Five takeaways from American Banker's AI Summit

October 17, 2023 9:00 PM

As artificial intelligence changes the banking industry, institutions — whether at the forefront of innovation or still in the exploration stages — are thinking about how the technology will impact them.

At American Banker's AI Summit on Monday, bankers, consultants and vendors shared how their institutions are moving forward with generative AI, fintech partnerships and security. AI is the technology that banks are most excited about to create competitive advantage, according to recent research from American Banker's parent company, Arizent.

Banks are investing in and investigating AI uses across functions, and regulators are beginning to apply more scrutiny as the technology emerges.

Industry leaders spoke about evaluating key AI uses, adopting emerging technologies and monitoring fraud from different angles across three panels.

Here are a few of the main points from Monday's summit:

Daniel Tepper/Bloomberg

Banks are testing many generative AI use cases

Many banks have been using some artificial intelligence for certain processes for years, but since ChatGPT hit the mainstream last year, some banks are working on ways to integrate the emerging technology.

Michelle Grimm, Fifth Third's senior director of conversational AI, said the bank is planning to start slow with generative AI, beginning internally. For example, Fifth Third is evaluating how generative AI can make job descriptions more effective in recruiting talent.

"One of the things that we've really been focused on is this idea that we need to almost crawl, walk, run," Grimm said. "So start with something small, let's test it out and prove it out. And one of the things that we're looking at is, 'What are some of those internal use cases that aren't necessarily customer facing that we can prove out the technology and what it can do for our employees?'"

Amir Madjlessi, a banking industry advisor at Salesforce, said in the same panel that most of the company's banking clients are starting with internal applications of generative AI due to uncertainty with potential regulation.

Nima Ghamsari, founder and head of Blend, also highlighted using generative AI as an assistant for tasks like coding, which banks such as JPMorgan Chase have been testing. Blend recently launched a product called Blend Copilot to assist loan officers by evaluating customers' financial information.

Close up on two people as they shake hands.

SFIO CRACHO - stock.adobe.com

Vetting AI fintechs requires strong due diligence

Most banks, especially smaller institutions and credit unions, don't have the resources to build technology in-house, so they have to partner with fintechs to apply AI. But choosing if, how and with whom to partner isn't a perfect science.

Lance Senoyuit, a financial services executive consultant at software company SAP, said he advises clients to be diligent and take their time since banking is highly regulated.

Thomas Novak, chief deposits and payments officer at Visions Federal Credit Union, said the institution has a fintech collaboration framework that it's developed to evaluate potential partners.

"In the build, buy or partner equation, we're a relatively small financial institution in the grand scheme of things — about $5.6 billion in assets here in the Northeast," Novak said. "There's only going to be so much bandwidth that we have for internal development or to really work on our own internal models for a particular business case."

Novak said that Visions asks a series of questions as part of its due diligence, similar to its onboarding process for non-fintech vendors, including if the fintech has enough data to train AI models, and if the fintech has a business model advantage.

Chris Barlow, vice president of digital channels at Union Savings Bank in Danbury, Connecticut, said the bank outsources all of its technology development needs. In 2022, Union Savings set up a Fintech Council to evaluate the bank's best uses for fintechs.

Novak added that the bank typically partners with mature fintechs, instead of early stage startups, and uses vendor management guidance from the National Credit Union Administration. Earlier this year, bank regulators released interagency guidance that puts the majority of the risk responsibility on financial institutions.

Stuart Cook, chief innovation officer at Valley Bank, said the Wayne, New Jersey-based company uses a venture fund to invest in fintechs.

Rido - stock.adobe.com

Governance has to be a team effort

Creating governance frameworks is essential to responsible adoption of AI, experts said.

Swarup Pogalur, who leads digital capabilities, AI and machine learning engineering at Wells Fargo, said it's important to outline how AI will be used, and how it won't be. He added it's important to ensure there's a human in a position to make decisions.

"You have to produce controls upfront," Pogalur said. "This is where the classic AI to generative AI will be very different. In classic AI … after the model is developed, you would go back and do all of this testing and validation. But as you go into the generative AI … you bring in all of these various functions in your institution, whether it's legal, risk, cybersecurity," early in the process.

Pogalur also said that banks should have systems in place to take immediate corrective actions if AI models start "falling off the rails."

Grimm said Fifth Third put together a cross-functional AI governance council that meets weekly to discuss how to balance the business advantage of the technology with potential risks. She added that sometimes there's conflict about which applications of AI to prioritize, and how, but the group is designed to keep everyone in the loop.

"We're a bank, and risk always seems to be top of mind," Grimm said. "People want to be quick to say, 'No, we can't do that.' But we're trying to say, 'What can we do then?' How do we get people comfortable [enough] to allow us to take some of these use cases that have been bubbling up and saying … we can try this with these controls."

Proving ROI is easier said than done

Investing in AI isn't easy, or cheap. Madjlessi from Salesforce added that many banks may have other challenges to solve before they can get the maximum value of AI.

Madjlessi said he's seeing larger clients carve out resources to invest in AI, mainly in areas like internal bottlenecks and inefficiencies. He added that banks can also shave off costs in areas like employee training by making the technology more intuitive and easier to use.

"In this environment where capital expectations are rising, the fight for deposits is real," he said. "I think there is this reality check of, we can't do all things. But gosh, if we don't get started, we're going to be so far behind that irrelevance might be a problem, too."

Wells Fargo evaluates AI investments across three tenets, Pogalur said: top line increase, like attracting more of a customer's wallet share; operational reduction by improving productivity; and risk management.

Grimm said it's important to have a viable and specific business case, including expected returns. She added that Fifth Third is exploring large language models in a cheaper way, by starting with small test cases and utilizing internal data scientists.

Novak from Visions Federal Credit Union said it's key for organizational buy-in that a fintech partner aligns strategically and creates efficiencies across the institution.

AI can solve, and cause, problems with fraud detection

While AI can help develop forms of fraud, like deep fakes, it can also be part of an additional security layer for banks to detect fraud. Oscar Gonzalez, senior product manager of digital channels at Citizens Bank, said AI can't be the only method banks use to prevent fraud.

Cook, of Valley Bank, said AI can also drive false positives, or erroneously flag an account for fraud. Valley Bank is working with a fintech called Refine Intelligence that uses AI to mitigate the amount of false positives.

Cook said the bank wanted to better understand the false positive process by building a learning model around its interactions with customers that generate fraud alerts. The idea was to reframe fraud detection from "finding bad guys" to "finding good guys."

"We tried to reframe that problem," Cook said. "The focus has always been on, 'How do you make your current process better to reduce those false positives?' And it just struck us that we're on this constant search for the bad guys. … What if we turn this upside down and actually said, 'Is there a way here that we can drive greater understanding of who the good guys are and who the good customers are, and automate some of that process?'"

Gonzalez said Citizens also works with a vendor on this challenge. He said it's a balance between adjusting the AI algorithm to limit false positives without increasing risk and the balance between controlling the AI algorithm without increasing risk. He added that the bank has a team to verify alerts created by the system to mitigate impact on clients.