Signals-Based Approach Raises Bar on AML Transaction Monitoring
With David Griffiths, Director of Regulatory Affairs, Eventus
How has the current alert-based approach to AML transaction monitoring evolved, and what are its limitations?
It hasn’t changed that much since Anti-Money Laundering first became important around 2002, when the Patriot Act required banks to heavily invest on AML programs. In the market today, people still talk about Know Your Customer (KYC), having a risk score on the customer, and putting that into an alert that says, “If X is greater than Y, then alert.”…this is transaction monitoring today.
It’s fairly procedural. A bank can have hundreds of thousands of alerts generated per month, triggered by a transaction from a high-risk jurisdiction or individual perhaps, or a third transaction just below $10,000. You need an army of people to trawl through these alerts. For example, a bank’s surveillance tool might provide an alert for each transaction in a high-risk jurisdiction, or for an individual that requires heightened supervision, or if a transaction is the third transaction just below $10,000. Today some firms use machine learning to cut down on the noise of the alerts, but it’s still the alert with no context that is itself the problem, because when you generate an alert, it must be investigated – no ifs, ands or buts.
And because banks have different businesses – retail banking, correspondent banking, private banking, all with different business flows, trade flows, and transaction flows – it’s difficult for traditional transaction monitoring alert logic to amend parameter logic to account for the differences in those flows.
What is a signals-based approach to AML transaction monitoring?
AML transaction monitoring is a complex discipline, dependent on the processing and filtering of masses of data, which themselves are constantly evolving and increasing in step with regulators’ demands and the methodology of bad actors. A signals-based approach leverages technology to enable customization and tailoring of alerts, which in turn meets the market’s longstanding need for greater agility and efficiency in AML transaction monitoring.
Through a clean-sheet approach to deployment and service, firms can achieve efficiency, scale and flexibility – placing the control, without additional burden, back into the hands of the monitoring team as they face today’s challenges and look forward to the challenges to come.
What are the advantages of a signals-based approach?
A signals-based approach allows you to be more specific. With this technology-based approach, if somebody’s done a third transaction that is just below the radar, you can choose whether you want to treat that as its own alert, or whether you want to combine that into a macro alert, generated by multiple signals specific to that location or client type. With this approach, you have the flexibility to add context from reference items, like external customer risk attributes, but the signal itself is also an indicator of customer risk, which also improves KYC.
Standard alerts can be slow and inefficient. You can add AI or machine learning to reduce noise and false positives in those alerts, but AI and machine learning require vigilance in training and re-training the models. This becomes even more complex in a world where banks and business flows can be very different. A signals-based approach is a paradigm shift in how people are managing AML transaction monitoring, i.e., through a system that can take multiple inputs and produce one single alert.
Signals are smarter and more efficient because the approach is more specific and more tailored to your business than a generic risk model. A signals-based approach enables a higher level of risk capture at the transaction level based on trade flows, as compared to just taking in a KYC risk score that somebody else has done.
What are some specific applications of a signals-based approach?
At Eventus we do a lot of work with crypto firms. Some firms will generate a risk score for a particular wallet, but what doesn’t exist at the moment is that feedback loop where the transactions, the trading activity, feeds back into that score. A fundamental aspect of AML and KYC is enhanced due diligence, where customer risk is continuously reassessed maybe once or twice per year – you can’t really do it more frequently due to the volume of alert data. But by being able to generate a signal, in addition to improving the alert generation, those signals can improve enhanced due diligence in a timelier manner.
For the past 20 years, there has been a standard list of alerts that everybody runs. Being able to create your own risk type or alert type gives much more flexibility based on specific types of clients, firms, and banks. That flexibility will lead to improved alert generation, better risk models, and improved turnaround time and conversion rates.
Is a signals-based approach more of a ‘tomorrow’ story?
People have been wanting a signals-based approach for years.
The whole point of this paradigm change is to change the way you are assessing risk. It won’t happen overnight, because an AML cycle within an institution is typically around five years. But within seven to 10 years’ time, I expect a signals-based approach will be much more widespread.
What is the outlook for adoption in Asia specifically?
‘Explainability’ is key. When you start talking about artificial intelligence and machine learning, people think it’s a black box – compliance officers will look to ensure they understand how it works.
A signals-based approach to AML and KYC is not a new type of AI or machine learning. It’s a different way of thinking about how you reach the end goal of identifying your risk. There’s nothing ‘funky’ about it. We’re not changing KYC – we’re improving the transaction monitoring capability by enabling you to build your own alerts.