As banks continue to invest in artificial intelligence software — at the end of last year, 40% of respondents in an American Banker survey said they planned to increase AI spending in 2025 — the question arises, are they seeing a return on this investment yet?
In a follow-up American Banker
Here's a closer look at these four hurdles to AI benefits, and how bankers can overcome them.
The cost of modernizing core technology
For just about half (49%) of respondents, the cost of upgrading their core system is the primary impediment to getting a return on AI.
Many banks have older core systems, in some cases decades old. These systems are written in programming languages like COBOL that young engineers don't know. It's not easy to feed data from them straight into AI models. And often these older systems operate in batch mode, so real-time data feeds are not possible.
At the same time, upgrading any major system is expensive and comes with risk. For instance, in 2022, VyStar Credit Union in Jacksonville, Florida,
"For banks that are thinking strategically, AI is going to be an accelerant" to core modernization projects, Celent analyst Alenka Grealish told American Banker. "It has to be, because they realize that the AI flywheel spins fast, and modernization is a first turn of the industrial flywheel. It's really hard. It takes a lot of investment with not as much return, until you get to the full turn, then they can start spinning."
Banks that get 12 to 18 months ahead could potentially stay ahead, she said.
"It's a sustainable competitive advantage for banks, because they have to be very cautious and careful when they use third party models or build their own models," she said. "They have to make sure there is explainability, no bias and that everything is done in an ethical manner."
The need for modern cores has to do with the ability to access and extract the data needed to run AI solutions, Brian Gibbons, principal at EY, told American Banker.
Some banks have been working to modernize their underlying data infrastructure so they can access, extract and integrate data from core platforms into AI models, he said.
"I think those companies have a head start and an advantage," Gibbons said. "Companies that are just getting into that now are realizing that a huge amount of their AI investment has to actually go into modernizing their data platforms and environments."
Michael Abbott, global banking lead at Accenture, pointed out that deploying AI doesn't always depend on having a new core system.
Some use cases, like summarizing customer calls into a call center or improving marketing communications based on behavior patterns, don't require a core system upgrade, he told American Banker.
For many other use cases, a data management upgrade is needed more than a core overhaul, he said.
"We are seeing an influx of people looking to modernize their core," he said. "But it's not AI driven. It's more regulatory and compliance driven, and end-of-life systems driven."
The high cost of AI talent
A third (33%) of surveyed bankers said the high cost of hiring people with AI skills gets in the way of seeing a return on their AI investments. Across the 50 banks tracked by research firm Evident, the AI talent stack has grown 12.6% over the last six months, the firm said in its 2025
"There is a war for talent, for sure," Gibbons said. "With advances in some of the tooling and capabilities, I think there is opportunity on the horizon where the high cost for data science and engineering workers may subside a bit. But I think there are not a lot of human beings who can really do that end-to-end design work and consider that end-to-end design in the context of regulations, in the context of cybersecurity needs, in the context of the governance, both data governance, model governance, etc."
Some banks are taking data and tech savvy executives from other areas with deep institutional knowledge and putting them in charge of AI, Gibbons said.
In Abbott's view, banks need to help employees gain the needed AI skills.
"What I'm seeing more and more is a lot of banks putting in training, hackathons, things where people in any function, whether they be in a mortgage group or risk group or compliance group or legal group, where they're encouraging innovation in that group, encouraging people to put their fingers on the keyboard and learn how it might change the way they're working," he said.
Grealish said that with more employees using AI to help them with their jobs, the workforce will shrink overall. "The future of work is attrition that will not be replenished to the same degree," she said. "That's just the natural evolution when a technology matures and can replace humans. So while it's like, whoa, look at the price tag, I think it's more a mental shift that the average compensation is going to go up and we're going to have some prices on skilled labor that is close to a VP that owns a certain line of business product."
Some bankers are looking for the ultimate hybrid, she said: "somebody who understands business, product, customers and who knows some engineer speak."
Data governance
Among the surveyed bankers, 33% said rising data governance and data management costs are preventing them from getting a return on their AI investments.
Most banks are product-centric, and have data organized by products rather than by customers, Abbott said.
"Very few banks can tie a customer's products together with their channel experiences and touch points," he said. "Banks want to have a digital memory of one, whereas, right now, unfortunately, they have a digital memory of all their products. And that's where they struggle." They also need permissions and rights to use customer data in AI models, he pointed out.
Grealish said banks are just starting to do data governance.
"Before, data just sat in a database," she said. "It wasn't asked to perform in an AI predictive model, or a gen AI employee-facing assistant that wants to integrate all the documents on some complex product, and all of a sudden, AI can't read the PDF even, because there's images and other things that aren't translatable. So I think it's more an awakening of, oh, we have to spend more here to get up to speed."
AI necessitates application programming interfaces, cleansed data, and data governance and explainability, she noted.
"That does require this long journey for many banks, to go from legacy core data that's not really consumable by models because it was meant just to record a transaction to data that's dynamic and can feed into models to refine them so they can perform better in a banking context," Grealish said.
Gibbons sees a trend of companies deciding that rather than feed their data to AI models, they will partner with core enterprise software providers who are embedding AI in their platforms, "and it effectively allows you to bring AI to where your data already is," he said. An example would be migrating to a customer relationship management software program with AI capabilities.
Vendors' prices going up
Increasing prices from vendors is another perceived obstacle to getting a return on AI. About 18% of respondents worry about elevated prices from AI model providers, 17% say rising prices from app providers are the problem and 11% cited increasing prices from cloud providers.
Grealish pointed out that over the past two decades, banks have moved from a fixed cost infrastructure to software as a service and cloud computing.
"All of a sudden, what had been an advantage, if you were at scale, was you built and owned these big pipes," she said. "When you move to a variable cost structure, there's not a big upfront cost, but you're not going to get that nonlinear scaling."
However, she believes competition among SaaS and cloud providers is bound to drive prices down, and then banks will be able to get lower unit costs as their volumes increase.
