Fighting Realtime Fraud with AI and Behavioral Modeling – Aman Cheema [Episode #16]

Podcast Intro (00:00:00) – Hello. Welcome to P.I.T. Exchange, a podcast by Currency Research. Join us as we discuss the latest in payments, innovation and technology with the industry’s most innovative thought leaders. Today’s payments are changing and moving around the world faster than ever before, and P.I.T. Exchange gives you the knowledge and insights to keep up. Sit back and relax as we join Currency Research, exchanging ideas with today’s special guest.

Shaun Ferrari (00:00:01) – Hey, everybody, welcome to the next episode of The P.I.T. Exchange. We’re recording live here in Kuala Lumpur at our Currency Research Payments Week. We hope you’ve enjoyed some of the broadcasts we’ve brought you already this week. I’m thrilled this morning to be joined by Featurespace. Hello. Hello. How are you? Good, good. I hope you’ve enjoyed the the week so far. It’s the last day.

Aman Cheema (00:00:23) – It’s been a great week and, very entertaining, very thoughtful. A lot of good. A lot of good conversations.

Shaun Ferrari (00:00:30) – Awesome. Glad you’ve glad you’ve enjoyed it. And it has been it’s been a busy week. I mean, I think we did the two events back to back, the central Bank Payments Conference and now the Global Payments Summit. And thank you for being so involved in both of those events. There’s a lot, a lot going on in the whether it’s cross-border or, just AI you’re just talking. There was just a session inside on a keynote with AI and how that’s making folks evolve their what they’re doing.

Shaun Ferrari (00:00:56) – I know you all are heavily involved in the AI space. Maybe just take a second and tell folks what your what Featurespace is all about and, yeah. Start there. Yeah, yeah.

Aman Cheema (00:01:07) – Absolutely happy to do so. So Featurespace. We are a, machine learning AI based company, but we apply those technologies for solving and solving fraud. So preventing, preventing fraud, as the sole purpose of the company. That’s why we exist. That’s the only thing that we want to do is help nations protect their citizens and businesses and investors from, from being defrauded.

Shaun Ferrari (00:01:36) – Awesome. Yeah. And I mean, I know I said in one of your sessions, earlier this week, some of the stuff you’re doing to work on that fraud prevention, is really, to me, it seems groundbreaking in terms of kind of where you’re putting, putting your technology and what is looking at and how much it can help identify fraud that’s not already being captured, by the current system. So maybe just a quick like, you know, what do you think the, what, give it maybe a very quick overview of where we are in terms of real time detection of fraud right now versus where we can get to.

Shaun Ferrari (00:02:11) – Yeah, or need to get to.

Aman Cheema (00:02:13) – Give some context behind the problem. So, the speaker just now, Farhan, CEO of PayNet, really, really great presentation. And he put a stat on the slide that says a quarter of the world’s population has been, subject to fraud and scams. A quarter of the world’s population, over $1 trillion has been, defrauded and stolen from hard, hard working, hard earning, people, businesses around the world. So that’s the context of the problem that, you know, Featurespace is, is trying to address and, and is addressing. But how we go about doing that. The company was born out of the University of Cambridge in the UK, and the two founders, professor Bill Fitzgerald and David Excel, they effectively invented adaptive behavioural analytics. And that’s the science and the math that’s behind a lot of the machine model, machine learning models that are out there in the world today. And what these models allow us to do is really understand what is normal behavior.

Aman Cheema (00:03:25) – So the angle here is historically fraud, fraud detection used to be around, at best writing rules into some sort of computer system that were quite binary. And that said, you know, Shaun, if he spends over $1,000, that doesn’t look right for Shaun. So stop doing Shaun. Anytime Shaun does $1,000 transactions, stop it, decline it. Very binary rules remain limited, but sometimes can be relatively effective. So we say if you don’t have anything, have rules. At least, at least you got a good chance to catch something. What adaptive behavior analytics does is it tries to understand you and create a profile of you, and therefore understands what is your good behavior. And therefore, if you deviate from your good behavior says, that is not right. Like Shaun typically does his payments on a Friday and you did it for this amount, and it’s typically from this device and it’s for, from these two of these merchants, etc.. So builds a pattern. Understand your pattern, but then all of a sudden you start doing $1 million payments every Monday, and it’s to various countries around the world that’s a bit odd.

Aman Cheema (00:04:36) – Right? So, so this is a bit odd. It’s not a normal Shaun behavior. It flags it in the system. It sends it back to to whoever your bank is and says maybe you want to look at this transaction or actually probably decline it. Right. So that’s what these models are doing. Right. and the power of these models that you don’t have to go back into these models and retrain it. So for example, your pattern changes, the fraud attacks change. You don’t have to go in and have real hard code the system again. It’s learning. It’s always learning your behavior. It’s always learning your patterns. It’s always learning the the fraud attacks that are out there. So you don’t have to go in and, you know, have the machine learning, machine learning model working, sitting in the background and say, well, I’ve done it. Now I can just walk away and just the the bank will be safe, it does that but as we always say though, it’s always good to have a human in the loop to keep an eye on the model to make sure it is always working.

Aman Cheema (00:05:30) – But the essence of the model is always retuning. It’s always self-learning, and therefore the detection rates of fraud don’t drop off. They’re continually improve.

Shaun Ferrari (00:05:40) – Yeah and I think that was one of the fascinating things in hearing you discuss what’s, what’s going on in this space over the week, is that I always kind of assumed that you brought up the retraining of the model piece, that it would be a pretty massive undertaking, for an institution to be like, okay, I need to input or inject one of these models in my process flow, and then it’s like, okay, well, I gotta train and I gotta do this and get all the data. I think that was one of the big takeaways that I had was that no, because it I’m probably oversimplifying, but if it’s because it’s based on kind of your personal behavior, it’s not necessarily needed to take a whole lot of time to learn the specifics of that institution.

Aman Cheema (00:06:21) – It’s correct. So the, the, the way these models have developed and, you know, Featurespace began for 15 years and only been doing this, you know, 13 for for 15 years.

Aman Cheema (00:06:31) – So we’ve got a lot of experience, right, of of building these models and seeing how they perform, combining with our expertise in payments and frauds. The models that get quite, quite sophisticated pretty quick, but it sounds very complex. Well, I don’t think my team will be able to understand these models, and I think we’ll be able to deploy these models. We’ve done a lot of work in making sure that organizations can use these models very quickly and very effectively. So we’ve hidden all of that complexity behind the scenes. So now we’re in situations where, where you can take a model, most of these models can be pre-trained and they can be deployed in, in sometimes in days. Yeah. Yeah. It depends on the organization situation. but we can take a cod’s take a model that’s detecting cod, cod card fraud, get deployed within 30 days. And because it’s pre-trained it doesn’t need a lot of data. So in fact, as soon as this 30 days implementation is done, you can start to detect fraud.

Aman Cheema (00:07:31) – Yeah. Simple. and over time, it just gets better and better because it’s learning the getting the data from the systems and it’s continually learning. So we’re in a world like that now where I guess five years ago, ten years ago, these systems would have been quite complex to install. There would have been there was required a lot of data for training. and so the effectiveness probably takes six to 9 to 12 months. Right. We’re not in that world anymore. So we’re very passionate about finding fraud, detecting and fraud helping as much as we can, as soon as we can and the innovations and developments we’ve done in the tools are just pushing us to that world where we can help detect fraud, now. Yeah.

Shaun Ferrari (00:08:13) – No, that’s great. And I think one of the things, and one of the topics we’re going to have in an upcoming event that we’re doing all around central banks and AI, and in London in September, we’re going to have a topic on kind of responsible AI and bias and all that sort of stuff.

Shaun Ferrari (00:08:26) – I’m just curious from your perspective in terms of, you know, the some degree of inherent bias in these models? how do we how do you deal with that? Or how should the industry think about dealing with that?

Aman Cheema (00:08:39) – It is a very good question. And you have bias. And then the next set of generation of models where you apply generative AI. So Featurespace has launched the world’s first large transaction model. So what generative AI has done for language and for audio for visuals has done for the world of deepfakes. As we saw in the presentation day from Farhan. We have done, taken that mindset and that philosophy and applied it to the world of payment transactions. So, talent LTM is a large transaction model, and with those models now you have better place to understand and correct for bias. I don’t think it’ll ever disappear. Right. Just the nature of things are, but you enter also world of a hallucinations, right? So it’s like, is this real or is this not right? So it’s a great, like, these tools are great and they will get better.

Aman Cheema (00:09:38) – but these are the few things that we just have to be conscious of when we deploying this technology. Yeah, well, and as you say.

Shaun Ferrari (00:09:43) – There’s there always has to be that human aspect to where, you know, people to keep an eye on things like that. And yeah.

Aman Cheema (00:09:48) – It’s a big believe is that, you know, you still have a human in the loop. It doesn’t matter how good your models are performing, how good they are detecting, is good to every now and again just to check in, you know, just over it? are you performing the way you should? Have you gone off piste or even. How can we improve? Yeah. So the human in the loop will always be big fans and believers in that. Just don’t let the machine just sit by itself, but it’s a very interesting time. we talk a lot about now, how do you how do central banks and networks around the world take advantage of this technology? Right. So as we’ve seen with some of the exercises that we’ve done with networks recently, is network should be seen as a last layer of defense.

Aman Cheema (00:10:31) – At the moment, they’re not considered any part of the defense strategy of detecting and managing fraud. Some countries nations are on that thinking, and we are helping them with with that. Right. And we have two clients, one on cards, network obviously card network and a country that’s doing real time payment network. They’re seeing the correlation between real time payments or fast payments and and creating forced as a new term to go after and they realize that they can do more for their nation and by putting in fraud detection on the network. So that’s an interesting development. And we see, you know, the strategies now, hopefully more nations deploying and saying, okay, commercial sector, you should be doing your level one, level two or layer one, layer two defense. And the network should be doing the last layer of defense. And you know, doing their for the nation is doing as much as it possibly can. Right. It always could be more for sure, but as much as it possibly can to help its nation, help its citizens businesses, to, protect them, protect their money.

Shaun Ferrari (00:11:36) – Yeah. I mean, I think that it is a nice kind of shift because a lot of times the responsibility in these things is kind of like pushed downstream, right, of like, oh, no, it’s the well, the commercial bank is the operator or the, the fintech is managing it for the commercial bank. And that’s where the kind of the responsibility lays. But I think you’re exactly right. I think the, the operators now, the central banks or whoever is the, the network operator is, is starting to take on a degree of responsibility, for sure, in partnership with the commercial sector, which is great because it’s, as you say, it’s it’s a huge problem. It’s only going to get more complex and more challenging, and everybody needs to try to solve it together. It’s not good enough to just kind of point the finger one way and expect that it will happen.

Aman Cheema (00:12:18) – Yeah. Yeah, absolutely. And we see some nations who are trying to deal with this problem at the network.

Aman Cheema (00:12:25) – Well, at a, at a national level through regulation. So in the UK we had a session to this this week, as you know, from the payment service regulator in the UK and the UK is a taken an approach of doing two things slow down the real time payment system, which sounds very counterintuitive. You want real time. And then there’s clear evidence that a country go in real time adds to GDP. Right. And international competitors. But they had to slow it down because the amount of authorised push payment for fraud that was happening in the UK.

Shaun Ferrari (00:13:00) – And I think just on that point, and when we say slow it down, the nice part about, I don’t know, nice in quotes is like it doesn’t mean slow it down by days. It’s like you just need to slow it down so some checks can be done, some some additional processes can occur.

Aman Cheema (00:13:15) – Unfortunately the UK did slow down by they did.

Shaun Ferrari (00:13:17) – But in theory with the technology you don’t need you.

Aman Cheema (00:13:19) – Don’t need to, you don’t need to.

Aman Cheema (00:13:21) – But they decided to slow it down so they can do the right checks and well, give a better chance for the commercial sector to write checks. The second thing that the UK has done is, made banks liable for authorised post payment fraud losses, right? Yeah. and you know, if there’s ever a stick moment that is an example of a stick moment, right? Yeah. but that shouldn’t be the only tools, right? We’ve proven, for example, if you do fraud detection at a network level, you can detect a lot of fraud that misses that. That just gets through the net. And you see the last line of defense, the last bit of the net could be the, the network. you can detect a lot of fraud. You can detect, different fraud patterns that you may not be able to detect at a commercial bank level. So with scamming and muling as a technique, you can detect that very well. If you can see both sides of the transaction, the inbound and outbound, as we see as we call it, you can do that at a network level very effectively.

Aman Cheema (00:14:23) – And most importantly, the exercises that we did with, Featurespace did with pay UK recently to answer these exam questions on how much fraud you can detect at a network level, you can, you can find a lot more fraud that no one even looks for, right? Right. Yeah. Because what you can.

Shaun Ferrari(00:14:43) – See, I say you can see the universe.

Aman Cheema (00:14:44) – You can see the universe. Yeah. Yeah, exactly. So like, just shy of 30 odd million of fraud that no one even bothers looking for because it’s really, really, really hard to find. we like technical challenges. We like academic challenges. We, data scientists and data analytics. Love to be challenged. And with that data set that we have from the UK. We went to town with it and we found the frauds that no one else was picking up. So the possibility of of saving, you know, pain and suffering from many losses is, you know, this is not a solution that needs to be invented. We’ve got it today.

Aman Cheema (00:15:22) – Nations can use this today, can use it as a layer of defense to protect its nations today. And that’s, you know, you know, hopefully the narrative, you know, lands and works, but this kind of way, no excuses and kind of like, shame on you for not using the tools. Yeah, yeah.

Shaun Ferrari (00:15:39) – No. Exactly. Well, cool. Thanks for sitting down with us and chatting for for a few minutes. I think it’s an amazing, it’s, you know, I don’t know if it. I think it was, you know, having your presence here really injected some of that needed discussion about fraud onto the agenda. I think we were talking before. It’s like it’s inherent in a lot of these discussions, but, you know, I don’t know that it’s definitely picking up agenda steam and people are talking about it because it is a problem that needs to be solved. So thanks for kind of pushing that issue and forcing that.

Aman Cheema (00:16:09) – I apologize if fraud wasn’t meant to be the intended, thematic of the event, and it’s now become the magic of the event.

Shaun Ferrari (00:16:19) – No, it it’s an emblem. It’s emblematic of issues that are happening, happening in this space that we need to address. So, so thanks for taking the time. Pleasure having you here. Enjoy the rest of the day here. And thank you for listening to another edition here in Kuala Lumpur. And we will chat with everybody soon. Take care.

Podcast Outro (00:20:28) – Thank you for listening to The P.I.T. Exchange, a podcast by Currency Research. Check out our upcoming events and publications at and join us for our next episode to hear what’s trending in payments, innovation and technology.

Welcome to another episode, recorded live from Kuala Lumpur at the Currency Research Payments Week. This week has been filled with insightful discussions, and today, we are thrilled to be joined by Aman Cheema from Featurespace, a pioneering company in the field of machine learning and AI-based fraud prevention. In this conversation, we will be focusing on the current state of fraud detection, the groundbreaking technology developed by Featurespace, and the evolving role of central banks and networks in combating fraud.

Understanding the Magnitude of Fraud

Fraud is a pervasive issue affecting a significant portion of the global population. As highlighted by Farhan Ahmad, GCEO of PayNet, during his keynote presentation, a staggering quarter of the world’s population has been subjected to fraud and scams, resulting in over $1 trillion in losses. This alarming statistic underscores the critical need for advanced fraud detection and prevention technologies.

Featurespace: Pioneering Adaptive Behavioral Analytics

Featurespace, founded by Professor Bill Fitzgerald and David Excell from the University of Cambridge, has revolutionized fraud detection with its adaptive behavioral analytics. This technology goes beyond traditional binary rules-based systems to create dynamic profiles of individual behavior, enabling more accurate detection of fraudulent activities.

Real-Time Fraud Detection: Where We Are and Where We Need to Be

The current state of fraud detection is a mix of traditional rules-based systems and emerging AI-driven technologies. While rules-based systems provide a basic level of protection, they are limited in their ability to adapt to new fraud tactics. Featurespace’s adaptive behavioral analytics offers a more robust solution by continuously learning and evolving.


The fight against fraud is an ongoing battle that requires continuous innovation and collaboration. Featurespace’s adaptive behavioral analytics represents a significant advancement in fraud detection technology, offering real-time, self-learning models that adapt to evolving threats. By adopting these technologies and fostering collaboration between central banks, networks, and commercial institutions, we can create a more secure financial ecosystem.

Stay tuned for more insights and discussions on the future of fraud detection and the role of AI in finance and more!


Share with your favourite platforms!

Disclaimer: While we embrace open dialogue and value diverse perspectives, it’s important to note that the views expressed by individuals in our podcast episodes are entirely their own. They may not necessarily align with the views, opinions, or positions of the organization they are associated with.