Anti-money laundering can benefit from AI—to a certain extent
Experts say that organizations should not expect to switch completely from a rules-based system to AI-driven AML detection (Getty Images/10’000 Hours)
When her financial institution introduces a new product, Marie-Andree Malo-Mongeau and her team at CIBC get to work trying to think like money launderers in order to figure out how criminals might try to game the system. “What are we trying to catch here? What’s the criminal journey?”
Malo-Mongeau, a CPA and the former senior director of the bank’s Enterprise Anti-Money Laundering (EAML) unit, was not only tasked with stopping the flow of dubious funds, but she has also spearheaded an effort within her bank to implement artificial intelligence/machine learning tools to either augment or replace more traditional rule-based screening systems.
It’s a sensitive task: bank regulators and internal controllers are accustomed to using rule-based systems to flag potential risks. While there’s been a lot of talk about the potential benefits of developing algorithms to detect anomalous funds’ movement patterns that might elude more conventional warning triggers, the reality, Malo-Mongeau cautions, is more complicated. “The deployment of AI and machine learning requires a lot of effort,” she says. “It requires a lot of expertise. It requires a lot of funding, infrastructure and so on."
Like constructing a building, the foundation is the critical first step. In the world of AML, that means investing the funds to clean up the mountains of data that’s accumulated in various parts of a financial institution including—increasingly—know-your-client information. Only when the data is machine readable and relatively free of glitches can a financial institution’s AML team begin to deploy algorithms designed to identify suspicious transaction patterns.
Organizations should not expect to switch completely from a rules-based system to AI-driven AML detection. “We've deployed hybrid models, which are a combination,” says Drew Galow, managing director, anti-money laundering (AML) model management, analytics and machine learning at BMO Financial Group. “At my institution, for AML transaction monitoring, we have transferred about 50 to 60 per cent of our existing rule-based monitoring to machine learning or advanced analytics.”
He says at BMO, the conversion has increased the productivity of its AML systems to about 25 to 35 per cent. But he cautions that financial institutions or money services businesses shouldn't assume that those kinds of metrics would automatically inform regulators. Galow says his group meets quarterly with regulators to walk them through the technical steps they're taking. "[Implementing a hybrid model] was a long journey and having [regulators] prepared was better than not."
STAY UPDATED
Get insights on the Cullen report, and the potential impact of its recommendations. Plus, hear from anti-corruption expert José Hernandez about what CPAs can do to fight corruption and money laundering.