• Saeed Patel, Group Director at Eastnets


• 09.03.2023 02:45 pm
#ai #data

The first ever piece of anti-money laundering legislation came into effect in 1970 when the US introduced the Bank Secrecy Act (BSA). Since then, financial institutions have been under pressure 

to stop those who aim to benefit financially from wrongdoing.  

In their bid to comply, traditional banks, challenger banks, insurers and more, have sought to spot money laundering and financial crime using traditional rules-based systems. These have checklists of conditions with pre-defined thresholds that when detected flag a payment as being potentially criminal.  

The rules are often based on criteria such as customer account profile, location, frequency or the recipient of a payment. For example, if someone living in the UK suddenly has large payments being made on the ground in Uruguay. Or if multiple payments happen in quick succession or to lots of new bank accounts. 

These rules have been relied on for decades and have become more complex and effective as time has passed. What’s more, they can work at speed, are simple to use and offer transparency. This last point is important in terms of human intervention when checking a flagged payment or when regulators need to audit a system.  

However, there’s major inherent flaw in this approach: financial criminals don’t play by the rules. As soon as a new one’s written, often in response to an emerging fraud technique, it’s circumvented. This means rules-based systems are always one step behind and out manoeuvred. 

In addition, the strength of the rules and effectiveness of the solution are often dependent on the skill of the person setting them. Incorrect or badly defined rules and thresholds lead to low quality of fraud detections. Alternatively, and perhaps more commonly, teams become over-cautious, leading to huge volumes of false positives. 

One of the consequences that banks face with inefficient rules-based sanctions screening and anti-money laundering (AML) transaction monitoring solutions is that they’ve had to employ an army of compliance people to investigate the high volume of alerts. In a market where compliance staff are in high demand, this is not a smart use of a valuable and scarce resource.     

Finding a solution 

This begs the question, what’s the alternative? How can financial institutions move from the traditional, yet outdated rules-based model? The answer lies in artificial intelligence (AI) and machine learning (ML).  

This elevates AML and anti-fraud software from merely a tick-box, rather binary approach to a far more nuanced and considered one. Instead of a payment merely being flagged as potential fraud, AI and ML can create risk scores and find anomalous behaviours as they happen. This reduces the number of false positives and need for human intervention.   

The technology can even predict potential threats before they happen, putting security and compliance teams one step ahead of the criminals. It achieves this by poring over vast amounts of data to predict fraudulent transactions in the SWIFT payment environment. 

It’s also incredibly efficient. Threat modelling powered by AI works 24/7, without downtime or the need for human intervention. This boosts productivity and results for fraud and compliance teams.  

The road to adoption 

Despite the compelling case for AI and ML, and pressure from regulators to adopt the technology, data from consultancy firm KPMG suggests only 57 per cent of organisations are using or plan to use the software to combat financial crimeⁱ. A sizeable minority of financial institutions are lagging.  

This is because there remain significant barriers. Cost is always a challenge, but more important is a lack of skills. Traditional rules-based systems have been managed by compliance teams with a legal background rather than people with an understanding of AI, ML and data analytics.  

This needs to change. While legal and compliance professionals will still be needed, it’s vital that financial institutions build multi-faceted teams adding data scientists, data analysts, forensic accountants and professionals with a programming background. Without this, it will be nearly impossible to take full advantage of the opportunity afforded by the new technology. 

It’s also incumbent on vendors of these new systems to do everything in their power to make the software as simple to use as possible to alleviate the skills challenge. They need to get the balance right between offering complex data insight and usability.  

This means looking for ways to present data visually and in a meaningful way. They need to marry mathematical and statistical expertise with the ability to show why certain data matters. Using graph technologies to find relationships in data through smart visualisations, providing link analysis is an important development that vendors must embrace. 

They also need to work with modern open source programming languages such as Python, which are excellent for AI modelling and simulations. The AI models and ML code are typically proprietary, which means a vendor will be a step ahead of sophisticated tech-savvy criminals. 

Managing the transition 

If financial institutions can build the right teams and work with the best-equipped vendors, the next step is transitioning from rules-based technology to AI and ML. This usually requires a move to the cloud.  

This is because the computing power required by AI and ML can be huge. Trying to use older on-premises infrastructure is likely to cause issues. However, there may be the option of a phased approach where existing systems are combined with cloud-based vendors’ AI. This will begin to illustrate the value of AI and help get buy-in from decision-makers prior to a full transition. 

To further help smooth the path, it’s worth financial institutions considering the cloud to securely share large volumes of confidential data with vendors so they can build a proof-of-concept efficiently and illustrate just how much value can be gained from a move to AI and ML.  

What the future holds 

It’s clear that there are challenges when it comes to adopting this new software, but it will become easier as the technology develops. In fact, some vendors are going a step further to combat financial crime by considering the use of conversational AI in their systems.  

This is when the software can interact with the team using it. Imagine a chat bot on a website or app. It can ask questions, make suggestions and help guide a shopper through an issue. The same principle is beginning to be applied to complex AML and anti-fraud software, with AI making suggestions. For example, “Have you looked at this data?” Or, “Would you like to consider this factor when making your decision about a potential case of fraud?”  

This could be a particularly useful development given the skills needed by financial institutions are in short supply and fraud teams may need a helping hand when understanding how to make the most of the new software they’re using.  

Reaping benefits 

If financial institutions can become wise to the benefits of AI and ML, develop the skills they need to deploy this type of software and then manage the transition effectively, they’ll be able to reap the benefits.  

Because we live in a world where AI is increasingly ubiquitous. It’s in our phones, used by the apps and websites we visit and deployed by organisations we interact with daily.   

Rules-based solutions are no longer fit for purpose. Criminals are, by definition, rule-breakers. It might not be easy, but only AI, ML and data analytics can stop financial criminals.