By Walt Williams
Banks are racing to deploy artificial intelligence in multiple business functions, and at least one study suggests that in one area — lending — the technology may be giving institutions a competitive edge.
A recent paper led by a University of Missouri economist compared the results of the U.S. Census Bureau’s annual technology survey with banks’ small-business lending patterns. The researchers concluded that banks with greater AI usage lent to borrowers who were further away from physical bank branches, according to a Mizzou summary of the findings. They also concluded that banks with greater AI usage offered lower interest rates and experienced fewer instances of default, even when lending money to distant borrowers.
“When implemented carefully, AI can help banks extend credit to underserved regions without sacrificing loan quality,” says lead author and Mizzou economist Jeffrey Piao.
Still, there remain considerable hurdles for banks seeking to use AI for lending. Beyond the cost of the technology, there are questions about risks and how AI can best be implemented.
“One of the biggest challenges, especially with credit admin and using AI, is data bias, both known and unknown, and the problems that presents,” says Meredith Piotti, principal at accounting and advisory firm Wolf and Company. Piotti participated in a panel on AI and lending and credit risk management at the ABA Risk and Compliance Conference in June.
Adopting AI
Banks should approach AI adoption in terms of the customer lifecycle, from when they first open an account to when they leave — “which is hopefully never,” says Harsh Pandya, VP for product management at compliance solutions provider Saifr.
“Step one is just attracting them,” he says. “So doing target audience analysis where you can accurately identify the folks that have a high propensity to buy into the type of service that you’re offering them, on the terms that you’re offering. AI is really useful here. There are lots of marketing solutions that you can buy off the shelf.”
The next step is to look at what your competitors are doing in the lending space, particularly nonbank fintech firms.
“It is probably helpful to think about what your application process looks like versus what theirs looks like … because those are all AI-first,” Pandya says. “They’ll create adaptive applications that are responsive to your previous response, and then just kind of update as you go, so that you ultimately have to answer fewer questions, and that will probably reduce the rate at which people abandon those applications in the first place. These are tiny ways that they get an edge in geographic markets where we have a presence.”
When it comes time to make a decision on whether to approve a loan and issue terms, banks will rely on information from credit bureaus, which were some of the first adopters of the machine learning at the heart of AI, he says.
“I would lead with the assumption that these technologies are being used in the scoring process, and when you’re making decisions using that kind of information, you may also be able to augment with additional information using artificial intelligence,” Pandya says.
Plan for risk
AI is only as good as the information fed into it, and that is where much of the risk can come from. One example from the academic world: An MIT researcher earlier this year inserted text into a research paper on AI that instructed large language model technologies such as ChatGPT to stop reading the paper after a certain point, causing the AI to hallucinate an incorrect answer when asked to summarize the paper’s contents.
Bankers need to be aware of the technology’s limitations before they engage with it. Krysti Cunningham, CERP, CRCM, SVP and chief risk officer at Security National Bank, says one piece of advice she follows is treating AI like a person, who can make mistakes from time to time.
“You have to do that because it doesn’t know what you’re thinking, just like an employee doesn’t know when you say, ‘Go build me this report,’” she said.
Many banks are diving into AI or are contracting with third-party service providers that use the technology, but relatively few are taking the time to spell out how the technology should be used. Piotti points to a 2024 survey by Bank Director that found only a third of bank respondents said their institutions had drafted policies on the use of AI, including acceptable uses, risks and ethics.
“If you haven’t done this yet, please make an AI usage policy,” Piotti says. “Also look at your third-party risk policy for AI in that.”
Understanding AI with ABA
New from ABA is a webpage consolidating all the association’s resources on artificial intelligence. It includes policy updates and resources, training, guides on topics such as generative AI, AI news, policy and advocacy analysis, agency guidance and additional updates. Find it all at aba.com/ai.
