Podcast: Using AI to Identify Fraud

Whitney McDonald 0:02
Hello, and welcome to The Buzz, a bank automation news podcast. My name is Whitney McDonald and I’m the editor of bank automation news. Joining me today is AML, right source Chief Technology Officer Phil McLaughlin. He’s here to discuss the need for anti money laundering practices, and advancements in AML. Technology.Phil McLaughlin 0:22
My name is Phil McLaughlin, I’m the Chief Technology Officer at AML. Right source. Amo, right source is a provider of managed services, which is people, financial crime advisory services, and then also technology platforms, and sort of the blending of those three offerings together in technology enabled managed services, and we support banks, other non bank, financial institutions, fintechs, all over the world, we have around 4000 investigators that work with our customers to help them stay compliant in the AML KYC space. And we’re bringing technology solutions to those customers, to help them be more efficient and more effective. And, you know, that’s really the the problem that we’re we’re all about, you know, trying to make the efforts that our customers and that that our, you know, internal teams are trying to accomplish as efficient as effective as possible.

Whitney McDonald 1:20
Great. Well, thanks so much for joining us on The buys, let’s take a step back here first and set the scene with financial or fighting financial crime today, you could talk us through really the need for this advanced technology, especially when identifying money laundering.

Phil McLaughlin 1:39
Definitely. So the the estimates that are out there today are that basically the current methods that we’re using for any money laundering, our lack, you know, are lacking, right, they fall short of what we really need to accomplish here. If you look at a number of estimates from the UN and others, it’s something like two to 5% of global GDP are, you know, between 800 billion and $2 trillion that are involved in, in money laundering, and we’re probably only catching maybe 5% of that. So despite the significant amount of effort that banks, regulatory agencies, folks likes us that are in the services and technology business, you know, there, there’s still a lot of room for improvement to make this stuff better. And then when you sort of look at the technology side of this, that the technology systems themselves that are helping are really not all that effective, they look at relatively relatively small amount of data, when trying to make assessments, they are really pretty simplistic in terms of the things that they’re looking at, like simple patterns, that sort of stuff, simple name matching. And we know that the the reality of the of the financial crime space is a lot more complicated than that. And so really, technology needs to come in and help improve this. You know, again, the way to think about this is, this is largely today a very human intensive effort, the tools alert or highlight certain characteristics, but it’s really left to the investigator really left to the human being to do the vast majority of the legwork, do all of the data synthesis, do the evaluation, make a conclusion, draw a recommendation, document all of that. And it’s a very, very time consuming process. So the degree to which technology can be employed to help make those human beings more efficient and effective. That is, is where we’re going.

Whitney McDonald 3:35
Now, before we get into where we’re going with, with new technology and advances in technology in this space, maybe we can talk through what exists today. What are some best practices in tackling, identifying and in identifying money laundering today?

Phil McLaughlin 3:52
Sure. So I think we’re, we think about this, kind of from a current state future state sort of thing, right? So really, the goal is gonna be to improve the level of automation and to include or improve the level of efficiency with the investigators. Like I said, a lot of the processes today are very limited in terms of what they look at. So you know, as you’re thinking about as people are thinking about, you know, how would they improve their process, looking at more data, automating anything that they can the robotic process automation capabilities are out there are a good place to start in terms of, you know, thinking about how to make things better. Expanding the frequency of monitoring again today, because it’s a very human intensive process. Things get looked at maybe on a once a year basis, once every six months basis, if there’s things that we can do to make that an ongoing, continuous monitoring type of a solution that lets us find things faster, and allows human beings to flow focus on the things that are really salient as opposed to separating the wheat from the chaff so to speak. Again, a lot of the tools that are out there right now, or are very limited in terms of their technology or their their detection capabilities, a lot of them are rule based. So, you know, the simple rules that are capable of being implemented in these kinds of solutions are, are very limited. And that’s really why, you know, the broadening of the of the technology platforms and the algorithmic content and moving towards AI, and some of these other things are so important to help us, you know, begin to tackle these problems in a more efficient way.

Whitney McDonald 5:41
You can’t talk about anything in technology right now without talking through AI. Right. So maybe you could expand on that a little bit. Why is AI well suited for this type of technology? And how can AI fit into this puzzle?

Phil McLaughlin 5:55
Thing, AI is exceptionally well suited to the AML challenge. The thing that’s great about it is, is that, you know, as people now are starting to have a pretty broad awareness, some of these AI tools and techniques are really approaching the ability to emulate, you know, the more advanced features of human cognition, right, so they are really able to, not only, you know, do what we consider to be really relatively simple things, but but much more complex levels of thinking much more complex levels of inference of summarization, those kinds of things. And, you know, being able to figure out even with traditional AI techniques, you know, be able to, to do anomaly detection, figure out what’s notable, and, you know, separate the needle, find the needle in the haystack, so to speak. There’s a bunch of different flavors of AI that are sort of relevant here, you know, two good examples are natural language processing. So if you think about what an investigator has to do, to go read news articles, read various documents and artifacts, and try to infer and connect and synthesize all the connections there. It’s a huge amount of work and the degree to which you can get knowledge from text and understand it and present it to a person in a way that is easy for them to then internalize and take action on. That’s just a super, super big force multiplier. And then, you know, the more traditional, you know, machine learning models, whether they’re classifiers, or whether they’re other types of, of neural networks are really good at at, you know, training to be able to figure out things like entity name, or entity type from an entity name, that’s one of the problems in money laundering is that the, the banks and financial institutions know a lot about their customers, because they vetted them in the onboarding process, but they don’t know much about the counterparties or other related parties. And so the amount of work that can be done to to, in an automated sense to try to collect information on those related parties and counterparties is going to make the total understanding that the investigator has that much more clear and allow them to, you know, more, resolve those issues or solve the cases in a more timely manner.

Whitney McDonald 8:18
Now, we’ve talked through the technology, the opportunity for advancements here the need for solutions like this. Can we talk through where AML right source fits into this and how the technology works?

Phil McLaughlin 8:31
Yeah, sure. So as I mentioned earlier, email is a provider of technology enabled managed services, as well as software solutions to banks, fintechs, and other institutions that have regulatory requirements to help oversee the safety of the global banking systems. We have 1000s of investigators working in the field on KYC, suspicious activity monitoring, you know, those around the globe, really, across the all the different global geographies, in addition to you know, providing sort of these AI LED technology solutions. So we’re really all about trying to bring this great technology along with great people to our customers. You know, one of the things that I would say to somebody who’s looking into trying to embark on, you know, putting their toe in the AI for AML waters is, make sure you work with somebody who knows AML because if you’re just going to work with somebody who knows AI, you’re going to end up paying for their learning curve. And there’s so much nuance in terms of the data and the risk bearing characteristics that are that are relevant and important in the AML space, that you really want to have a partner that understands that stuff. And so, you know, we think we are, you know, the best of the best in that regard, really having, you know, strong practitioners, coupled with that AI technology, you said bringing that AML AI, sort of blend to the our customers.

Whitney McDonald 10:07
Now speaking of a customer, maybe you can talk through or identify some use cases who would use this? How would you get in? How would you integrate maybe talking through what that entails?

Phil McLaughlin 10:20
For sure. So our customers and our solutions tend to follow the customer lifecycle. So think about your relationship with your bank, you open your account with a bank, they onboard you, they make sure you’re not a bad guy, they make sure you’re who you say you are. Once you’re on boarded, then you can start transacting. So there’s some, you know, transaction monitoring that’s going on the so called suspicious activity monitoring. So we’re helping in that regard. There’s also sort of know your customer monitoring that goes on through the course of the lifecycle. So let’s say you’re a bank, let’s say you’re a corporation, and you’ve just had a change over in your board of directors, and you want to understand, you know, you’re the bank wants to understand, is this new person on your board? Are they a good guy? Are they a politically exposed person? Do they have? Is there negative media about them? Is there some other risk that should be surfaced related to, to this district board member. And so we have tools and techniques that allow us to monitor changes in those activities, identify that a change has occurred, evaluate the parties involved, to see if there’s a risk event that we need to surface, and then we’ll surface that, then then, you know, we also help with more broader just workflow across that whole client lifecycle, helping customers to manage that full trajectory from onboarding through monitoring through suspicious activity detection, periodic monitoring, and then to offboarding. So it’s, it’s all the stuff that you’d think about in terms of, you know, that full lifecycle.

Whitney McDonald 11:59
Now, quantifying here some savings that that someone that a bank might benefit from, from this client might benefit from this catching fraud examples of successes here.

Phil McLaughlin 12:14
Yeah, definitely. So like I mentioned, the big banks do a pretty good job of understanding who their customers are, but it’s this community of related parties where there’s often a lot of insights that can be gained. And also just like, understanding sort of the specific nature of the activity and trying to identify if something is anomalous. So for example, we have, you know, a tremendous number of our customers who’ve seen, you know, instances where they’ve identified risk in in Counterparty. So for example, some buddy might be have negative media associated with them, they might be a bad guy, they might be a politically exposed person, that kind of stuff. Some of the more interesting ones, when you start looking at the AI techniques, the more advanced AI techniques is looking at things like inconsistent line of businesses. So if you’ve got a banana, or steel company, and they’re buying iron ore, that makes perfect sense, right. And if you’ve got an iron, steel company, they’re paying for bananas, that doesn’t make sense. So the tools and techniques are able to learn by looking at a massive amount of data, what kinds of relationships are appropriate, what kinds of relationships are inappropriate or consistent with what one would expect. And they can highlight that to the investigator that this, this company seems to be doing something that is counter to what one would expect given, given what we know about them. We’ve seen a number of instances of that with our customers, we’ve also seen the issue of money going the wrong way. So let’s say you’ve got a we’ve seen an instance where there was a casino, and they were getting transacted with a company that makes computers and so you would expect to see the money flowing from the casino to the computer company, because they’re purchasing computers to use in their Casino. That would be a perfectly reasonable use case. But what we saw is the money going the other way. It turns out that after further investigation, the the gentleman who was the head of the computer company had a bunch of different activity that he was involved in. And you know, we were able to help surface that particular instance, we’ve seen other instances where companies are related to risky parties or risky jurisdictions. So let’s say that people are concerned about doing business with any buddy who’s not only in Cuba, but doing anything related to Cuba. And so we’re able to detect, for example, that there are companies in Venezuela, who are arranging travel to Cuba, which is not illegal in the context of what they are doing as a company but But, but the US banking folks would want to know that that party is has a relationship with Cuba and is doing something there. So there’s, there’s a lot of those kinds of instances where, you know, we’re able to surface relationships or surface characteristics about the related parties that help make sure that the, the, our customers understand what that full picture of risk is. And it just wouldn’t be practical for humans to do all the legwork to hunt each and every one of those things down. So, you know, at the end of the day, it’s really coming back to automating whatever we can, for the investigator, making the investigator giving the investigator, you know, the, the best point of departure to resolve the investigation as they can. So I the analogy that I like is, um, let’s say, doing an investigation is a 100 meter dash, you know, if we can start a client at the 50 meter line, or the 70 meter line, and all they’ve got to do is get to the end, then that’s, that’s, that’s the goal. And that’s, that’s really what we’re seeing with our customers, they’re seeing a significant amount of savings, in terms of the amount of time that it takes. And it also puts the investigator in a lot better position because they’re able to then instead of doing all the legwork, all this grunt work of doing Google searches and searching for names and structured databases and searching, you know, downloading transactions and building pivot tables, and totaling in sub totaling all this stuff to see what’s going on. We can give them all of that prevented, we can give them all of that, in a human readable narrative, supported with all the documentary evidence, and it really lets them the investigator focus on using their training their experience, their their education and, and an expertise in actually understanding if there’s financial crime there, as opposed to being an Excel expert or a Google search expert.

Whitney McDonald 16:59
Now with with these use cases, and working with clients and and all of that what you just discussed, what are you working on when it comes to innovating in this space and forward looking maybe just to the end of this year? What am all right sources is working on I know, we talked through AI opportunity and machine learning and of course generative AI as a as a buzzword as well, maybe you can share a little bit about what you’re looking into?

Phil McLaughlin 17:26
Yeah, for sure. So, the good news for us is that we’ve been really bringing AI to the financial crime flight now since 2015. So we are well versed in how to use and employ these different techniques to to solve the problems. We’re looking right now, working in a couple of different areas, one major area that we’re looking at is we’re rolling out the next generation adverse media solution that we have. So really helping, you know, our customers very effectively and efficiently get surfaced articles, news articles content from around the world, that might indicate that they’re a customer or a related parties involved in something that would be risk bearing, we have a tremendous amount of natural language processing and other artificial intelligence techniques that are baked into that, and we’re gonna see, you know, a two fold improvement, at least in terms of the efficiency with with with which the investigators can adjudicate the articles as well as a significant drop in false positives. All of these adverse Media Solutions, try to do their best to give relevant content, but it’s a hard problem to solve the next generation of our stuff that we’re bringing out is going to do a fantastic job of that. We’re also we are working in a number of different areas with with LLM with the generative AI techniques. You know, the way we think about this is, this is just another tool in the ever evolving AI toolbox. So, you know, when when we talk about AI, it really spans the gamut of all the different things that can fit in there, right, from natural language processing to more traditional, supervised and unsupervised machine learning to the new LM and a whole bunch of other, you know, techniques that are in this toolbox. And so, you know, our view that L is that LM is is just another tool that we can utilize to help solve problems. The work that we’ve done with LM M’s and we expect to have some of these use cases in production in the next few months, has largely to do with with inference and reasoning and summarization, like those are the things that the algorithms are really very good at. So asking the LLM, read this article and tell me if this entity is a good guy or a bad guy. They’re pretty good at that. Looking to do knowledge extraction, taking the LLM and saying, you know, tell me how old the subjects in this article are or tell me what jurisdiction in there that are in, those are very easy things for humans to do. Not very easy things for some of the traditional AI techniques that we’ve had out there, and, but are something that LLM ‘s are very good at. So, again, we’re looking at a number of different areas having to do with data inference, summarization, those sorts of things. And we’re going to be peppering them essentially, throughout the solutions, we’ll be sort of using them to augment the existing capabilities. A lot of the techniques that are there could have AI techniques are often layered. So you may start off with one technique, and that may get you 50% of the answers, then you may need to go to a second technique with that is different or better to get to another 25%. And then you need to go to a third technique to get you in another, you know, 10, or 15%. And so the way we think about these MLMs, in the short term is, is them just being another layer another tool to help fit into that tapestry of, of solutions that we’re using, you know, in the big picture, our view is that, you know, these, the MLMs are here to stay, they are going to become more and more important tool in the toolbox. Like I said, they’re not going to replace everything. They don’t do everything, as well as some of the other techniques. But I think that over time, we’ll see them becoming more and more prevalent. I also don’t think that in this space, at least LLM ‘s are ever going to just entirely take over the the process, right. There’s always going to be the need for human judgment, human intuition, human training and experience to be able to adjudicate the final outcome. And while the LMS can definitely help with efficiency and effectiveness, they’re they’re never going to be maybe never too strong. But in the near term, they’re not going to be sort of the standalone, you know, Uber AI solution that that answers the questions for us.

Whitney McDonald 22:12
You been listening to the buzz of bank automation news podcast, please follow us on LinkedIn. And as a reminder, you can rate this podcast on your platform of choice. Thank you for your time and be sure to visit us at Bank automation news.com For more automation news,