different flavors right? It's out there and the numbers can be big. The other thing that we want to keep in mind is not only is it auto health life. Er it could be anything right. But the takeaway is as payers and and providers, and claim
Scott Heath: sort of investigation, or, quite frankly, just claims in general. Is. The process is that if you pay and you pay incorrectly, it is extremely difficult to get that money back. And so, as such, we really really need to keep our net promoters
Scott Heath: scores high, we need to be as as good as we can in the claim process, and quite frankly, after a couple of the claim conventions that we've been through this year. Speed is of the essence. However, we now need to look out for where this fraud occurs, we need to find it fast. We have to look for non-trivial, complex networks. And then, more importantly, we still have to do more with less. Well, most most of the folks that are out there today
Scott Heath: departments are under
Scott Heath: sort of stress to make sure that you do it faster. And you do all these things well, if you do it wrong, there's almost no way to get that money back. And some people just sort of call that part of doing business. Well, we think today we can do better. And so part of that, what we're gonna talk today is
Scott Heath: looking at, regardless of what line of business you're in, or, or, again, where you are. It is part of this claim process that's so incredibly important. And how do we suss it out. How can we look at things like upcharging and up coating disaster or storm chasing? We'll see a little bit of that today.
Scott Heath: Bring fraud rings or collusion, and and some of these other elements. Those are all real right. And then there are probably more lists of these, depending on what line of business again. Those those things that come at us in different ways.
Scott Heath: The good news that we we are bringing to you is, is now this premise. Of what sort of technology can I bolt onto? And why? Why is this so hard to do when you look at the complexity behind the scenes?
Scott Heath: You know, we have lots of data. But this paying first, and the connection or the disconnection of these dependencies is very difficult. Right? If I have SQL. Databases, and they're they're separated from each other. There's no way to sort of identify when I paid, if I paid properly. And then this idea of going back in time. It's very hard to look at things the way they were, or to go back and triage how those things happen. But again.
Scott Heath: it's ultimately leading to this scoring. If I can simply share with the human things in in near time or real time, it makes an enormous difference. And then being able to tie back time and that data dependency of the of the connections of the information.
Scott Heath: And then, finally, when I see that in history and then in real time. Obviously, this is an enormous problem. And so what we hope to do today is demystify some of that and bring to you some some good news. Now.
Scott Heath: back in the day we've all seen the movies with the the red lines. And this case was that, and this is over here. Well, we can do better. And and that today in in what we call graph analytics. Now, that's not a pie chart graph.
Scott Heath: That is a style of data. That neo for Jay is quite frankly the leader. And and again, Michael is gonna walk us through more details. But at a at a macro level for business people, you don't have to rip out what you're doing today. This is a bolt on relationship. But what graphs do that are intrinsically different is instead of saving the the the again, if you're not tech technology based
Scott Heath: the one to many big tables, table walking, lots of outer joins. Lots of complicated words means expensive. And sometimes for what we're looking at in these connections can be very slow. But that doesn't mean we have to throw that out. What we're now saying is with the Neo for J, which now saves those relationships and at scale allows me to think in more of a logical way. But inside of
Scott Heath: what we'll see inside of neo for Jay, then, is this ability to do logical becomes physical. That helps us both understand it. And where we're going here in a minute. Is this sort of concept that whether you're in
Scott Heath: parts of insurance.
Scott Heath: underwriting, or whatever it is, everything is actually connected to everything else. It's about how the data is connected. And this is really where a graph will stand tall. This can make an enormous amount of difference both in understanding the complexity of a claim and a dispute and multiple claims and rings.
Scott Heath: or all the way over on the right now, where those things came from as they worked their way to the left. So if I have a claim, where was it? What were the deductibles. What were the risk? Factors? What was underwriting? And where was all of that? Over on the left? All of that can now very concisely be held inside of a graph database and at our fingertips to do lots of different kinds
of use cases. Not just lost waste and abuse
Scott Heath: potentially fraud. But I can actually start to look at positive use cases, too. And that's really sort of the other exciting thing about this. And so when we see this
Scott Heath: we see now over on the left. I can do things like customer journey upsell cross, sell chargebacks. But what we really came to talk about today is perhaps fraud, waste, and abuse. And where does a claim and dispute turn into fraud? Or how can I start to identify in very simplistic ways? Well over on the right now, what we're getting from Neo for Jay is. Imagine now, if I had a smart system
Scott Heath: that could indeed look for patterns and look for what if, or predictions, or giving the human a scorecard that we'll see here in a minute for a live demonstration, where they could be prompted to say why and what
Scott Heath: and then I can see it very concisely and I can plug into my existing claim system, or I could bolt onto any of the the systems that I may have today. And that's really the power of this is Number one. Be able to find it and to be able to do it in in a much faster and more concise way. Right?
Scott Heath: So again, some people like things from top to bottom, some things like things from left to right. But what we see now is the entire sort of capability of a graph database. But we're really segmenting it over here, and maybe we want to use it for other use cases. But today we'll focus on that. So that's kinda setting the table right. We're trying to mitigate our risk and decrease our cost.
Scott Heath: For what that is. Now, where does this fit in the overall landscape? In one of the conferences that we were recently at connected claims. They very clearly called it out. That claim fraud is simply a bolt on to our claim process or in a technology stack. It is part of that early warning system that radar that we can give either
Scott Heath: investigators or just rank and file claim administrators. Right? Where are those items? And can we put this to work? Is what we're gonna see today. But the ecosystem has a place for that right? And and that's really what what we wanna show today is, why is this so powerful? And how can it work in line to save us that time. Save us that money decrease our false positives, etc.
Scott Heath: Now, one of the products that we integrate with is something called guidewire. Well, they do a full claim management process. And, as you can see here is that there really isn't a call out. And what we want to show today, then, is during that claim intake
Scott Heath: in the assigning. And the evaluation process is really where we're looking at today. Focusing is that if we can automatically notify. The teams that are working on those and simply in English, point out what the anomalies are and where there may either be a potential issue. Or if there's a problem, or even if it's all the way over to fraud, that's really sort of how we do this. But again, a good backdrop on sort of where this fits in the overall process.
Scott Heath: Now, what is this graph thing? Right again? Michael's going to talk to us more about this specifically. But what we've seen over time, then, is, is back in the way back machine. Here in the s
Scott Heath: SQL. Was very, very powerful we were able to do things we had not been able to do. But then came the thousands, these dimensional bi tools. Well, I can look in my rear view mirror, and I can build really interesting connections. But looking in the future was really hard. And that's really sort of the age of this graph database that's come into line is now
Scott Heath: I can look backwards. I can look at current, and I can start to look at those patterns very simply and very easily inside of our neophy, instance. And now I can start to go with this projection of things like Llm.
Scott Heath: That that ability to sort of move up, talk to the human, the investigator, or the claim manager. And now I can start to find those patterns with a much higher accuracy, and I can decrease those false positives. So now I can start to cash that check of saving that money or denying the claim before it actually gets paid. If there are enough anomalies, or quite frankly, turning it over as potentially fraud to folks that may need it.
Scott Heath: Now, what does that really mean, then that means that if you simply throw some good machine learning at it, and we've had a customer. Tell us that was that look, machine learning is great. The problem is that it's difficult to do it at scale. But now, if I can couple that with a graph database, and I can start to look at. How do I run? Supervised and unsupervised? It gets even richer, and that's where we start to see that boost. Then, if I couple it with a human
Scott Heath: right the different team members that are out there and get the human in the loop. This is an enormous boost. And, by the way, we've done some empirical things. We'll see here in a minute. Where it can be even higher. It can be up to % better than more of a linear or simple. If, then, kind of logic, and we don't lose any risk. And I think this is a really really
Scott Heath: important sort of note here is by using the graph database. And again, we'll see some of that. What Gartner says here in a minute. But the takeaway, the real empirical data is this can make a really demonstrable effect to your top and your bottom line, as well as making a lot of folks lives easier. So
Scott Heath: how did we do this? So what I'm going to do now is I'm going to step into a little bit of what as practitioners we might look at, and then I'll hand it over after this section to to Michael.
Scott Heath: So in this side.
Scott Heath: what is this sort of process? Well, we ingest data, we match it. We do that today, right? Most of the folks in technology do a lot of that today. But what if now, I were to augment that with this graph data structure, what would that look like? Very simply, graphs are very good at connecting disparate data. In that sort of mental map that we saw previously, and where we start to do that now is, we can incorporate some of the tools from Neo for J.
Scott Heath: And now what I'd like to show you today is the connected suite of tools that would work for claim managers where we can now see those algorithms. I can see how mere sort of mortals, if you will, can wield this incredibly powerful new technology and that it is fairly straightforward to use. And it doesn't upset what our flow or what our processing of those claims are. And so the beauty of this now is sort of plugging it in as we're as we're perhaps a a system that we already have.
Scott Heath: Now, how does all this fit together so logically, even though you may use guidewire today? Or you may use one of these other tools. That's okay. These tools can plug into those. Or if you don't have one of those, well, obviously, we have. We have tools for claim management as well. But again, we're using that data from others. We're connecting the dots, perhaps on a verisk or an open corporates against guide wire and claims and data.
Scott Heath: And what we're now able to do is make smart connections. And we can now here in the middle, start to use that power of the Neo for J. Graph data Science Library, which we'll hear about here in a minute in addition to machine learning. And what that does now is that gives us this very simplistic way for for different kinds of roles
Scott Heath: to start to go and and investigate that, and right increase that accuracy and the velocity of our claim processing many times. That's up to or % on the speed or the velocity of that. So it's now helping us in a couple of ways.
Scott Heath: Now, when we get into sort of what is behind the scenes. And again, Michael will talk a little bit about this. But these different kinds of algorithms are extremely powerful. So to be able to find connections or similarities of people that have been trying to either defraud us as a ring or as an individual looking for recommendations or shortest paths to other
Scott Heath: kinds of of connections, etc. That's really where neo is gonna help us here to go do that. And what do these things look like? Well, what we're gonna see here in a minute is there is something that we call a logic or an alert builder that a super user can do they can simply use these similarity of patterns. And we're gonna see how we do that for a couple of different objects
Scott Heath: to be able to do that. But this is really now what neo for Jay is bringing to us underneath. And so a quick version. And again, this is something no one would ever see. But it simply looks at that claim data over on the right, and it actually starts to connect those dots. And what we may see is things in blue or green are good, and those are the ones that sail right through the process, and maybe the ones that are in red or magenta at the bottom are ones where addresses were. Were you reused by witnesses or addresses, or incongruent for witnesses and doctors, and those kinds of things that what we see then is, it's very easy for that pattern detection.
Scott Heath: and then for new claims coming in, we can find it. And then what it looks like on a screen is simply reject. You don't have to worry about all the complexity of what just went on. But we're able to surface that in a very simplistic way. And so what we see then are simple kinds of of user interfaces. These addresses are not correct. Show me what that looks like, right.
Scott Heath: this connection of claims and not just this claim. Maybe there were multiple claims through several carriers, something that we can pull in from Verisk, where we can make those connections and realize. Well, this is not just a singular event, and it might be happening over time in other areas. So again, this power is really getting us to that next level. So that's really sort of what we're trying to bring to you today is talking about how to increase that
Scott Heath: speed. Use this power to go, do that. And without further ado, then that's what I'm gonna show here in a minute. Right? Is this top layer is simply that we now have clean persons. Maybe you're in customer ,
Scott Heath: and and you have other kinds of products and services. And then you can see that in there. And that's really what we're trying to show now is the normal business of insurance and claim operations. Stay the way they are. And we're now bolting on. And then, if you're a graph data science person, or you're in sort of more of that, it user. The Neo J products have a wonderful set of user interfaces, so that now you can build and test those kind of logic elements. Now that you can start to incorporate into production.
Scott Heath: So this is kind of that logical layer for the technology folks that are in there. Neo, for J is is basically connected to the expiro we call it a semantic layer, so that you're able to sort of use it out of the box, and I think that's the other sort of big component here is. There's not a huge sort of ramp or or time curve. This can actually be done pretty quickly.
Scott Heath: Couple of different modules. And again, your mileage may vary on these, but but really, what we're saying is, they are flexible, and they are connectable. To the existing claim management solutions that are out there today. But again, they're there for ones that you need, or if you don't need all of these, you can. You can pick and choose.
Scott Heath: Now, what I'd like to do is switch over to the demo
Scott Heath: as I'm switching over here. Let's see if I can do this
Scott Heath: alright. So in our first demonstration, what I wanna do now is sort of share. What this would look like. Now, I'm gonna show different roles in my demonstration. The first role is more of an administrator or a super user. In this instance. What I've seen then is, I am an insurance company that is looking at property and casualty.
Scott Heath: and auto and home. And so in this vignette. What I see then, is a dashboard.
Scott Heath: The dashboard is showing me a couple of different elements. Number one is. It's showing me geography and across the bottom. I'm looking at time now during that weather event. In this case it was a series of violent storms. I see.
Scott Heath: That ingest of different data I may have loaded Verisk data. I may have loaded national weather data in this case. That's what I've done here. And I can start to see an overlay and I'm seeing a risk probability that I have already run inside of Neo for J. So it's already risk scoring for me here in a minute. I'm gonna show you how I built that with sort of a drag and drop window. But in this sort of dashboard. I'm now seeing those different elements. What I'd like to do
Scott Heath: is actually start to dig in. So now I see where there were in red where there were events where there should be claims. And then things in yellow where there might be lesser claims and definitely things in in blue. There shouldn't be any claims. So I'm now using again that graph capability. But maybe I want to zoom in here.
Scott Heath: What I can start to do now is zoom in and start to say graph database, hey? Where are their likely predictions of potential fraud? So, for instance, somebody is claiming their car was total and there was no hail in that area. Well, that is something that we're gonna look for. We'll see how we build those algorithms. So in this case, this super user is starting to see sort of where those are and what I can do then is I can zoom in. And now, what I wanna do is show me specifically
Scott Heath: where I am looking at Hotspots. And I'm actually now going to potentially go deploy claim adjusters out in the field. And so what I want to do then is, where should I send them? Number one? So I can use the graph to optimize my deployment
Scott Heath: of those different claims, persons right? And now what I can see then is, well, that is the density of what's going on. The second thing that I've done is, I see now the power of an Llm.
Scott Heath: Or a generative AI. What it's saying, then, is, I wanna see where these are high high probability of what those are, and what I've done now is, I popped up and said, Would you like me to go build an alert right? Would you like for me to go find
Scott Heath: that other data to go do that. And so what I wanna do now is, actually I wanna pop over here. And I can see now that I can go build those different kinds of alerts. In this case I perhaps wanna go look for maybe there's in this particular instance.
Scott Heath: I want to go see what's going on. Maybe they're I'm looking for in this case a syndicate of persons.
Scott Heath: Well, what I can simply do is drag the data from my Neo for J data source. I can say, well, in this case I've grabbed an insurance agency you can see over here. I could grab a block over here for a customer or any other kind of data. But what I'm doing is I'm building this logic tree, and I'm saying that if there was an auto and maybe there was a connection, and there were previous claim damage in that same visual region. What I can start to do then is is, see that
Scott Heath: the other element that I can start to do then is is is go through the different process. So if I wanted to test this.
Scott Heath: I could go do that. But down here at the bottom is really where again, these network kinds of things are showing what we're doing. What I'm able to do then, is I can go down and look at these network algorithms from neophy. So things like clustering. In this case I'm looking for, strongly connected to a fraudulent claim in in the previous. I could even do fuzzy matching or cycle detection.
Scott Heath: Now, Michael's going to talk about sort of what's going on behind that. But these are those elements that are now allowing this alert logic to be so much more productive. Now I can sort of skip ahead here.
Scott Heath: and I can say, Test this
Scott Heath: and I can go see what my accuracy is. And again, I'm I'm running that in a test environment. But now what I've done is I've run that test and then I can either install it. But in this case, now, what we're gonna see here in a minute is I wanna run that I want to automatically, perhaps suspend that. And I wanna pop up a queue
Scott Heath: to my claim administrator and say, these are the or things that I would like for you to do, and then you can make the judgment. But I have brought this now to the human's attention, that if there was an auto damage, and again, this this screen that I'm showing here is typically for a super user.
Scott Heath: Now, I want to come back to my other screen over here. And what we can see then is, I'm basically going to sort of fast
Scott Heath: forward through some of those elements. But now what I can do is now that I've got this short list here. I can go in and say, well, somebody filed a claim in an area here. That is not correct. Right? What is that? That? Perhaps II wanna do here. And what it's done then is that alert that I had showed you previously has now popped up a quick list of things for me to do to say, well.
Scott Heath: as I'm looking at the claim, obviously, I have my regular claim elements. I can see that. Well, you're outside of the region.
Scott Heath: right? You shouldn't be having that kind of damage, and it looks like now the recommendation again back from Neo, for Jay has told me that. Well, you should walk through these, and I can see over here that I've got a little chat. Bot! That's saying. Well, I found these anomalies here I found that this homeowner, it basically has done some things previously where he was trying to get away with some things. I see that there was some geographic
Scott Heath: location that's outside of the band of what's going on. And again, my neophy risk algorithm has told me what that is. But what I'd like to do is, let's just step in to go see that.
Scott Heath: And now what it can do is again the graph database is bringing me that recommendation to say, I think you should reassess this, and it looks like there's a few elements in here that are indeed problematic. And so I'm gonna walk you through sort of what those are, and the next one, then, is this geographic anomaly.
Scott Heath: Now, what you can see over here on the right is that that property in question is indeed very wide of where all the heavy activity occurred. And so when we get our adjuster online, we now have effectively sort of strikes
Scott Heath: for why this? This may be a problem, and then we see some other detail down here on on sort of what their claim is. And then, finally, what I can do is I can go use that graph database and time, and say, this person has indeed done some other kinds of elements that are down here as as I'm stepping through this. And I think that's sort of the the key here is that I can say, well, this one over here in Red says.
Scott Heath: well, sure enough, he tried to previously claim something in his home. That was clearly already existing right? So we had denied that claim right, and we can see that there is an auto one down here, where, indeed, they did something very similar, where they were trying to claim something where it was next to a neighbor. And so what we start to see now is the power of the graph
Scott Heath: is not sort of in your face. But what it's doing is it's behind the scenes. And now you can start to see sort of where and why? This is so incredibly powerful with with kind of what and how this fits into the the scenery. So what I'm gonna do now is actually come back
Scott Heath: over here
Scott Heath: and share what that application is
Scott Heath: here, and then we'll go back to our previously listed Powerpoint. So as we're shifting over here. Now, what we're gonna do is we're gonna look sort of behind the scenes.
Scott Heath: And I'm gonna pass the baton over to you. Cycle.
Michael Moore: Thanks. Scott. Yeah, that was terrific.
Michael Moore: Yeah, so so what I thought we would do in this next section is you know, Scott, gave a really good rundown of the kinds of inferences and the kinds of experiences. As part of an analytical investigative flow. That the expiry connect solution provides. And so I thought what we would do is
Michael Moore: take a peek a little bit under the hood and get an understanding of exactly how the graph is. PA is creating these insights. So graphs are, really coming into their own. We're seeing the analyst community responding very well. To to advances in the space. And we're S, and you're seeing statements like %
Michael Moore: of
Michael Moore: all data and analytic innovations are going to be powered by graph just within the next few years, and we certainly are seeing that to be true across our customer base.
Michael Moore: Another in interesting quote here is that finding relationships and combinations of diverse data, using graph techniques at scale at scale will form the foundation of modern data analytics. And we're we don't have time to touch on it in this particular. webinar. But one of the things we're absolutely seeing is that all of the interest in genai and combining those conversational experiences with
Michael Moore: knowledge graphs in order to make very trustworthy and reliable applications.
Michael Moore: For enriched customer experiences. And so I think that that's gonna be a a big area of growth particularly in the insurance industry. So let's go to the next slide, Scott. So let's sell just level set. So when we talk about a graph, you know, we're not talking about charts. We're talking about this, this interesting data structure. And the data structure is composed essentially of types of data. We have nodes. So you can think of those as records. And so here I have
Michael Moore: collection of records. I have employee records, company records and city records. And then those records are then connected by relationships. And one of the differentiating features of graphs is that those relationships are stored in
Michael Moore: memory and on disk. And you can put data on them. They have directionality, and they also have a semantic type. And so you can see, for example, here that this company
Michael Moore: has a CEO. And so there's a relationship there. And so we know that this employee is, in fact, the CEO, because that that is the relationship that's being described here.
Michael Moore: We also see that this company is located in a particular city, and you can see that there's a relationship there. That specifies this. Now, this is a really toy example. But when you load millions of records into a graph, you end up with a fully connected network of data. So a fabric of data that you can then query, and you can go as deep and wide as you would like
Michael Moore: with the with your queries in a graph, because all of the logical possible relationships are being constructed and stored. And it's computationally super cheap to discover all of the connected data from a given starting point.
Michael Moore: And so you can. You know, you can go across multiple hops, you know, , , as deep as you want, and do the kinds of operations that would be physically impossible to do in legacy, SQL. Or no SQL. Environments.
Michael Moore: So that's a little bit about graphs, they naturally handle complex data. And so you'll, you know, data shapes where you might have wide data. So think of like a view. You know. Maybe I want to understand something around a particular object. All of the kind of data shapes where you've got complex, many to many linkages, hierarchies, recursion, deep paths. All of those kinds of analyses are easily performed in a graph.
Michael Moore: and then and then, in terms of how it fits into the It portfolio, as as Scott mentioned earlier on, we typically see customers putting graphs on top of, you know, legacy data silos or
Michael Moore: or data warehouses and data lakes where data has been say, centralized, but not necessarily mobilized. And and then, of course, there's a whole world of analytics of whole graph analytics where you can look at the topology of the graph run algorithms. And then and then infer additional relationships. And so we see this happening for things like link prediction or vector similarity.
Michael Moore: First, semantic search, for example, neo for J is the leader in the graph database world. We started this category about a dozen years ago. We have a huge community of developers. Over , now and and we are the, and we are the top offering and have over % market share.
Michael Moore: Let's go to the next slide.
Michael Moore: we're widely adopted. across a a wide, a wide range of vertical. So all of the North American banks are using us. All the top aircraft manufacturers, all of the top automakers, most of the top retailers, most of the telcos most of the pharmaceuticals and and in insurance. out of out of insurance companies. And on the Fortune are already using us.
Michael Moore: So next slide. And so
Michael Moore: what? What is the value that we're providing? So when you begin to connect your data and you create these knowledge graphs, where you're where you're connecting and mobilizing data from disparate domains. all kinds of value is released. And so you can do, we can support use cases around data driven discovery and innovation. Hyper personalization for retailers or app or healthcare providers, decision making and you know which is the
Michael Moore: topic of of this discussion around fraud prevention we've got. We've got banks using us for anti money laundering. We have intelligence agencies using us to, you know, identify bad actors.
Michael Moore: data integration. And then, of course, a ton of work around data science where the graphs are actually being used to either predict and solve problems inside the graph or or engineer new features that can be then exported into traditional deep learning architectures. Let's go to the next slide.
Michael Moore: The neo for J is is a really capable system. It it is able to support. major workloads associated with the data management. So yeah, you can run transactional workloads which are sometimes known as Oltp type workloads. You can run large scale analytics for reporting and analysis. So that would be called an olap workload. And we can also do machine learning workloads with our large portfolio
Michael Moore: of of graph algorithms. And so it's a really unique system in the sense that it can support all major analytics workloads all in the same environment with no movement or export of data.
Michael Moore: Let's go the next slide.
Michael Moore: And so some of the core components of Neo J is, we use what's known as a native graph architecture. And so that means that the data is stored on this in graph format, which means that there's no mathematical limit to the size of the graph. And indeed, we have customers that are
Michael Moore: running graphs that have billions of nodes and hundreds of billions of relationships we have, we have the ability to ingest data from a variety of different data types. We can, we can handle these hybrid workloads that are that have both transactional and analytical
Michael Moore: demands, and we have the largest set of integrations with in terms of tooling and drivers to support a full enterprise ecosystem.
Michael Moore: We're unique in that. We have this very large library of algorithms that are that run in a in a dedicated analytical
Michael Moore: memory space that doesn't interfere with your ongoing read, write operations. And of course we have a giant community of developers, which means, if you want to go down the path of building out a graph coe in your in your organization. You it's it's not difficult to find a highly capable and trained, you know, for J developers. So the next slide?
Michael Moore: So so let's dig into a little bit of
Michael Moore: of
Michael Moore: some examples around fraud and fraud detection. So I think everybody on the phone is probably a herd of the Panama papers, right? So this was a big leak of
Michael Moore: of information pertaining to offshore entities that stood up.
Michael Moore: A
Michael Moore: and many of those offshore entities were stood up for the purposes of tax evasion.
Michael Moore: And so
Michael Moore: one of the things that the organization, the is the international consortium of investigative journalists did is they took all of all of those documents, and they loaded them into a neophyte graph. And they basically
Michael Moore: constructed a graph that had this relatively simple design of entities with addresses and officers and intermediaries, and
Michael Moore: with this they were able to identify some very significant, politically exposed individuals who had, in fact, been staffing cash offshore in these in these special entities.
Michael Moore: and so and very difficult to unravel. All of the different layers of these of these corporations. But you put it in a graph, and it becomes very clear and you have the ability to essentially walk across all of that data and discover who is the actual ultimate owner of a set of accounts. So that's one example. So the next slide.
Michael Moore: if you think about the insurance industry graphs are actually applicable across a ho, a wide range of the data of business processes in insurance. All you know, beginning with things like personalized recommendations. You know, doing things like bundling around. You know, marketing and advertising sales lead generation. The actual process of underwriting and risk assessment. We have customers that are using graphs to improve
Michael Moore: their risk assessment capabilities. Managing policies. Determining, for example, what does a customer actually own across across across my different
Michael Moore: product lines and doing things like looking for. You know, major life events. Right? So if you know, if you have a kid that turns , hey? Guess what you know. You're going to have a new driver in the family very shortly.
Michael Moore: Things like that claims processing fraud, detection. We'll we'll touch on that in just a bit. And then, of course, things like customer support and renewals and retention.
Michael Moore: And so graphs are very good and can provide deep insights across all of these areas.
Michael Moore: But the next slide.
Michael Moore: So, for example, we have a customer that is using neophy to understand agent efficacy.
Michael Moore: And this is the design roughly of what that graph looks like. And
Michael Moore: and this particular and insurance company has a really large independent agency. And so it's very important to them that they have the ability to actually
Michael Moore: send information to agents about what their next action should be relative to data that they understand about that policy holder on on their side. And so
Michael Moore: and they're able to do things like householding. They're able to look across all of our product lines. And they're able to basically work with the agent and say, here's what we think this particular customer might be very interested in.
Michael Moore: And so that's one example.
Michael Moore: Let's go look at some fraud examples.
Michael Moore: And so in the world of claims, fraud, particularly in auto
Michael Moore: we see some really common patterns, and
Michael Moore: and the power of the graph is its ability to actually connect the data in such a way that
Michael Moore: details that might not be obvious from an individual claim when they're actually connected in in groups what we would call a subgraph. Those details jump out very quickly, and you can see that this is a really anomalous pattern. And so one of the one of the types of fraud that we see a lot
Michael Moore: I'm being explored with, perhaps, is this business of fraud rings where you have individuals who are
Michael Moore: occupying different roles
Michael Moore: across a set of seemingly independent claims. And so in this model, you might say, you know, we have say accidents, and so there's a node for accidents. And you see that there's there's a node representing different cars, and then different persons.
Michael Moore: and you can have relationships that describe
Michael Moore: what was the role of that person relative to that car and that accident. And so you see here, I have several people, and this person, number one was, was both a driver
Michael Moore: in one accident and a witness in a second accident.
Michael Moore: and similarly person . Here was a witness in the first accident, but was a passenger
Michael Moore: in a second accident.
Michael Moore: And then, for example, you might find that, you know, they're getting legal representation from the same attorney, right? Or maybe they're going to the same physical therapy provider.
Michael Moore: And so this ability to basically understand, hey, this is a really unusual situation that there's this individual that was involved in so many different claims, but seemingly, you know, performing different roles in those claims. So graphs can very, very quickly identify these kinds of patterns, because you can just write a query and say, Show me, you know, show me all the people who have had who have participated in accidents and across multiple roles, say, within the last years.
Michael Moore: and and you know and so
Michael Moore: you'll you know, you'll get an act answer in just a second or from the Neo. For J. Graph. Let's look at another claims fraud, example. So this one is a little bit more detailed.
Michael Moore: But this shows you really the kind of inference that we can do. So. Here, let's assume that you have
Michael Moore: a similar structure to this graph. But we've actually done some other things we've we've actually exploded out. Say, for example, the driver's license ids and the phone numbers and the addresses of all of the participants. And then we've joined the data.
Michael Moore: Using those common elements. And out of that we might get a graph that looks like this. And so here you can see we have a set of cars. We have a set of individuals. And we have a couple of accidents.
Michael Moore: Now, what's interesting right off the bat is, you can see that person, one and person actually share the same driver's license. Id.
Michael Moore: even though they're located at different addresses. You can see that right up at the top there. And they also share a phone number. So
Michael Moore: go to next slide, Scott.
Michael Moore: So one of the things that we can do in a graph is when we see an interesting and anomalous pattern is we can actually set a whole new relationship. And so that relationship might be this shared Ids relationship, and it calls into question which one of these individuals is actually
Michael Moore: has a true identity, and which one is pretending to be somebody, that they're not.
Michael Moore: Similarly, we we can run some analytics on this graph. If you go to the next slide
Michael Moore: and we can, we can deduce that because there's multiple accidents here.
Michael Moore: That
Michael Moore: that.
Michael Moore: And and we, you know, we see, for example, the fact that this person with a sketchy identity was involved in an accident, one on the on the left side.
Michael Moore: and then was also and involved in an accident number over on the right side of the slack flag there.
Michael Moore: That also calls into question those accidents.
Michael Moore: And then, similarly.
Michael Moore: we might see additional linkages with other individuals who are present at one or the other accident. and we'll call into question. Say, accidents, you know accident number
Michael Moore: down at the bottom.
Michael Moore: and this is an example of the kinds of inferences that you can do, I think if you go to the next slide
Michael Moore: right? And so then you. And then the other interesting thing is is that
Michael Moore: while we identified person one and synthetic person, number on the basis of their Id.
Michael Moore: The linkages and the involvements in these other accidents are also revealing to us
Michael Moore: person number , and Person number , who also are suspicious participants in this group of accidents.
Michael Moore: and this would be an actual indication of a fraud ring. And so now we know that you know, person one for
Michael Moore: and
Michael Moore: are potentially colluding
Michael Moore: false claims. Let's go to the next slide.
Michael Moore: And so it's exactly this kind of logic. That we've got a you know. We have a number of customers that are implementing this at scale, and I'll tell you an interesting story. One of our customers. We did a Poc for them. We turned the graph on.
Michael Moore: and within just a couple hours of the graph going into production, they immediately identify auto claims, rings
Michael Moore: and and if one of them was a garbage collection company and the other one was a landscaping company, and just within a couple of hours the valot, and the and they estimated that those rings were each costing them something like $, a year.
Michael Moore: And so right there basically paid for their new overj license. So pretty cool. And so some other benefits is, you know, when you have this kind of analytical capability, this ability to actually look at how things are connected. This really does help cut down on things like false positives that are being
Michael Moore: that are being generated by more simple rules-based systems. It allows your special investigations units to to operate more effectively.
Michael Moore: And you'll end up with a significant drop in fraud payout, and and you'll see the Roi around that. Let's go to the next slide.
Michael Moore: Zurich, insurance uses us as well.
Michael Moore: and
Michael Moore: and without going into the details and the specific details of their implementation, I've got a few things that we can share with you on that. So they claim that they've they've saved themselves over , h of investigative investigator hours, and these are some of the some of the statements that they're making. You know about this implementation, you know. Previously they had a lot of data that didn't have a lot of con context.
Michael Moore: but their ability to look at things like bank accounts and addresses and customer data together was seen as a major benefit
Michael Moore: because they can link them holistically in the graph.
Michael Moore: And they can also update that in real time and see the shifts in the graph
Michael Moore: which helps them with their rep with their reconciliation. So the next slide.
Michael Moore: And so
Michael Moore: you know, they're able to basically bring together, you know, the same kinds of information that we've been showing, and they're also enriching this interestingly with data from external sources, like national databases, blacklists, and other economic data like credit scores to provide additional context around the individuals in their databases.
Michael Moore: And so, you know, they love the ability to rapidly identify issues. They can see the context and they can drill down to a specific claim. See what else it's linked to. They can compare it to past behavior, get a full understanding of everybody that's involved in the claim.
Michael Moore: And then a
Michael Moore: their investigators like this solution so much that our stakeholder there shared that, you know. If we were to
Michael Moore: take it away, there would be a huge outcry. So anyway, the point is is that this kind of capability really is helps. You find, you know, those needles in the haystack do it at scale, and be able to really improve the productivity. Of your investigators as well as actually manage down your false positives and your payoffs.
Michael Moore: So the next slide and I think we are done.
Scott Heath: Super. Well, Michael's thank you so much for that. And and I think what what we're hoping the audience has seen today is a couple of different things, right? Which is
Scott Heath: this really can be
Scott Heath: fast. It can be incredibly accurate, and it can really move the needle. What you've seen today is is like literally one of thousands of kinds of use cases in a couple of different ways and in in kind of where we're going. Here is that with expiro and neo for J. We can do these kinds of things at a different level. We can make them fast. Obviously, neo can scale the performance.
the machine learning those kinds of aspects that bring that to it. And then, hopefully, what you saw today was expiro and the ability to make it easy
Scott Heath: whether it's a chat capability it's bringing up for analysts or other kinds of sort of more technical side folks or
Scott Heath: claim adjusters and very simplistic kinds of customer focus team members. It doesn't have to be overwhelming. And it really should be easy and fast to sort of bring all that together. And then here at the end, look us up online. Neo, obviously it has an enormous
Scott Heath: sort of self-learning capabilities that are out there. There's code. If you're interested, lots of great things that are going on out there, so certainly go check out all these links, and there's even more online with Neo.
Scott Heath: And then for expiry we're in most of the major clouds
Scott Heath: we can install, either on premise or the cloud, but certainly look us up as well, and as we're getting to our next slide here, we'll open up the floor for questions. If you have any questions, please enter them in the box.
And we will be sending out the recording of the video
Scott Heath: at the end of this week, and we'll be able to do that from there. So we'll take a minute here and basically identify questions. If you wanna send those in? Michael? Looks like one came in for you. Which is
Scott Heath: how much data do I need to run the algorithms on to touch, to start to get accuracy information.
Michael Moore: Well, you could run the algorithms on any amount of data.
Michael Moore: the
Michael Moore: and the
Michael Moore: what the algorithms do is essentially they teach you
Michael Moore: insights around the structure of the graph. And so you know, maybe you want to start. You know, you could start off with a small number
Michael Moore: of of data points as a Poc.
Michael Moore: and then and then you could scale out to, you know, millions or billions of records. The the algorithms are always looking at the structure of the graph
Michael Moore: relative to sort of its global average structure. And so it looks that they do things like count relationships, or they look at linkages or isolated communities.
Michael Moore: And so they're usually pretty good at detecting that even at, say, medium-sized volumes of data.
Scott Heath: Okay, looks like you got another one here. It says, I'm a relatively new.
Scott Heath: a machine learning team member. How hard or easy is it to pick up neo for Jay
Michael Moore: Neil, for J is very easy to learn, so any any competent it team that has sequel under the belt can learn. Probably. I don't know % of of our cipher query language. In a weekend, and you know, maybe a couple of weeks to get the rest of the
Michael Moore: of. You know the advanced pieces. Under their belts. It's a very straightforward very intuitive it's easy to build these graphs.
Michael Moore: and we have a ton of
Michael Moore: course, work that's available for free on Graph Academy. For anybody that's interested
Scott Heath: Super, the next one looks like it's probably for me. How long would it take to run a prototype for what we saw today? Well, that always depends on your data.
Scott Heath: If you're using a guide wire or one of the more common data systems or your your something where your your data is known. It can be weeks. We can do rapid prototypes and literally or days. If we actually wanna do it behind your firewall. There might be a a bit more sort of setup if you will, but it's typically again, to weeks, perhaps
Michael Moore: depending on how specific you want to get, but it's very easy to do and also just let me chime in here. So one of the one of the big advantages with working with an experienced solution. Provider like expiro
Michael Moore: is the acceleration and and speed to business value. And in, you know, today's budgetary.
Michael Moore: you know, climates. It's really important to quickly show your business stakeholders the power of what can be achieved. And so.
Michael Moore: and and you know, in our organization, like Expiro, with their very well thought thought out solutions. Can can really accelerate that whole process.
Scott Heath: Yeah, it's a great point, Michael, looks like there's one here for you. It looks like, do I need to set up any neo infrastructure to go through the training and graph academy for you, Michael.
Michael Moore: you're welcome to but they those are sandboxed experiences, and so they they have. they have sandboxes associated with them.
Michael Moore: you can very quickly install neo for J. On your desktop we have a desktop ide. And and so you can create a local environment, or even more easily. You can just sign up for our aura graph database as a service and get a free account
Michael Moore: right from the neofroj.com homepage
Scott Heath: right? And it looks like the last one here is probably for me. It's talking about guide wire. So there's ways to do it. Guide wire. Well, it depends on your license. So if you do own guide wire today, there is an Api
Scott Heath: that you can use. That we would simply, on your behalf would plug into depending on where your source data is. We can also connect natively to the tables.
Scott Heath: Or if your environment is a bit more sophisticated. And you're using a Kafka or one of the the sort of message buses we can plug in there, or sort of the last, most tried and true version is an excel spreadsheet that gets exported. And typically we we pick up wherever your system is and what your licenses. But we can do any of those, and then
Scott Heath: your last one is, will you be able to effortly ingest the data? Or is it more of the gooey side of things.
Scott Heath: That is is again, we can do either or both, so we can. The way the front end is designed. It's a rest protocol where we can leave the data inside of a guide wire or we can actually do a full integration. It's really about your organization and and sort of how far you want to go. But the answer is one of many ways.
Scott Heath: Okay. well, that looks like it. For questions, Lisa. I'll pass the baton back to you to close this out.
Laura Smith: Yeah, I just wanted to thank everyone for attending. And thank you, Michael, for joining us today. We will be sending out the recording as I mentioned, via email. If we have your email list. Otherwise we'll also be posting it to the expiry Linkedin Page, where you can also access it. So. And if there's any other questions,
Laura Smith: this is your time. Otherwise thank you for attending.
Laura Smith: Thanks, everybody. Thank you. Everyone.