Discai and KBC Group featured in CFO Magazine

Follow the money: KBC brings AI into the fight against money laundering

This article was originally published in Dutch in CFO Magazine.

Risk management and compliance are high on the agenda of every CFO and CRO. But in the financial sector, both terms carry particular weight. Banks must comply with increasingly stringent regulation, including measures designed to combat money laundering and terrorist financing. At the same time, they are facing criminals who are becoming ever more professional and sophisticated. In the fight against financial crime, AI offers an additional tool that can help make a real difference.

Read this article if you want to know:

• How artificial intelligence can help tackle money laundering.
• Why anti-money laundering systems generate so many false positives.
• How AI regulation affects transparency and freedom to innovate.

Money laundering remains a major problem. The National Bank estimates Belgium’s shadow economy at around €27 billion a year, equivalent to roughly 3.8% of GDP. Yet it is extremely difficult to determine how much money is actually involved. Broadly speaking, authorities are believed to detect only around 10% of all laundered money. Of that total, barely 1% is recovered. In short, it is a major problem and an extremely difficult one to tackle.

“Criminal behaviour is increasingly digital and therefore increasingly hidden,” says Fabrice Deprez, CEO of Discai. “Crime is becoming more professional and more sophisticated, partly through the use of new technology. Banks have to carry out controls to combat that crime, while often relying on traditional technology. That creates an increasing workload that is becoming unsustainable.”

Europe tightens the screws

The workload Fabrice Deprez refers to is partly driven by AML regulation, short for Anti-Money Laundering. It requires banks to monitor transactions, identify suspicious activity and report it to the relevant enforcement authorities.

Europe is stepping up the pressure and recently established a new EU agency, AMLA, the Anti-Money Laundering Authority, to harmonise and strengthen AML supervision across the European Union. In recent years, several European banks have been fined for failing to comply fully with anti-money laundering rules. And the consequences go beyond financial penalties. Organisations that fall short on compliance often suffer reputational damage as well.

Banks that do invest heavily in compliance must find the right balance between additional staff, more and better data, and new technology. KBC recognised that challenge and started developing an AI solution in-house, a project that ultimately led to the creation of its subsidiary Discai.

“The traditional way of working requires a lot of time and people,” says Fabrice Deprez. “In principle, banks have to investigate every transaction flagged as atypical and determine whether it may be suspicious.”

Focusing on behaviour

Traditional rule-based technologies screen customer transactions using filters built around predefined scenarios, such as transaction amount, transaction type, timing and location. This process identifies transactions that appear “atypical”.

In practice, however, it also produces a very large number of transactions that turn out to be legitimate, known as false positives, simply because the scenarios being used are generally fairly broad. Typically, only around 5% to 20% of the alerts generated turn out to involve suspicious transactions.

But the other potential money-laundering cases, which ultimately prove to be legitimate, still have to be investigated thoroughly by bank employees. That takes time.

Discai’s starting point is that AI can make those controls both more efficient and more effective. “Follow the money” is the guiding principle.

“We developed technology that looks not so much at the transaction itself, but at the behaviour and full profile of the individual. That means looking at the entirety of their transactions as well as KYC data, or Know Your Customer data, such as identity, customer and risk information,” Fabrice Deprez explains.

“That allows us to filter the overall workload, namely all transactions that need to be monitored, much more effectively and therefore process it more efficiently.”

Where a traditional anti-money laundering approach may fail to detect unusual behaviour, Discai is designed to identify it.

People still matter

A crucial element is Discai’s ability to show exactly how the technology works and how it arrives at a particular recommendation. After all, the technology is being used in a highly regulated environment, and regulators require transparency when AI is involved.

“Look, building a machine learning solution in itself is not that difficult,” says Fabrice Deprez. “But being able to explain, in black and white, exactly what is happening behind the scenes is much more complex. And in the context of AML, that explainability is absolutely crucial.”

Discai grew out of a bank, so the importance of explainability was understood from the outset.

“It is not enough to identify who is suspicious. You also need to be able to explain why.”

At the same time, new regulation, such as the AI Act, is creating a stricter framework.

“That is a positive development,” says Fabrice Deprez. “Even if it limits our freedom to develop technology to some extent, it also raises the overall level of safety.”

“The idea that AI can simply solve everything and immediately deliver major cost and FTE savings is simply not true.” – Michael Wittenburg, Chief Compliance Officer at KBC Group

Transparency is central to that approach. When Discai filters suspicious cases from the wider pool of activity, the process is never entirely automated.

“There is always a human in the loop.”

At the same time, the company has continued to optimise its models through greater automation and standardisation, meaning they now operate significantly faster on average than they did at the start.

The search for efficiency

Discai’s story illustrates the pressure banks are under. They have to protect themselves against money laundering while at the same time responding to additional regulatory requirements. Balancing those two demands remains difficult.

“If you are responsible for compliance, you constantly live with the concern that you may be missing certain risks or signals,” says Michael Wittenburg, Chief Compliance Officer at KBC Group.

“You cannot afford to overlook attempts at money laundering, but at the same time you do not want to be overwhelmed by false positives.”

As mentioned earlier, traditional scenario-based and rule-based systems often prove relatively ineffective in practice. They tend to rely heavily on expert knowledge, are less efficient at processing large volumes of data and are not adapted quickly enough to new risks and changing behavioural patterns.

It is estimated that 90% or more of money laundering activity remains undetected.

“It is an extremely complex issue. In a world of globalised, internationally interconnected payment flows, a single bank ultimately represents only one small part of the picture,” says Michael Wittenburg.

“But what we can do as a bank is improve our effectiveness and efficiency.”

That is where new technology such as AI can play an important role.

Compliance is a demanding field, including for the professionals working in it. When employees carry out anti-money laundering checks and are continuously confronted with large volumes of false positives, the work can become frustrating.

“New technology that supports us can bring the best of both worlds: employee satisfaction as well as regulator satisfaction.”

Moving away from a rule-based approach, however, is far from straightforward.

“Before you can do that, you need the right data and the right technology, combined with sufficient transparency.”

Step by step

That created a new challenge for KBC. The bank not only needed to identify the right data, but also develop a deeper understanding of the work carried out by anti-money laundering investigators, determine which parts of that work could be automated and, at the same time, take regulatory requirements into account.

“That is how we got started,” says Michael Wittenburg.

“It was certainly not some ‘new dawn’, as the marketing machine often likes to portray it. Using AI does not suddenly transform AML overnight, once and for all.”

But there was still something of an “aha” moment.

“For us, but also for the regulators. Supervisors were very familiar with the rule-based approach, but they still needed to build the necessary knowledge and skills around AI.”

Based on those first findings, KBC decided not to abandon its rule-based solution, even though that had initially been the intention.

“We realised it was not a question of either-or, but of combining both. If you model and validate everything correctly, AI allows you to detect genuinely suspicious behaviour in a more efficient and effective way.”

At the same time, AI is not a miracle solution.

“Even today, we still have false positives. The idea that AI can simply solve everything and immediately deliver major cost and FTE savings is simply not true. What it does do is help us monitor transactions more effectively and efficiently, while continuing to identify new and previously unknown money-laundering techniques.”

Since introducing AI into AML, KBC has been able to focus more on relevant alerts. The savings generated are reinvested in additional controls for activities that carry a money-laundering risk.

This allows the bank to uncover more cases of money laundering.

Realistic expectations

Michael Wittenburg believes it is important to be precise about what AI can and cannot do.

“We have also had to adjust our expectations,” he says. “We have become more realistic. AI absolutely offers a more efficient and effective approach, but we still have to operate within our risk appetite. Our objective is not to reduce the total number of alerts, but to reduce the number of false positives.”

That allows the bank to focus its attention more effectively on transactions that genuinely present a problem.

“But as I said, that requires a different way of working. Change and transformation management become an important part of the process. That is essential if you want to succeed.”

Compliance teams must also adapt their strategy, both by attracting new profiles and by further developing the skills of existing employees.

“In a pilot environment, an AI application may appear to work perfectly,” says Michael Wittenburg. “But in practice, conditions are never perfect.”

“Building a machine learning solution in itself is not that difficult. But being able to explain, in black and white, exactly what is happening behind the scenes is much more complex. And in the context of AML, that explainability is absolutely crucial.” – Fabrice Deprez, CEO Discai

A holistic approach

Step by step, KBC now intends to expand its use of AI further.

“The intention is not for this to remain a one-off initiative.”

KBC plans to broaden its approach so that AI can be used in a more holistic way, rather than only at the level of the individual transaction, and can also be applied to other types of fraud or abuse.

“But we are aware that this cannot happen all at once. We are also not limiting ourselves to AI alone. We are looking at other solutions as well in order to achieve better results. And yes, technology can help us do that, but it cannot replace people.”