What if a bank’s compliance software could relay messaging in natural language?

That’s the question anti-money laundering fintech Lucinity is hoping to answer with Luci, its new generative AI-powered co–pilot for financial institutions. The tool was built in collaboration with Microsoft Azure’s OpenAI and uses natural language understanding to turn data into easy-to-understand insights and automate labor-intensive tasks, according to a Lucinity release.  

“This is more than just an OpenAI integration,” Theresa Bercich, vice president of product at Lucinity, told Bank Automation News. “It actually is a whole system that sits between what the user does and what OpenAI then returns.” 

Luci has the power to automate tasks currently performed by analysts, using AI to comb through data in search of suspicious patterns and to present those patterns in an intuitive format, according to Bercich. 

“This is where, really, the productivity boost comes in,” she said. “You actually have, in context, productivity that just really, genuinely reduces review times from hours to minutes.” 

After detecting suspicious behavior, Luci can use ChatGPT-style natural language processing to automate the time-consuming generation of regulatory reports, Bercich added. 

The Iceland-based company is conducting customer trials with Visa’s Currencycloud and Iceland-based Arion Bank, according to its website. 

“The Lucinity platform has already made a significant difference in our [anti-money laundering] compliance efforts, and we are excited about the impact Luci is poised to have on our operations, particularly in enhancing analyst consistency,” Arion Bank Chief Compliance Officer Andres Fjeldsted said in a statement to BAN.

Behavioral modeling

Luci analyzes consumer behavior to detect suspicious patterns like rapid movement of funds — a common tactic used to obfuscate the source of laundered money — and then score the transaction based on how likely it is to be part of a financial crime, Bercich said.

This behavioral modeling mirrors the tactics used by anti-fraud fintechs like Sardine and Sift, which use AI to flag suspicious transactions in real time.

Bercich said she believes these two parallel industries “will eventually collide,” citing a recent incident in which criminals used a deepfake to imitate a British chief executive and transfer hundreds of thousands of pounds into a Hungarian bank account as an example of the crossover between fraud and money laundering.

Human touch

In these high-stakes situations, accuracy is paramount. That’s why a person will make the decision on whether to escalate a transaction to a suspicious activity report and pass it along to a regulator, Bercich said. 

Beyond keeping Luci in a secondary role, Lucinity has taken additional steps to decrease the likelihood of a hallucination, or a response made by a chatbot that is not justified by its training, according to Bercich. These include turning down the “temperature” — or potential for creativity — of the chatbot and reducing the scope of the encounter to the immediate context of the transactions.

“It does not need to hallucinate because the questions, the prompts that it can get, are made so that it has the data that it needs,” she said.

Though Lucinity does not yet have its own proprietary large language model like some of its generative AI-powered peers in other verticals, it plans to release one within 12 months, according to Bercich. 

The company plans to roll out its co-pilot to the public in the third quarter.