Tech companies are ramping up their artificial intelligence-based software for anti-money-laundering work and crime detection. The latest example is Oracle, which is rolling out AI agents for financial crime investigations.
Banks have used AI in financial crime investigations for years, to help gather research and sift through thousands or millions of alerts, separating the actual indicators of crime from the false positives, and to help find sketchy behavior that might at first look normal.
Traditional AI systems for AML and fraud and financial crime detection, from providers like ThetaRay, Quantexa, Hawk, Feedzai, Nice Actimize, SymphonyAI, ComplyAdvantage and Oracle, have begun making a difference in this line of work. In a June 2024 blog post, Jennifer Calvery, group head of financial crime at HSBC, said the bank is using AI to check 1.35 billion transactions a month for signs of financial crime across 40 million customer accounts.
"As new financial crime tactics or trends emerge, we teach our AI what to look out for," Calvery said. "As a result, we're able to find and tackle financial crime faster and more thoroughly than ever before."
HSBC co-developed the system with Google and began using it in 2021. According to Calvery, the bank is now finding two to four times more financial crime than it did previously, with greater accuracy.
"Historically, we had a high number of false positives, meaning that we were calling customers unnecessarily to ask them about what turned out to be completely legitimate activity," Calvery said. The rate of false positives has dropped 60%, she said.
Where agentic AI fits in
Agentic AI goes a step further than traditional AI. In addition to finding and analyzing data, it can execute follow-up workflows, make recommendations and draft suspicious activity reports.
The agentic AI the technology giant is adding to its Investigation Hub Cloud Service on Thursday contains workflows that do things like collect evidence, according to Jason Somrak, head of financial crime products for Oracle Financial Services. Before joining Oracle, he worked in anti-financial crime roles at KeyBank and PNC Bank.
"It's all about, how do we help ease the time of investigation, at least for financial crime, or make decisions?" he said.
In a typical financial crime or AML investigation, a human investigator will take the suspicious activity alerts generated by an AML or fraud detection software program and investigate the riskiest-seeming transactions. For instance, if a customer sent a high-risk wire to a high-risk country, the investigator might look into the source of the funds, the recipient and other relevant facts.
"They're basically trying to figure out who, what, when, where and why," Somrak said. "In doing that, they have to go and look at all their internal systems. They have to follow their procedure. They type up a long narrative explaining what triggered the alert, and then what the evidence tells them about it, and then ultimately, they make a decision on it."
AI agents can automate much of this. They can query internal and external data sources, collect evidence, follow an investigation plan, recommend an investigative decision and write a complete narrative of what they've done and why they feel the transaction is normal or abnormal.
Where humans fit in
The abnormal transactions get put in a case management file for human review. If the human expert thinks the AI agent missed something, they can kick it back and have the agent layer take additional steps, and then the human ultimately confirms or denies the AI agent's recommended decision.
"The largest banks have 4,000 or 6,000 investigators doing this work," Somrak said. "We believe that with this agentic workflow, you can automate at least 80% of that work. We think long term — it might take five to 10 years for banks to get there — we think we have the ability to shift the industry from trying to throw a bunch of bodies at this work to try to make sense of which of these events are good, and you turn the humans into labelers for a decision engine."
Financial institutions should take a phased approach to adopting AI agents into their regulatory processes, according to Michael Shearer, chief solution officer at AI-based financial crime software provider Hawk.
"In addition to looking for consistent and explainable processes, they should remain focused on effectiveness, harnessing this new technology to optimize the time of their skilled investigators in their pursuit of genuine financial crime risk and not simply further automate the processing of volumes of ineffective alerts from legacy platforms," he said.
