Anti-money laundering (AML) systems of the future will be based on artificial intelligence (AI) and machine learning and will be predictive, bringing down the cost of AML compliance by up to 80%, a financial services conference heard.
Through a combination of anonymised data sharing between financial institutions and an algorithm that looks for suspicious transactions based on machine learning, financial institutions are likely to become much more efficient at stamping out money laundering activities from the sector, panellists at the Sibos conference said.
“All of the supply chain applications in the future will be predictive in banking,” said Tom Siebel, chairman and chief executive of C3.ai. “AML will be predictive. These rules-based systems that all these systems are based on today are like 1979-level technology. When we place AI on top of these rules-based systems, we can reduce the cost of AML compliance probably by 80%.”
Many banks are cautiously looking at ways of making their AML compliance processes more efficient through the use of technology. At Deutsche Bank, most of the current work involves machine learning and processes such as natural language processing (NLP) for tasks ranging from anomaly detection and pattern recognition to document recognition, said Rafael Otero, chief information officer and chief product officer at Deutsche Bank.
“We have a lot of use cases internally,” Otero said. “We do a lot of stuff actually in document recognition; how can we digitalise the documents and the document flows that we have?”
Natural language processing technology would, for example, be used in trade finance to get context out of documents and better understand their intent, he said. The same technique could also apply to AML.
“There’s lots of semantics that we can extract from speech, from texts, from social media, from CRM systems and from cable instructions, where we can get real meaning that informs the anti-financial crime processes and AML processes,” he said.
C3.ai is involved with a number of large financial institutions on projects to apply AI to processes such as AML to improve efficiency, Siebel said. To drive down costs and improve efficiency across the whole sector, firms should work through international organisations such as SWIFT to find ways to share data anonymously, he said. That could feed into a common AML algorithm without giving away personal data or competitive advantages, he said.
“It is in everybody’s best interest to collaborate, and there are no privacy issues,” he said. “Where we can develop machine-learning models that are very accurate at predicting money laundering, there is no reason why we shouldn’t publish that machine-learning model and make it available to the next firm. There are no privacy issues there. We’re not dealing with any identifiable information. It’s just a machine-learning model. And it’s not a place where one bank might have competitive advantage over another.”
“It’s in everybody’s best interest that any money laundering events be minimised. They don’t just do it at Deutsche Bank or UBS or Bank of America. There are bad actors in rogue nations that play this game, and they ring every doorbell.”
A machine-learning model would not need to contain any information about a customer, he said. An AML machine-learning model could be developed on Deutsche Bank’s data, further refined on UBS’ data, and so forth, without anybody’s data being visible to anyone, Siebel said.
“The model itself becomes very precise at identifying whether a transaction or a series of transactions and activities in fact constitute an AML event,” he said.
About the author
Trond Vagen is a senior editor, European financial markets regulation, at Thomson Reuters Regulatory Intelligence.