Verafin launches genAI copilot for fincrime investigators

Verafin, Nasdaq’s anti-financial crime subsidiary, is launching a generative AI-based copilot to help improve investigator efficiency in researching and identifying financial crime. The product, dubbed the Entity Research Copilot, will be integrated directly into Verafin’s existing anti-money laundering (AML) solution, at no additional cost to users.

Rob Norris, VP and head of product strategy at Verafin, tells WatersTechnology that the bulk of work around fighting financial crime and providing AML measures is repetitive and requires considerable manual research. He explains that until generative AI burst onto the scene in Q4 of 2022, there were few options to increase user efficiency in this area. But as soon as ChatGPT started making waves, Verafin began experimenting with genAI-driven solutions in Q1 of 2023.

“We’re big believers in the idea that you have to start with a customer problem, instead of starting with a technology and try to fit it to users,” Norris says. “So we actually had some interesting things that we wanted to solve for years that [generative AI] did unlock.”

Norris explains that the very first feature that the team at Verafin automated was a research tool built into the Entity Research Copilot that can quickly find and collate many sources used for investigative purposes after the trigger of a bank alert. He says that this task, which typically took minutes, was reduced to being completed with a click of a button. The repetitiveness of this task, which often involved reading articles on the affected party, listings on corporate registries, and details accrued through search engines, has been simplified significantly.

“It does a search—not through a browser, but through a secure API-based search—and then we’re feeding the results into a large language model, and we have specialized prompting that we do on behalf of the user,” Norris says. The LLM that Verafin is using is a commercial foundational model, and while the genAI component of the large language model has not been created or trained in-house, it is fine-tuned by the team at Verafin.

“The prompt has been very carefully crafted over months to reduce any errors,” Norris says. “We use something called retrieval augmented generation to not use the model’s past training but to always feed it the articles that it’s operating on to reduce errors.”

While the copilot is Verafin’s first venture into genAI-based models, Nasdaq has been experimenting with the technology across other areas of its business. At the company’s Investor Day in March, Nasdaq CEO Adena Friedman said that generative AI “is a whole new opportunity for us.”

Also at the Investor Day was Hazel Dalton, Nasdaq’s CTO of financial crime management technology, who talked about genAI applications within Verafin, and referenced similar use cases to those mentioned by Norris.

“We’re leaning into generative AI,” she said. “We’re using large language models to reduce time spent on research and documentation, and believe that will have an attractive ROI.”

Verafin’s copilot is coming hot on the heels of other firms making similar moves in the genAI space. Last week, at a conference hosted by Bloomberg in London, Man Group’s CTO Gary Collier detailed use cases for genAI at the hedge fund, including its own copilot, Alpha Assistant. In BlackRock’s Q4 earnings call, CEO Larry Fink said the company was internally building its AI copilots. At the same time, Microsoft’s corporate vice-president of worldwide financial services, Bill Borden, told WatersTechnology in January that the tech behemoth is looking to open up its tech stack to allow users to more efficiently make their own copilot models.

Greg Slayton, director at ACA Group’s cybersecurity specialist division ACA Aponix, says that testing genAI models within large organizations with proven track records in security and competence is a good way of “testing the waters” with the new technology, considering its rapid rise over the last year.

“AI technology and LLMs have come out in the past year—they’re blazing fast and there’s been an adoption [speed] that’s incredible, maybe faster than any technology that we’ve seen so far,” he says. “We’re seeing things that have traditionally been called AI in the background like machine learning and predictive analytics take the forefront being involved in these copilot models, which help people be not just faster but more accurate.”

Slayton notes that as the use cases for the technology are myriad, so too are the potential pitfalls. Efficient and safe data regulation and governance in the creation of large language models is important, especially with the compliance and regulatory issues thrown up by the industry’s rapid adoption of emerging technologies.

While Verafin’s Entity Research Copilot will not be used for increasing quant efficiency at investment firms or re-tuning code at banks, Slayton thinks using the copilot model in an anti-financial crime context makes sense.

“It’s exciting if it works, and if it works well, it’s great,” he says. “We know that bad actors are doing bad things, and they’re going to continue to do bad things and they’re going to use these AI technologies to do bad things. We’ll be behind the curve if we’re not using these technologies to prevent that.”