How can banks best apply technology to help with their financial crime compliance processes, particularly when the rapid advancement in the space?
AML RightSource Creative Director Elliot Berman sat with CEO Frank Ewing and Head of Product David Buxton at AML RightSource to discuss exactly this issue.
Firstly, the difference between types of financial crime – namely fraud and money laundering – is often convoluted and needs to be identified to see where technology can specifically add value:
‘Money laundering cases have a big problem which is that you don’t really ever know if you’ve found one,’ David suggests. ‘Unless you happen to get the FBI involved and they knock on peoples’ doors and tell you that you’re right. It’s very difficult for you to do anything other than say, “based on what I know about the totality of the world, this doesn’t quite fit, they weren’t getting personalities involved that don’t work with the transactions, with the volumes, with the purchases”. It doesn’t make sense.
Whereas fraud is a different thing. You can build more on an outcomes-based basis rather than on the basis of applying a set of intuitive human criteria. So you end up with quite different systems to do that, and you end up with fraud, frankly, being an easier problem to solve because you can apply the wealth of different advancements in technology over the last well, 50 years – but broadly speaking more like five years in terms of machine learning and artificial intelligence – to let systems go wild and actually work out what the characteristics of a fraud are that are hidden in the data.
With money laundering, frankly, the sets are too small. If you try to apply that kind of outcome based logic, you have to be looking at smaller, more granular areas to make sense of it.’
Frank believes that no work in the financial crimes space can be achieved through technological advancement alone, as human intervention and judgement proves to be just as salient in investigations. Mainly, technology and human capability in tandem will improve processes:
‘I think we all can agree that there will always be an element of human capital involved in the work. But the work can be done better. If the goal of the work is actually to find and suss out nefarious characters and nefarious schemes and activity, well, certainly technology is probably the best avenue to accelerate that journey and to get better answers along that journey.
I look at technology as friend, not foe. And I think that’s something that we should all embrace. But like anything that is an accelerant or an advancement, whether it’s a medical drug or a procedure or anything, it’s a continuum and it’s not a panacea, you know? And so you have to view technology, in the sense that it’s going to make life better, but it’s not going to solve every problem.’
David then goes on to say that ‘You’ve got lots of different human actions – some of which are easily reducible to technical problems. So, a good example might be something like reducing the noise in a dataset. From a human point of view, you might be saying, look, I can tell for sure that these things are irrelevant, and I can train an algorithm to do that. And then there’s some things where actual human judgment is really important. Being able to really assess the context of the situation with reference to all sorts of different parameters, which might not be easily capturable by some sort of technical system.’
In terms of how technology has assisted financial crime compliance, David is not so sure whether the outcomes of financial crime have changed as successfully as the advancement of technology itself:
‘I started Arachnys, which is now part of the AMLRS family, 10 years ago or so. I guess that the way I feel about it, broadly speaking, is that although technology has really improved, at least if we’re talking about anti-money laundering, I don’t think that financial crime outcomes have really improved in that period. We as technologists have to look at ourselves in the mirror and say – what’s going on? I think at Arachnys, our solutions have been to say that it’s not just technology that’s the solution. It’s all about having a sensible way of embedding technology in a broader process.’
For financial institutions, solutions are providing difficult to come by due to amalgams of different systems, as Frank suggests:
‘The thing here, and this is no fault of banks or financial institutions, is that if you want machines to do the right thing, it’s kind of the old proviso of garbage in/garbage out. And banks largely are the amalgam of other banks. Ultimately what happens is they don’t really have all of the information to come up with precise information to come up with the right kind of inputs to have a good result. And so what you wind up with is having a lot of half-baked solutions that don’t really ever come to fruition because they’re not informed by good information.’
And so what solutions can banks implement to help with their financial crime compliance processes?
‘I think the future is in investment in actually getting good information and data into the system because that’s the only way you’re ever going to get good output,’ Frank suggests. ‘And then second, the world understands – especially in the AML fincrime space – to a bank, to an institution that the conversion yield on a set or series of algorithms is typically somewhere between like two and 7%, the notion is always to shrink those false positives. But, it’s more than that because there’s also false negatives and there’s also just things that you don’t have the background kind of data information to understand what you’re not seeing. Now you’re only looking at 10% of the iceberg and there’s a whole lot of stuff.’
Ultimately, handing off the more manual, basic techniques to the machines will give greater time and resources to the talented analysts to deal with these problems. David says:
‘When I think about the machine learning possibilities, I’m less enthused about the ability of machine learning to really replace human judgments about the salience of the specific point of a piece of information to a specific individual or entity. I’m much more enthused about the ability of computers to solve really quite generic problems. Those are reasonably well understood and well-formed problems that have application in multiple different domains.
Anomaly detection is one thing. I think there’s a lot of places in analyst roles where they’re doing road work which doesn’t add value, as part of the track towards an objective, which does add value. It’s a bit like the way that in order to climb Everest, you don’t just need to climb Everest. You’ve got to trek to base camp.
A lot of the actual time and money invested in analysts is actually spent with them not performing the role that they’re uniquely suited to, but instead doing essentially administrative tasks, you know, it’s that kind of preparatory work, the slice of expertise and human judgment, which they’re good at.
I think a lot of where technology can actually help is paradoxically in the stuff which is not particularly specific to financial crime or even hard, it’s just very basic trimmings around the side. And part of that is just stringing it all together and interfacing with the broader processes, the business.’
Interestingly, to conclude, Frank and David look into the (strangely positive) effect the covid-19 pandemic has had on anti-money laundering:
‘The one really interesting anecdote about that is that various customers have told me that the pandemic was actually one of the best AML environments of the past decade or so. It was like that Warren Buffett quote that only when the tide goes out that you find out who’s been swimming naked. It wasn’t until this broader macro shift that it was possible to see how behaviour changed.
So it may have actually created tons of new frauds and all sorts of things, but interestingly it created an environment in which anomaly detection was able to come into its own in a funny way.’
You can listen to the whole podcast over on the AML RightSource website.