Scott Heath: Everybody starting to file in while we're having folks join here very excited for today's presentation Mike Booth and I, who are presenting are super excited about what we have just a couple of of housekeeping elements. Today we are indeed recording the session.
Scott Heath: We will be sending out an email afterwards. With this recorded session, if you have any questions during the session, there's a chat box. If you could enter that also at the end. Mike and I will be covering any sort of questions that you might have. And we're super excited that you're attending today.
Scott Heath: Without further ado. Looks like we're right on schedule. We'll go ahead and kick off.
Scott Heath: My name is Scott heat. I'm part of the expiro fraud graph and analytics team. I wear a lot of hats. But today we're going to be talking a little bit about what makes
Scott Heath: Llm spatial analytics, graph technology just in so incredibly powerful. And then Mike and I are going to walk through a couple of different scenarios, and then we're going to try and show as much as we can. So with that, Mike, you want to introduce yourself. Thanks everybody for joining today. My name is Mike Booth.
Mike Booth: I've been in the analytics space for about 25 years. Most recently the last 3 years with Tiger Graph having moved over to Connecticut about 5 months ago, really anxious to to get started and look forward to to speaking to all of you. Thanks, Scott.
Scott Heath: sure. So one of the first things that I'll do is, I'm gonna talk a little bit about why and what is sort of going on in the back drop. What? Why is all this so important? And then I'm gonna talk a little bit. About the secret sauce. What is it that expiro in Connecticut are bringing to the table? Kind of what does it fit at a logical level. And then, finally, Mike is is actually gonna walk us through sort of the neat
Scott Heath: of why, the Connecticut capability built into expiros. Connective platform is really such a home home run. So without further ado, we got a couple of different types of of things going on in the market. Right? What we're seeing is
Scott Heath: no longer, is it? Aml? Is it purely one sided? People are bad guys are attacking us from one. Vector, what we're finding is they are actually hopping over the silos so they may come in and cyber. They may start to, then open a brokerage account. They're starting to do trade so we need to do trade surveillance. And then they're laundering the money out through Aml, and they're doing other kinds of things. And so the numbers are staggering. I don't think I think we need to
Scott Heath: anybody. But the point of this is the sophistication of this sort of different element or attack vector is is really what's doing is now, when we look at this, whether it's brought in Aml, they're attacking us in different ways. And now, what we're seeing is all the way down to the point of sale, and this false positives is becoming overwhelming. Most everybody in some of the surveys, and
Scott Heath: we're both members at a cams for anti money. Laundering is acfe that we'll see here in a minute. But what we're seeing then is
Scott Heath: this collaboration. Really, the only way to approach that now is, perhaps with a technology to be able to do that. And our friends in the government and and and sort of the people that are governing these different elements depending on where you are, are now scrutinizing this, you have to be faster. You have to be more proactive. And the fines are actually ratcheting up. They're getting enormous, right? And some of our customers that we've worked with, you know, 600 million to a billion. Those are pretty staggering.
Scott Heath: But the one sort of item that you will get from both of these and a quick plug here, please come see Mike and I at the a camps in Las Vegas. if you're interested we'll be giving Demos. We also have a swag bag so sorry. That was like free marketing plug there. But the number one thing that if you look at the side by side is folks that are working on fraud and folks that are working on anti money laundering. It is this introduction of new
Scott Heath: technology. But, more importantly, it's collaboration. What's going on in a credit card. First, party trade is also manifesting over an anti money laundering. So it's taking this technology capability and integrating it. That is primarily the focus. In addition to giving the human the mirror mortal on the business side of an investigator, they have to be able to wield this new thing called Llm or AI, or machine learning
Scott Heath: or graph, or this visualization, all has to be wielded by mirror mortals. And that is really what we're seeing here is that it doesn't really matter what you're doing.
Scott Heath: Everybody needs to be talking. But we have to talk the same language. And so in technology worlds, we call that humans in the loop in the old school it's called the human is many times smarter than the system. But the systems now have this enormous capability. And that's why we're here today. Because we want to see what this technology can do for us. Now.
Scott Heath: when we look at what we're gonna do today. What is expiry connected for financial crimes? Now, obviously, it's a technology layer, and it is using the Connectica platform. But over on the left, regardless of which group you're in. Whether you're in the Fi, you're working on cyber, you're doing trade sanctions or compliance whatever that is. You need to be here in the middle. You need to see it. You need to be able to detect it, and you need to be able to go about your business.
Scott Heath: Without upsetting what you're doing. The other element over on the right, then, which is where Connectica helps us, is trying to find these patterns, taking massive sets of data over 10 years and looking for not just the needle in the haystack looking for which haystack has multiple needles in it. It's a very different style, right? We also have this ability now to do scoring and prediction. And now we have large learning
Scott Heath: models that are in the mix. Now that we're gonna talk about today. But all of that culminates here in the center, which is, if I can't see it as an investigator, and it doesn't make sense to me explainable. AI, and I can't see it on a map, or I can't, you know, sort of interact with it. Then it really has no value. And yes, it's complicated and it's cool. But I haven't solved the problem. So that's really kind of why we're here today
Scott Heath: now lay down in a different way. This kind of a model
Scott Heath: can also have adjacent use cases. Now, Experieno works on many different kinds of use cases with Connecticut, and you see them at the top
Scott Heath: where this really starts to be. The meet of where we're going here is we have to look at data from Kyc. We have to look at the positive that you may not be fraudulent. But I need to look at everything down here in this world, and that's why this problem can become so large, and then we need to share with our friends in audit and compliance and regulations. So not only do we have to do all these things right, but we also have to play with other use cases
Scott Heath: now
Scott Heath: where we started. This journey was a long time ago, right way back in the nineties, we started DC SQL. Tools. Well, that's great. A lot of the systems that are out there today. Still have a massive investment and capability in sequel. Well, we bring you good news today. Right? We're not talking about throwing it out. How do you use it? How do you uplift it. How do you modernize it?
Scott Heath: Then, in the 2 thousands, we started to see this bi tool revolution. I can drive everything for my tableau dashboard. Well, what happened there was is now. I started creating sequel on top of sequel on top of sequel, and it wasn't very flexible, and it wasn't very fast.
Scott Heath: Then, in the 2,010 era, we start to see graph databases. Now, this was a real insight, and we've been working on these kinds of layers for quite a while, and what we see then is you can. We'll see that here in a minute, a massive up, uptick and capability.
Scott Heath: But now, when we sort of tie together geospatial time. Algorithms, machine learning. Llm, that's where we are today. And it's a bit overwhelming. But what we bring you again, is this sort of capability to increase your accuracy and lower your phone positive by a radical nature.
Scott Heath: And this isn't just us talking right? We have basically, this is from our customers.
Scott Heath: and some use cases where simply doing machine learning alone gives you a modest lift.
Scott Heath: Some of it, maybe, again, your mileage may vary, you might get more than 5% in there. Some cases you can get a big number. But where the money really starts to sort of impact. This now is when I can get into using graph or algorithmic machine learning, both supervised and unsupervised. In addition to pure machine learning. We can start to see a bigger boost. Then, where we get our big pop
Scott Heath: is now visualizing, letting that human into the loop and giving them a feedback to these things via G. Ge Geo. Time, space, etc. And that's where we're starting to see this massive lift. And for a quick case study
Scott Heath: in this. What we found is an existing sequel based system. And it was a large name that everybody probably knows and uses, and what we did was, is, we bolted on
Scott Heath: to that system, and we were able to do that in a roughly an 8 week time scenario, and from that proof of value we were able to identify a massive reduction in false positives, and what we see over there is, there was no additional risk. In fact, we were able to squash very unproductive linkages in the way we did that. Now, we're not knocking sequel.
Scott Heath: What we're saying is by using what we're talking about today, we can bolt onto what you already have, and Mike will talk a little bit more about sort of how that works together. Now
Scott Heath: that's all. Great Scott! That's pretty cool. We're excited. Now let's put our our investigator hat on right. So, regardless of where you are in that stack, whether you're a technical machine learning person in an Fi, you or your first line investigator, or your investigation matter. You're completely overwhelmed. Right your boss at the top there saying, you've gotta do more with less.
Scott Heath: You've gotta be better with your case processing. And obviously you're swarmed with false positives. You've gotta lower it right. And we want you to be better, because what we don't want to do, then next time we get in front of the government, right is trying to explain things. Where we're not very good. Now, on top of that. Now you're catching all the things from the left. So everybody seen the hype trained about artificial intelligence. It's gonna save us all unsupervised
Scott Heath: learning graph algorithms, etc. And what you find then, is that is very overwhelming to the investigator. Right? This this poor woman here, you see, with the the bullhorn shouting in the woman's ear, right! Who's the loudest? Well, they're all important.
Scott Heath: and then you go back to what they actually have to do. They have to go find evidence. They have to go find where these things happen. They have to find things in time. They have to find hidden connections. They have to find out where things were geographically. And now we have IoT devices in the mix and and those kinds of things. So what we start to see now is this sort of overload right? We need to do it all. It's all important. But how do we do that right. And that's kind of where we're going next.
Scott Heath: Now, part of this is what we see in one of our examples in in one of our customers was. They were syn, thinking that the cyber group, the Aml group, and the credit card group over here on the left, they all thought things were fine, and what they found was well, each of them were running along with what they thought was an acceptable fraud, but when they compared notes across them. This is where the silos got them into trouble.
Scott Heath: What happened then was is it was actually a 20 million dollar fraud ring that was acting like they had come in. They had done account takeover. They were acting normal. They got loans. They were doing normal things. They started doing brokerage, and by the time they were done they had hit all the different business units. And the number one thing we mentioned
Scott Heath: was that collaboration. Well, one of the technology capabilities that we'll see today has that ability to be able to share that? And how are we able to segment it by pods or by business unit? And then at the same time run those analytics over the top. And the answer is, what we're gonna share here in a minute was, how did we do that. But again, all of this is part of that investigators needs. Now, what we're offering now
Scott Heath: is the ability to number one streamline right? So if you screen it and adjust the data.
Scott Heath: you can segment it. And now you can start to run either alerts that you already have today in different silos, but then you could bring them back together with machine learning, and that is where we can then start to parse this out to different business units or different users. And so what we're showing now is a similar process.
Scott Heath: And now that investigator is using the power of machine learning is using the power of an Ll is using the power of geography. And some of these time series elements. But they don't need to care. And that's what we're going to show you here in a minute as well. So when you put this together
Scott Heath: right, this is why you need all of those. And the business user investigator doesn't really care. They just know that I need to search. It needs to be fast. I need to be able to run sequel style queries or olap, because I need to see what correlations are to figure out if it's even fraud. Yet
Scott Heath: right? The other thing now is with our streaming device data and and sort of the what that looks like. I've got to be able to look at every point of sale, every brokerage transaction, everything that goes on in real time, because I don't know if I'm getting scammed right now, and I need to make stops to it or or not right did I overlay that with time? What went on in time? Was this really
Scott Heath: compliance issue? Or was this true fraud or where is it in time and space? And those 2 together start to? Now give you a very different view, materialize view as more of a technical term. But that allows me to do simple math over billions of rows of data, etc., and then finally, I get to how do I feed my machine learning, and then my generative or my my query. Now.
Scott Heath: now in English, what that means is I need to ask simple questions of the data. I need it to be fast. I need it to be real time, and then I need to overlay it with time and space. And it's just that simple. However, that simple business request from an investigator or an Fi you is extremely complicated, which is why obviously crimes from Connected and Connecticut is so much different. Now.
Scott Heath: now we're gonna get into a little bit of the secret sauce. So how is it we start to do those things? What are the sort of tips and tricks that we can do. Well when we start to look at the data in that first chain is looking for screening, and those kinds of things. I can't wait till later to figure out my screening. If I find a similarity in the ingest of my data at scale now, I can start to find and tag people very early in the process that could be
Scott Heath: connected or central to or similar to which one of these algorithms are doing 2 known people, and I can do that much earlier now in the process. Which means that when that alert fires it's more productive, right? And that's where we start to see the secret sauce. Where in Step Number one, I can find the pattern. What did I find? I found that Scott
Scott Heath: and Mike Booth may be collaborating, but I don't know. Then I can run another pattern and say, Well, guess what they really are doing that. And I'm starting to overlay this higher order of math.
Scott Heath: What I then can do is create something called a community. But now, where the real money comes in, as far as saving the money and detecting is using time
Scott Heath: and then using space down here for that, because now what I can do is very powerful is tie those things together. Now I can increase my accuracy and lower my false positive by this recipe, if you will, the secret sauce
Scott Heath: for what's going on behind the scenes, and then finally, we'll see a vignette later. Here, where we show how the human can interact as an investigator where all of this really cool stuff I just talked about can be handled behind the scenes, but it feels like. And they are similar with the systems that they use today. And that is really also a very subtle point, but very important. The investigator doesn't need to know all this data. They just need to know that they trust it, and that these capabilities are there.
Scott Heath: So
Scott Heath: now what we'll do is we'll talk about a quick demo, and then I'll hand the microphone over to my counterpart, Mike.
Scott Heath: So in this demonstration
Scott Heath: there are different users. And we talked about them. So if you're on the phone today and you're a technology person. You're sort of the group at the bottom. And what kinds of things can we do? Well, this is where the data models happen. This is where those algorithms are run. This is where the queries and that materialize view the geography and the time series. This is where all of those things happen.
Scott Heath: Unfortunately, we're not gonna talk too much more today about that today is about our investigators on our business team, but certainly reach out to us. And we can talk about that. Now. Mike's gonna touch on a little bit of that.
Scott Heath: What we wanna talk about now is, if I'm a more technical investigator down here at the bottom. How far can I go? What else can I do? I can see an awful lot. Inside of the tool. And I can start to go see where those things are, and we'll take a walk with with what that looks like.
Scott Heath: Also, I have the ability to do Kyc. And screening right? So if there are things that I wanna continue to look for, or block or monitor, I have the ability now to drag and drop my way. Using the power of Connecticut inside of this business user interface to build alerts. How do I build an interact elements? Right? Then, if I'm a line of business, maybe what I want to do is I may have a larger group of either investigators
Scott Heath: or the red team or the Blue Team or I may be sort of interacting with customer 360 persons, right? And so I also can have my set of screens.
Scott Heath: or interfaces that allow me to do that. And then, finally, if I'm the head of risk or head of compliance, I can start to look at, or there different kinds of trends and and elements that that go on there. So again, what we're looking for, there is a sort of full interactive capability because everybody needs to interact with what the time series, the Lom and some of the other components.
Scott Heath: So when we see our demonstration, what I'm gonna do, Dan is. I'm gonna walk you through a couple of the different modules. Right? I'm gonna show how we're gonna build a set of alerts. We're gonna look at them and introspect them. And then we're actually gonna go investigate them.
Scott Heath: Then what I'm gonna do is once I'm in case management. Now, what I'm gonna do is I'm gonna overlay that with the Llm. Ask it a question in English right? Sort of ask it a question that an investigator would ask, and then let the technology behind the scenes
Scott Heath: go, make the connection go, do those really cool things? And and again, Michael is gonna talk to us a little bit about that. So inside of the case manager. Now, I can do typical things. Now, if you have an existing case manager, we bolt onto those. So the typical ones that we work with or activize work, mantis and hummingbird. And again, those are fairly common. And Mike's got a slide on that as well.
Scott Heath: Okay. So without further ado, we'll switch over to our demonstration. While we're doing that.
Scott Heath: Let me see, I'll pull that up over here. Let's talk a little bit about sort of what's going on. Inside of this. Now, there are 2 different views that I'm gonna show you here. Right? So the first view here, which is the title of our Webinar is that I am an investigator. But I am typically a level 2 investigator or a level 3 investigator. Right? So I'm not quite that sort of expert. I'm kind of new. But what I'd like to do
Scott Heath: so again typical sort of front end. Here everything looks kind of similar. And again, if you have an existing case manager, it might be very similar. But what I see now is, I see this alert right? I see these alerts that are coming to me here. And then the one that I'm I'm concerned about is something that might be going on here and serious. So this might be same castles is what this is actually based upon. But in our
Scott Heath: suspect here. What I start to see now is, I see what's going on. And I have a view, but as I dig in what I can now start to see as well. I have found that first ring of connection. I have found a party. That is basically problematic. I see my first introduction to time or time series on the bottom.
Scott Heath: So now what I can do is I can start to go back and look at atomic connections or other elements. But in this case what I see is the the. The system has found that there was a connection to Golden C, that was considered to be risky. But what I'd really like to do now is I want to interact with this in my connection to an Lm, I am again sort of a basic or a novice user.
Scott Heath: What I wanna do is I have some canned queries here. That are pre-built for me, but we call it jetpack, and then and other people call it co-pilot. Or, again, whatever you wanna call it, a lot of people have different names. But the point of this is, I am a mirror. What I'm looking for now is, why is this shipping container connected? And it's been flagged for me here, and what I see then is, well, I wanna grab this query, and that's
Scott Heath: seems good to me, which is, show me what those connections are for this person. And then what are within 3 degrees. So each of these circles and lines is effectively a a degree. But as I click on that, what it found for me now in a very fast manner. Right is I connected now those bad actors, bad transactions, and what they may be, and the risk that is associated
Scott Heath: with them in a specific timeframe. So I've used time series. And now what I've used is this nodal or graph connection. Now, what I see is, there's more to it. I see that this golden C over here is actually connected, and you can see the circles here for what we call the blast radius within 3 degrees. Now that is now shown me that. Well, there's something else going on here, right? What I wanna do now is, let's go look at this.
Scott Heath: Well, sure enough, this shipping container had a leak. Well, this is in effect adverse media. And so what I may want to do now is actually go out, and you can see over here. That I one of my pre-safe queries over here is I have adverse media.
Scott Heath: Well, it's not just adverse media that I'm looking for, I want to see what else is going on with it. And so now, when I find that
Scott Heath: I have found now a summary, and within this summary. What I want to do then is add that to my case. So now my Llm. Was done is 2 different things. It's run a query for me which was extremely helpful. I didn't know what those 3 degrees of connections were. Did the work warming behind the scenes. The second thing, then, is, if I am going to go investigate this. And I have this adverse media, I'm gonna use these bullet points now that I can either
Scott Heath: plug in at my metadata level. Inside of my case, or I can simply use it as a synopsis, so I can use that in 2 or 3 different ways. And so now, what you can see is just in 2 simple ways. I've been able to sort of ask the data and the questions right? So now that I have that, what I'd also like to do is now what I want to use. I wanna use my Geo. I wanna find out where that was. My third sort of canned query. Here
Scott Heath: is, show me where that is in respect to what that is on the golden sea. And now what I can see is, I can see the full sort of fraud chain here. What I see then was, and this is again partied on the Sandcastle beta. But this is a real person, by the way. In that investigation, and what it was then is that CEO
Scott Heath: was working through a shell company and then was trading. And now, by using the Connectica map, and the Geo capability, I'm able to look at the entire event. So now, what I have done is, I have created a much stronger case. I have used sort of that simplicity to do something that was extremely complicated, and that was our first vignette.
Scott Heath: And so what I wanna do now is I wanna back up and say, Okay, Scott, that was pretty cool. But how did we get? How did we get there? How did you set that up? And so what I wanna do now is we'll talk about. So what we're gonna do here is we're now going to effectively. Look at, how do I build that that alert? Now? I can bring data from different or disparate places. I can bring it from Optimize Oracle or
Scott Heath: Moody's, or wherever that Equifax look up may be here. And what I'm doing that is, I'm going to build that 3 degree of separation for that sanct sanctioned entity.
Scott Heath: What I can do now is every piece of data that that we saw there, the ship, the party, whatever your data set may be. I am now effectively building this logic tree that includes Boolean logic. Geo. Logic temporal for time. And then also these higher level solvers. And so what I'm doing now is I'm basically building these.
Scott Heath: In this case, I built that 3 degree of separation. And I can walk through this. So this now is allowing this business user again to use the simplicity of the graph database and some of the other elements. Now, what I want to do is I actually want to go look at my dashboard. And inside of this dashboard I can start to see other kinds of elements. I can see smaller kinds of
Scott Heath: elements here I can see what what's going on, etc. So within this, now, what I want to do is I can zoom in
Scott Heath: up here, right? So I can get that up here. But that maximize
Scott Heath: I can start to move through this. So now, what I wanna do is I want to go look at that same view. I want to see path finding options. So now I'm using this graph data capability or solver behind scenes. Now, I can see where things are going again. I see my time time series. There are 2 kinds of time series in how we do this, we're using temporal time or atomic time. So I can see sort of groupings or clusters, or whatever
Scott Heath: called the subgraph in this case, and I can see down here that I have that time selected, and then what I can do is I can start to see things in more of an atomic or overall view. So again, time has 2 different capabilities.
Scott Heath: The other thing that I may want to do is I can dig in. So now, whether I'm using an Llm. Up here to do those summaries right, I can. I can produce those, or now I can dig in, and I can see more specifically where these events are going.
Scott Heath: So what I've shown now is how the Llm. How the time, space. How the Geo. How all those different elements all tied together? Because, remember, the importance here is as effectively somebody who's an investigator. I want to see what those different elements are in in one level. So again, this was meant to sort of get the juices flowing if you will. And now what I'm gonna do is I'm gonna hand that over to
Mr. Pike.
Scott Heath: And why don't you tell us a little bit about kind of what's going on behind the scenes.
Mike Booth: Yeah, absolutely thanks, Scott. And and again, thanks everybody. If you could go to the next slide, I'll give you a a brief background on Connectica. So ironically, you know, in 2,009 the founders of Connecticut started with a project within the Department of Defense to merge satellite imagery with geospatial data.
Mike Booth: The challenge is the that geospatial data was of of mountains and and path. You know, sort passages between Afghanistan and Pakistan. So to the ability to take the surveillance data coming from the satellite
Mike Booth: and map that to a map where there were no landmarks there were no marked streets was a real challenge. So what Connecticut was able to do, and what our founders were able to do is pull in that satellite imagery and determine the path of the terrorists, the speed at which they moved, where they stopped, what location they were heading in and allowed the military to. Then, you know, plan a coordinated attack based on their movements in an area that was almost impossible to map.
Mike Booth: And and that was the start of of Connecticut.
Mike Booth: If you fast forward. 14 years. We were the technology behind the most recent
Mike Booth: event, where the the balloon, coming out of the weather balloon coming out of China, crossed into the Us. Border and and crossed across the United States, we were able to detect altitude, speed, wind, direction, and location. All of the pertinent data that the Department of Defense needed to determine where and when would be a good location. To shoot that balloon down Connectica was the capability
Mike Booth: on behalf of the Faa and Department of Defense to provide that information, a combination of time series input and geospatial data. Now that's great. It's not all that relevant to what a lot of commercial businesses are looking, you know, looking to do with this technology. And what we've done in the commercial sector is quite compelling as well and in terms of Ford, we're helping them determine. For autonomous driving, their lightning pickup truck!
Mike Booth: What roads are are best? What are the IoT? What? What's the IR IoT data feeding back to Ford, based on a particular location on a map. Right? How is all of that information impacting? How Ford releases autonomous driving for the Ford lightning. In the case of Liberty Mutual, we're working with them to detect when a hurricane hits
Mike Booth: Florida as an example. It's a category. 5, we're looking at path and we're overlaying where the what properties are gonna be impacted so that we're doing a predictive claims approach to understanding where the impact of of of a claim is gonna, be based on the path of the storm, in the, in the insured properties that are in that path, and the third one is the Us. Postal service
Mike Booth: understanding across all of the postal carriers. What is their what is the logistics, and what is their route. What is their route planning when there's a road outage when there's a road shutdown, or when
Mike Booth: their changes to addresses all of that information is optimized at a mail carrier level on a day-to-day basis on behalf of the Us. Postal service. So again, different examples of using time series and geospatial data well beyond how we get started next slide.
Mike Booth: So let's tie all this in. If you look at oracle and nice optimize, you know they are to their credit sort of.
Mike Booth: And you know, count context and others in there as well. The de facto rules, engines that have been used for fraud investigation. Right? So they're they're predictive rules based. They are batch based in terms of loading data. And they they do a they do a good job from a predictive rules engine in determining where fraud exists. And, as Scott mentioned, we complement. We do not, in in most cases, certainly aren't interested in ripping and replacing that technology, but rather augmenting it
Mike Booth: with the use of Connectica to then overlay graph and and real time. Ingest and Geospatial Time Series based capabilities to allow for a more detailed view well beyond what a predictive rule can provide. So taking those
Mike Booth: potential cases that need to be investigated and digging into them very quickly to determine. Is it a false positive? Or is this a case that we need further investigation, and to do it all that those 2 columns on the right or the middle on the right. You're looking at technologies like Etl from informatica and geospatial with precisely, and sensing with IoT data to cobble all
Mike Booth: of that together would be a herculean task to try and get done where Connectica offers this capability and bolts on to Oracle, Optimize and Quantx and others in pulling that in. And I'll show you how we do that from an architecture standpoint as we go forward next slide, please.
Mike Booth: So we're considered the speed layer right? So being able to take real time. Ingest is the is the heart and soul of what we've done in the hardened part of our our database, and and we've since you know, since creating the capability, overlaying with geospatial in real time we've added graph
Mike Booth: Llm and machine learning. So being able to take all of this data, all of this streaming data, this bank transaction data and overlaying it with a criminal database and address, information work, history, credit score sanctions list using graph capability, using a large language model capability and machine learning, coupled with the case management capabilities from experiment. This is how we pulled it together, and it really backs up what Scott has just showed in the demonstration
Mike Booth: providing for the visualization, the materialized views and such in better understanding. And I'll talk a bit about why geospatial
Mike Booth: is is such a key element to to fraud, investigation, and what we see, and why there's strong interest in that next slide, please.
Mike Booth: So when you take a look at best in class time series and spatial analytics, think about this for a second. You've identified a potential a fraudster, and you've looked at the transaction information. And now you've got an IP address, and perhaps you've got a device. Id. You're looking at the IP address, and you're looking at an address and the addresses in in an apartment complex.
Mike Booth: And you don't know whether this person how they travel? Do they take a bike? Do they walk? Do they take the subway? Do they drive a car? How do they get from Point A to Point B. Do they go from home? Do they go to the coffee shop where they perform another fraudulent transaction? Did they go into work where they do a third fraudulent transaction, this time on a cell phone versus a laptop.
Mike Booth: All of this information can be tracked and managed in real time. As these transactions occur within Connecticut. So being able to tie all of this location data device, Id IP address and look at it on a map, so that we have a very good idea or a pattern of what this individual's doing
Mike Booth: is really compelling couple, that with the fact that this person now lives in an apartment complex. He's part of a fraud ring of other individuals using IP addresses that are also located at this apartment complex. You start to paint a very clear picture of what's going on within the fraud ring, and in tying together, otherwise tying together transactions that otherwise wouldn't be completely obvious next slide, please.
Mike Booth: So let's talk a little bit about the large language model and the Gpt capabilities. And Scott showed some of that. But as as a fraud investigator does does their due diligence. They're going to start to come to a conclusion based on all of this evidence. And all of this information, using a graph model to determine that the person that you're investigating has affiliations with individuals in a criminal database they've shared a they've shared a former address and a work history with that individual.
Mike Booth: And there's from a geospatial standpoint, you know that they've they've stayed in the same area in in close proximity to others that are part of that criminal database. Well, think about taking all of that information now that is being that is being added to that particular investigation and asking questions of it with the Gpt capability within Connecticut. And now that that fraud investigator
Mike Booth: has further validation, or opens up additional questions to be asked about that particular case all done in real time. It's conversational query. It's taking all of this element of data, tying it back to the individual on in question, in having a quick synopsis of
Mike Booth: of the probability of of guilt or innocence. In in real time. And it's done through vector search in converged analytics. We talk about the geospatial, the Graph Time Series, all of which come together to help give a broader view of exactly what's going on with that particular individua.
Mike Booth: next slide, please.
Mike Booth: So you know, all of this is is validated. When you look at IoT, we call it IoT payment data. Transaction data is a is a form of IoT in terms of attributes. And if you look at what Deloitt has come out with.
Mike Booth: They're validating that this space IoT data is going to have geospatial capabilities tied to it. And it's gonna go up from from 10% 2020 to 40% in 2025, and let me let me step outside of the banking industry and use an example where I think we all can can see the validation and relevance.
Mike Booth: A farmer in in Illinois buys a a tractor, a John Deer tractor. They're on a 2,000 acre farm that tractor breaks, breaks down. A signal is sent over to the, to the, to the local dealer to take care of that tractor. The local dealer drives over to the farm has no idea where the tractor is, has no idea where the farmer is with the geospatial attribute tied to the IoT signal. Now, all of a sudden they can say, okay, they're at this
Mike Booth: particular quadrant of the farm. I can go out there and and and help this farmer out and getting it done this just one example. But this is this is the direction of IoT and and and IoT data as it gets tied to geospatial data. It's just a net. Natural evolution
Mike Booth: of IoT data next slide, please.
Mike Booth: So let's take a look at what this means from an architecture standpoint. Again, I'm going to focus on Orkle and nice optimize on the far left that the information that they provide in terms of a predictive model in identifying a transaction that may be deemed fraudulent, that information gets passed within into Connecticut
Mike Booth: couple, that with the third party data, that additional data, that static data that would be brought in by the bank or by the by, the organization around a criminal database. The Sanctions list, address, information, work, history.
Mike Booth: All of that information coupled with the transaction data, starts to round out the view on behalf of the fraud investigators as to what's going on and what is the relevance of this particular transaction towards a potential crime?
Mike Booth: The analytic applications that are contained within connectica. Allow for all of this capability from machine learning to aggregate analytics distributed key lookup. These are things that all go on behind the scenes, so that, as the as the fraud, workbench, and and case management capabilities that Xpero provide.
Mike Booth: allow that fraud investigator to make to to to bring this data together and start to come back and understand exactly what the issue may be, and whether it's it's worthy of further investigation or passing over to to law enforcement. So again, Connecticut sits in the middle as a additional capability that complements what's being done today with the predictive rules engine supporting fraud, investigation, and the last slide I have before passing it back to Scott
Mike Booth: is, you know, from a summary standpoint, the value of time series, and Geo. Geolocation is really key. I think Scott shared an example during the demonstration, understanding not only what's taking place and when it's taking place, but more accurately where it's taking place, where so much fraud is taking place in the Internet is really key. Taking advantage of that, IP address and device Id, and and looking at the cookie crumbs
Mike Booth: that these that these fraud that these fraudsters lay down is really key in in in Connecticut can add that scalability and performance is, is is paramount to who we are and what we do.
Mike Booth: You know the the majority of those customers that I've spoken spoken to the forwards in the Us. Postal service are dealing well in the tens of terabytes of data that that are being analyzed today by Connecticut.
Mike Booth: and from a machine learning and clustering, and then the large language model capability. Those those cutting edge applications on on the analytic analytics side are all incorporated. You couple that with the front end capabilities from experiment, the Ui, the the case management capabilities and it now takes a 30 year law enforcement individual who is a fraud investigator at a major bank
Mike Booth: and puts in their hands very, very powerful capability without having to be a data scientist quite frankly to to get it all done.
Mike Booth: But that's all I have. Thank you very much, for you know, for listening in, and I'll pass it back to Scott.
Scott Heath: Well, hopefully, that was you know, we had a lot of content to cover sort of in a quick synopsis here. That's kind of what we want to do next, and then we'll open up to some questions. So you know.
Scott Heath: we covered a lot of graph. Right? What you see, then, is the business of investigation of all types of financial crimes are very similar from an architectural or a data
Scott Heath: logical view. They were very different as you work your way up. So when I look at the data right at the bottom, I can't just have one type of data, and there is not one type of thing that will rule them all. That just isn't a thing, and as investigators, as as business users, if you will.
Scott Heath: we need all of it at our fingertips. Right? We need old school sleuthing, investigation capabilities. We've got to tie paper receipts and things together. Now, as we get more sophisticated with that, what you start to see, then, is this data connection layer at the bottom? If I do not have that capability, then, as I get up to the top up here to do my job, it gets
Scott Heath: time consuming or problematic. And now, when I start to look at this higher order of math.
Scott Heath: Llm. Helping me through sort of this maze of complexity. And I'm looking at, you know again, whether it's a graph algorithm or a similarity or a luvain clustering algorithm. I don't understand all of that. I'm not a technical person. However, I need to know that that is there. So you need to as a business user, ask for it by name right, being able to use machine learning graph data structures, time series, all those things
Scott Heath: that Mike just mentioned. I didn't need to know exactly how they all work. But as I work up now to the alerts, and then that capability, what expiry has done with connected is taken the stress, the bring damage out of all that complexity out of it. Right? It's flexible, it's modular. We now allow the business user and the technical teams to both work together in that stack layer that we showed earlier. And the takeaway now is.
Scott Heath: we can do very, very complicated things in a very fast way, and more importantly, at a cost structure, whether it's in the cloud
Scott Heath: or it's on premise or connecting to your custom systems that you have today. The point of this is, we are architected for this new generation to be able to bring this technology to bear. The other thing is we have modules, right? We have the ability to break this down even further. So if you're in the market for one of these different tools, or you wanna take it for a test drive. You don't have to turn them all on
Scott Heath: right. And if you want to connect to your existing case management, you know what I use ours. That's fine, right? The point of this is to allow this capability, this technology right being being able to to make it module or make it easy. Is really what, what we're what we're in the business to do
Scott Heath: right. And so this is our kind of our final wrap up slide here is really, it's about the false positives and the accuracy. Right? All of this is all window dressing, if we can't do that. And you saw early on that that some of these numbers are really, really eye popping right. But there's a lot of there's a lot of elbow grease that goes behind that. And and we certainly would love to chat with you about that. The other thing, then, is bolt on right. You don't have to throw away everything, and people that have spent years and years on certain types of technology stacks and folks that have coded and done lots of things. That's the beauty of what this allows you to do now is you now have the ability, both at the technology side and the business side, to go do that.
Scott Heath: Is it a silver bullet? My God, Scott, this thing can drive and make me a sandwich. And the answer is, gosh! I wish right. There's still a lot of investigation work that human beings do. And I think that's the the core premise here is technology can't solve at all. No, unfortunately, AI is not gonna replace all the investigators
Scott Heath: tomorrow. But what it is gonna do is make their lives better. Are there lots of other things we didn't cover today? Absolutely, that go into this. But I think the core take away, then, is, we think, we can help you where you are today. So with that we'll open it up for questions. If you have any questions down there in the bottom. please type those in
Scott Heath: let's see. I have a question here for you, Mike. The question is for the data structure and the vector search is this data that's been ingested into Connecticut? Or will this search work across other data sources.
Mike Booth: The data would be ingested into Connecticut.
Mike Booth: Right? So yes. So, Scott, the the data would be ingested into Connecticut. You know, for the Llm. For certain.
Mike Booth: We can pull from other data sources. But everything would be running within Connecticut.
Scott Heath: Okay, here's an a similar one, and I'll I'll answer, maybe the first part of this, and I'll throw it over to you with so many data sources and various jurisdictions. How do you maintain data, consistency and standards from the business user perspective. We use either your existing entitlements.
Scott Heath: which is, who can see what data inside of the system. And how do we contain that? And then we work with Connecticut, and I'll hand that off to you. To be able to create it more at the physical level. And and Mike, you want to answer, sort of how? With so many data sources in various jurisdictions, how do you maintain data, consistency and standards.
Mike Booth: Yeah, I mean, we write down a role based access control. We can manage, you know, sort of who sees what? As you're deploying this out. Say, you know, across multiple countries, you're the ability to then tie in and block information through our role based access control is managed within the system itself. And then, you know, the data feeds on a on a case by case basis would be done.
Mike Booth: you know, done at the user level. Right? So in in understanding what a user can and can't see would be would be managed in a you know, through a console that we have managed on a user by user basis.
Scott Heath: Okay? So I think the short answer is, both the business user layer and the database have consoles where you can set strong or or what the the conditions are for either entitlements or data access, etc. The next question that we have here is, we're a brokerage firm. And we are getting millions of transactions a day.
Scott Heath: How fast can can Connectica or the back end ingest data?
Mike Booth: Yeah, it's it's it's sub-second ingest, and we've got plenty of examples. In fact, citibank uses us for their global trade reconciliation.
Mike Booth: as a use case where we've had great success. But you know, II it's I'd be happy to talk in in more detail, one on one based on the type of data you know in in the you know, and how you're planning to ingest it. But it's done, you know we do a as another use case trade reconciliation. So, taking a look at a market transaction and mapping that to to a corresponding, you know, valuation within a within a market.
Mike Booth: you know, is is done on the fly with sub-second, you know. In most case.
Scott Heath: I'll answer the other part to that which is today we are doing trade surveillance. Aml credit card transactions where they are. Basically, as Mike mentioned, a sub millisecond. And one of the things about the architecture we we talked about today is we're not waiting in a serial pipeline for it to load, settle, and then run math, and then do things. We're actually running those in the memory or compute during the ingest, so that
Scott Heath: second, both ingest and alert logic that I showed are all happening in more of a concurrent model. So the the velocity on this is is quite amazing.
Scott Heath: Last question here. And and and we said, feel free to ask other questions is, how how would I get started. What kind of data would I need? So I'll answer the first part. And then, Mike, you can answer the second. So normally, what we look for is, whatever your use case. So if you're in bank fraud, trade, surveillance, Aml. Credit card, whatever that may be, is a requisite set. If you know that there is data
Scott Heath: in a month or a year, or whatever that may be. Right, we would need that kind of data. We would need whatever sort of transactional data transactions, processing data that you may have. So if you have. If you're looking at a transaction system where you already have alerts, that's the kind of data that we would need. And as long as you have that, it can be obfuscated or we actually have synthetic data sets that that are what we call realistic
Scott Heath: that we could jump in and look at a Poc, and so those things can be rapid. Those can be in, let's say, a week or so. Etc.
Mike Booth: Any other thoughts on sort of how to do a rapid prototype. Yeah, absolutely. Thanks, Scott. So we are, we support all 3 major cloud platforms. So if you're using any of those, we can certainly stand up on your instance with your data, if you choose to do something you know that where you don't want to allocate an instance
Mike Booth: we have an environment where you can pass your data through and and test it. Obviously, there's some sic INFOSEC challenges there. So you gotta be careful. But we can certainly do that as well. And then, you know, the third option would be, you know, an on Prem environment, whether you're running cpus or gpus and and standing up an on Prem instance, we're happy to help spec that out and get you started. But most organizations start off
Mike Booth: in the cloud, whether it's us hosting it on their behalf or them allocating, you know of their preferred cloud provider.
Scott Heath: Okay, next one here is, can this be used in crypto companies? Help reconcile crypto currencies? The answer is absolutely so today we're working with a large number of the current.
Scott Heath: The biggest firms that are out there, and that is one of the hotter ones is looking at crypto chant custody, etc., is absolutely tracking every portion of that as best we can and connecting it to that wider range of other data. So absolutely and and that is now becoming actually almost de facto or common in every used case that we have.
Scott Heath: Next question we have here is, what is the scale of the data solution can handle, since graph and geospatial attribute and graph require big storage and high computation. I'll answer the first part, Michael, throw it to the second part. So what are the common? So today we're doing?
Scott Heath: I don't know. Probably 12 of these today right now, and that is one of the fundamental questions, great question, right? Which is what we're doing then is we're we're looking at maybe 50 TB of macro data. This is unwashed
Scott Heath: transactional data over 2 to 3 years. It can go even higher, depending on if you integrate from multiple business units. That's one of the core capabilities of Connecticut. As Mike showed, a slide is almost as a feature store as a data warehouse it can handle, and some of the examples I think their highest one that Mike can tell us about is that Pentagon one where it's hundreds of terabytes.
Scott Heath: The difference is, you can now select those about whether you want to send 1 10, whatever the number is of that 100 TB
Scott Heath: to the graph solution. So you've now right sized a query to a graph database versus in in in previous days you would send all 100 TB to a graph database, and I don't care who that graph databases out there today. They're all great. We love them. But that's too much data in a graph database. They are very much about smaller segments. This solution really allows us to do that. And so the high computation has now been focused on things that are fraudulent or
Scott Heath: potential fraud capabilities that are a fraction of that data. So that's why the Connecticut layer is so incredibly important. Mike, your your turn?
Mike Booth: Yeah, thanks, Scott. You actually summarized it really? Well, and you know the other example of the Us. Postal service is, we can take all of their route optimis, all of their routing data across the Us. And look at all of that data and realize where there may be changes to a particular route.
Mike Booth: address those changes through the computational analytics and and and make suggestions back to those impacted mail carriers. Yet holding all of the relevant data, should a change happen in another location. So we do store we can store all of this data in a you know in in, you know, in a, as a, as a data store, and then pull in what's necessary. And and, as Scott mentioned, our largest customers are well, over 100 TB.
Scott Heath: The last one that we have here is, how do you handle complex queries from case managers. Great question. So what you saw then was the integration in a couple of the vignettes is whether it's optimized or hummingbird, and whether it's guide wire for insurance, or or whatever that may be, we handle connections to multiple kinds
Scott Heath: of case managers. Now, case queries can be simple in that. Show me that deeply connected, and geolocation, or to your point. Show me, those very complex multiple alerts and other attributes. They can be connected in 2 ways, right? We can connect it via a rest endpoint.
Scott Heath: So the way that the solution on the front end works is the queries from multiple data sources. So we can leave data and optimize. Leave data in hummingbird, and then take a predicate of a case or an alert. Ask Connectica the question of where is the where are those connections? Where is the fraud like we saw there, and the time series, and then return that as one nice rest endpoint?
Scott Heath: So the answer is the tool and the connectivity inside of what we're doing today. We call it the semantic layer which is why the the basically the connected financial crimes tool is so incredibly powerful. We simply can make those connections. We can take those complex queries in the semantic layer and a lot of that difficulty has been taken out of the equation for you, the customer, or you the technology person. So the beauty of that is, we can expand it.
Scott Heath: And we can. We can typically do very, very complex elements. And we can either leave the data in that source system as we said, bolt on, or we can carry some of that data for feature stores, etc.
Scott Heath: Okay, well, let's see if there's any other questions out there. How do you handle a complex with the Llm. So the Llm that we showed today
Scott Heath: comes from Connecticut and can be expanded into other systems. Right? So if we use the predicate data out of a case manager, we can use this to do it. Now, there are some
Scott Heath: rules that apply here, and we would want to talk about a baseline. Gpt model for what that is. And so we can use those components. And it depends on what you already have in our version. Obviously, it's built in and so some customers are choosing to keep simple things in their their older case managers and more complex things because they're kind of already built into ours. And so again, your mileage may vary.
Scott Heath: So whoever this person is that. Ask that question you might reach out to us afterwards. Okay, well, right on time. Now, that looks like all of our questions. Very lively session, Mike, I want to thank you for your time today. Great session. And then, as we mentioned before, all of this is being recorded, and we'll send out the recording to everyone. And obviously, if you have any questions, don't hesitate to reach out to either or both of us. And we would love to talk about your use case. Thanks again. Everybody.
Mike Booth: Thanks, Scott, thanks. Everyone.
