Koho rolls out AI tool to help investigate fraud and scams

Canadian FinTech startup Koho is using a generative artificial intelligence (AI) tool it claims will speed up its investigations of money laundering and terrorism funding.

The company says its internal tool now performs the “busy work” of anti-money laundering investigations and reduces the time analysts spend on an investigation from an average of 95 minutes to 35. 

“We saw an opportunity to build something that would make this more efficient—not to take the human out of the role of detecting and reporting.”

David Kormushoff, vice-president of technology and AI at Koho, said during a panel at Amazon Web Services’ Re:Invent conference in Las Vegas that the company was on the precipice of getting that investigation time down to five minutes, which would involve having the investigator check the tool’s work (BetaKit received travel and accommodation support from Amazon to attend Re:Invent).

“We saw an opportunity to build something that would make this more efficient—not to take the human out of the role of detecting and reporting, but effectively to have [the tool] be their assistant,” he said. 

Koho is regulated as a money service business and overseen by the Financial Transactions and Reports Analysis Centre of Canada (FINTRAC), which requires the challenger bank to detect and report suspected instances of money laundering and terrorist financing. 

But he noted during the panel that not only have bad actors become more sophisticated, finding ways to layer transactions and jump across institutions, the sheer volume of transactions that occur across financial institutions every day means the process of detecting and reporting on suspicious activity is very slow. 

“Most risk and fraud departments and financial institutions are seriously backlogged,” he said.

He said that Koho’s primary demographic, of paycheque-to-paycheque households, are “disproportionately affected” by bad actors. 

“And grandmas who get calls saying, ‘go to the store and buy a bunch of gift cards’—that’s not just fraud, it’s money laundering,” he added.

When fraud investigators receive an alert of suspicious activity, they need to dig into every transaction the individual made over the past year, identify indications of money laundering or terrorism financing, scour the open web to determine whether the individual has been reported, and then summarize those findings into a report for FINTRAC. 

Koho’s internal tool now does all that work, and has been built to cite where it found all of the information included in the report. 

“The analyst is not just assuming the AI is doing the right thing; it’s their job to spot check and make sure that job is good,” Kormushoff said.

Kormushoff told BetaKit that the tool is not trained on customer data, and is only given information on a per-investigation basis.

The company claims the tool is freeing Koho’s fraud investigators to be able to do more “high-thought” and proactive work, he said in the interview—such as figuring out how to detect and stop money laundering at the account opening phase.

The company built the system to be “very flexible” to adapt to any forthcoming regulations around the use of AI, with safeguards around customer data in place, Kormushoff said. Koho built its tool on top of Amazon Web Services, which he said keeps customers’ data within the company’s systems.

RELATED: Koho secures $190 million CAD in equity and debt to bolster banking licence efforts

Koho CEO Daniel Eberhard wrote on LinkedIn in October that the company has landed other “major use cases” for AI in credit, user success, and growth. 

“Our view is that every function is becoming technical but the hurdle rate to being technical is going to zero,” he wrote.

Kormushoff told BetaKit one application he’s particularly excited about is an internal tool that analyzes all the available information about a security alert to help analysts more quickly determine whether the alert is a genuine concern or a false alarm. 

The company is also starting to look at creating customer-facing generative AI applications, particularly focused on educating customers about how to deploy their capital, build their wealth, and get insights into their spending and alerts, if their spending trajectory indicates they might come up short at the end of the month. 

But Kormushoff said the company isn’t yet ready to roll out tools that interact with customers’ personal financial data. 

“The rate of change in the evolution and development of these tools is only increasing, so I think we’ll probably get to a point where we feel confident enough to put it in front of the customer,” he said. “But we haven’t today because … we don’t want to be lying to them or giving them bad information.” 

Images courtesy Koho.