There’s plenty of buzz for powerful new artificial intelligence models, but the less-flashy AI tools that financial institutions embraced in recent years are still reaping dividends.
Banks, credit unions and tech consultants all say traditional AI beats newer generative AI models at some key tasks — and that both have staying power. The industry is still in its early stages of adopting generative AI, such as large language models popularized by ChatGPT.
But traditional AI can be cheaper, is well-understood by regulators and is still making manual processes at banks more efficient, industry leaders say. Whether in credit underwriting or spotting fraud patterns, executives see room to keep deploying traditional AI even as they experiment with new models.
“It’s not a replacement conversation — it’s coexistence,” says Michael Sabado, vice president for IT enterprise platforms at Golden 1 Credit Union, a $20.6 billion-asset firm in Sacramento, California.
Generative AI is a “valuable complement” to traditional AI, agrees Lisa Shim, head of technology and innovation at the $35.5 billion-asset BankUnited in Miami Lakes, Florida. Generative AI shines in summarizing large reports or customer feedback, she says, as its strengths lie in “managing unstructured inputs” and synthesizing insights.
But traditional AI excels when there’s a “high volume of structured and historical data,” Shim says, helping with risk forecasting, segmentation of customers, spotting patterns of attrition and other tasks.
“While generative AI is indeed exciting, we view it as an enhancement to, rather than a replacement for, traditional AI,” she says, adding that using them side-by-side ensures BankUnited is “leveraging the unique strengths of each.”
Some banks have seen plenty of success by pairing the two together, says Kevin Laughridge, a principal at the consulting firm Deloitte. A traditional AI model, for example, can flag transactions that raise suspicions of money laundering — and generative AI can draft a report summarizing the findings.
But any form of AI is “not a silver bullet,” Laughridge says. Rather than suddenly deciding they suddenly “need AI use cases,” banks should examine which parts of their operations need a rethink and then consider specific ways that AI can help, he says. Otherwise, tech costs may soar without driving meaningful returns, he cautioned.
Structural Strengths
AI is far from new, with a Treasury Department report last year flagging its usage in the 1940s. But it’s nonetheless seen major advances and widespread adoption from banks over the last decade.
Traditional AI tools, such as machine learning to spot patterns in data, are now “deeply embedded” in core parts of the banking system, says Omar Akkor, senior director of banking market strategy and innovation at Moody’s.
That includes compliance-heavy areas such as credit risk scoring, fraud detection and monitoring of anti-money laundering risks.
“These areas demand rigorous control and predictable outcomes, making traditional AI’s rule-based and statistical models preferable to generative AI tools,” Akkor says, adding that “the more these tools are used, the better they have become.”
Golden 1 Credit Union’s Sabado, for example, says customer onboarding “moved faster and we needed far fewer manual reviews” by augmenting credit scores with machine learning. Traditional AI also helped in detecting fraudulent behavior, catching patterns that earlier static systems missed and reducing “false positives without loosening protections.”
Traditional models “remain the right tool for the job” in many key risk functions, Sabado says. “Those are the places where it really fits: you have known inputs, measurable outcomes, and you need repeatable, auditable behavior.”
Unstructured Struggles
But understanding unstructured data such as customers’ needs in chatbots or business clients’ financial reports are where generative AI outperforms, says Deloitte’s Laughridge. Gen AI can easily pull together the latter, saving the need for copious manual entry.
“That’s a place Gen AI does great,” Laughridge says. “You can scrub all that unstructured data, pull the right fields and then get it into the right spot.” Earlier AI models and bank chatbots wouldn’t have been “as fast and effective” as newer models, Laughridge says. “The improvement in quality is astonishing,” he says. “What you can buy out-of-the-box to do things today is incredible.”
Still Early Days
But newer AI tools are not perfect either, Laughridge and other observers say, which is vital given that bank supervisors heavily scrutinize the industry’s use of any models.
Shim, the BankUnited tech and innovation head, flagged the risk of so-called “hallucinations” when generative AI models spit out inaccurate information. Supervisors also require “explainability” of any model that banks use, she notes.
By now, bank supervisors are highly familiar with traditional AI models, says Greg Ohlendorf, CEO of First Community Bank and Trust in Beecher, Illinois. But generative AI is a “whole other universe” that smaller banks and their supervisors will have to learn, he notes. “I think that will come,” he says. “But I don’t think we’re there today.”
First Community Bank and Trust, a $197 million-asset bank with two branches, is holding off on leaning into generative AI tools for now. It’s tough to justify investing in those tools when “we don’t do enough volume” in transactions to get a big-enough benefit, Ohlendorf says.
But traditional AI tools continue to speed up processes within the bank, from repetitive tasks when handling payments to vendors to mass-transfering files when someone retires.
The bank hopes to “continue to find more of those examples” to use traditional AI before investing more in newer models, Ohlendorf says, even though both will eventually be “part and parcel” of bank operations.
Many other banks are in the same category. Bank Director’s 2025 Technology Survey, which publishes later this month, found that 62% of respondents are experimenting with generative AI in limited uses.
First Community Bank and Trust’s current use of traditional AI tools is not “revolutionary,” Ohlendorf acknowledges. But it does free up employees from mundane tasks they’ve long been accustomed to so they can focus more on checking in with customers, he says.
“None of this is rocket science,” he says. “Those are really simple use cases, but every bank fights those same problems.”
Previous