Artificial intelligence

When the largest U.S. bank needs compute, there’s little margin for error.

Executives at JPMorgan Chase don’t want IT teams waiting in line to access resources needed to spin up an AI use case, upgrade a customer service portal, enhance market forecast analytics or support critical business applications.

“We're putting a lot of effort into improving the ability of our software engineers to be productive as they do development,” the company’s CFO, Jeremy Barnum, said during a January earnings call.

As technology drives business outcomes, infrastructure availability becomes a competitive advantage — and a priority.

“For our capacity, we like to use a just-in-time model, which usually means planning years before we need the capacity — as much as five to 10 years out,” Darrin Alves, CIO of infrastructure platforms at JPMorgan Chase, told CIO Dive. “If you wait until you're only a year or two away, you won't have the capacity available when you need it.”

Alves oversees the digital systems that connect more than 600 JPMorgan offices globally and over 5,000 individual branch locations to a complex hybrid ecosystem of private data centers, colocation facilities and hyperscaler deployments.

“We have everything from mainframe to quantum computers and blockchain to public cloud and generative AI,” Alves said.

Like cloud before it, generative AI is transforming banking processes, as coding assistants help unlock data in legacy COBOL applications, smarter chatbots improve customer experience and agentic solutions ease workflow processes.

JPMorgan Chase is an industry leader in AI adoption. The company widely rolled out its in-house generative AI assistant LLM Suite last September and ranked first for AI maturity in Evident Insights’ analysis of the sector’s top 50 financial institutions last year.

Modernization investments are integral to AI success going forward, said Alexandra Mousavizadeh, co-founder and CEO of Evident.

“Whether banks are building their own AI or buying third-party solutions, the end result will only be as good as the underlying infrastructure,” Mousavizadeh said. “As the drive toward organizationwide AI deployment ratchets up, we’ll start to see which institutions have placed the right bets.”

Computing capacity

Major transformations call for rigorous planning and a modernized infrastructure foundation. There are multiple challenges to surmount, especially in banking.

“Any time you go through an application, you first have to solve for legal and compliance,” Alves said. “It’s got to meet our security and risk and controls and then it's customer experience.”

Infrastructure and spending are also key considerations, though they are further down the decision tree.

“We have to decide where it’s going to fit into our architecture and the last step is to optimize for cost,” Alves said.

Finding the right compute environment and the capacity to run generative AI use cases became an enterprise pain point last year as hyperscalers raced to deploy GPU servers and build out data centers.

In August, CBRE analysts estimated that nearly 80% of data center space under construction had already been leased. Gartner forecast that 40% of existing AI data centers will be operationally constrained by power shortages in the next two years.

“AI expansion is putting constraints on the entire supply chain for data centers,” Alves said.

JPMorgan Chase navigated around the hardware hurdles proactively.

“We’re a company that wants to make sure we can control our own destiny, for our own business reasons and for regulatory reasons,” Alves said. “So, we partner closely with the hardware manufacturers and the colocation and data center designers.”

The bank is also vigilant about refreshing its data center infrastructure regularly, as it expands capacity where needed and leverages public cloud.

“If we’re on a continuous refresh cycle, we can accommodate significant growth without having to expand,” Alves said. “The most efficient data center that you can build is the one that you don't [have to build].”

Hybrid equilibrium

While cloud remains a cornerstone of JPMorgan’s IT strategy, the bank retains a diversified infrastructure portfolio and invests broadly in solutions.

“You use the right tool for the right job,” Alves said.

Firmwide, the company increased tech spend by $1.5 billion year over year to roughly $17 billion in 2024, Barnum said during a May presentation to investors. CEO Jamie Dimon said JPMorgan Chase would have three-quarters of its data and 70% of its applications migrated by the end of last year.

“We were methodical in our approach to cloud, so we didn't go too quickly,” Alves said. “As we went, we learned that you need to be very thoughtful about which workloads make sense in public cloud.”

Certain banking applications align with on-premises capabilities, while others do not, depending on their function, data needs and scaling requirements. 

In some cases, a modernized mainframe is the right call.

“We process more than 10 trillion dollars in payments a day on our systems,” Alves said. “While there are other systems that participate in that, the mainframes are the workhorse that do the bulk of the heavy lifting.”

Developing AI solutions called for a different strategy. Initiatives were launched in the cloud, where piloting potential use cases was relatively easy and affordable. Now, as trial-and-error leads to practical applications, Alves expects to adopt a hybrid mix of infrastructure to support AI.

“The cloud gives us a place to do that experimental learning with no regrets,” he said. “If it doesn't work, we can shut it down very quickly. You can't really do that on your own infrastructure if you're making a significant investment in what AI requires.”

Banks were a proving ground for the value of digital transformation and cloud migration. Now, the sector is helping make a compelling case for generative AI, thanks to its mountains of ingestible data siloed away in cavernous legacy systems.

The success of banking rests on its ability to rapidly and reliably process millions of daily transactions at scale — an undertaking that inherently benefits from a combination of cloud and AI. 

“At the core of the bank, technology is increasing in importance, as it has been for the last five decades,” said Michael Abbott, Accenture’s global banking lead.

The challenge as the next wave of modernization crests is to leverage complex technologies to simplify systems and processes, according to Accenture’s 2025 banking trends report.

“Generative AI is going to impact every function within a bank — every single part of the job,” Abbott said, pointing to agentic solutions tailored to manage dispute resolutions and coding assistants trained to refactor legacy applications.

The impacts are already apparent across the industry.

Citi armed 30,000 developers with generative AI coding tools and rolled out a pair of generative AI-powered productivity enhancement platforms to its broader workforce last year. Goldman Sachs aims to furnish roughly 10,000 employees with an AI assistant by the end of 2025, the bank’s CIO, Marco Argenti, told CNBC.

A technology step-change

As enterprises worked through generative AI pilots, accuracy and reliability emerged as concerns, particularly for customer-facing use cases. While issues persist, large language model-based coding assistants have come a long way in the past year, Abbott said.

Tools that can help engineers refactor COBOL applications were, by Abbott’s estimation, 70% to 75% accurate a year ago. Now, they achieve closer to 98% or 99% accuracy in reverse engineering legacy code.

“It’s not just translating from COBOL to Java — now you can transform the application, move it into an [application programming interface], and forward engineering is going to improve in leaps and bounds this year,” Abbott said. “You’ll be able to forward-engineer the application into any code base you want.”

The economics of generative AI are becoming more favorable, as well, even as many companies struggle to achieve returns on their AI investments.

JPMorgan Chase President Daniel Pinto said last fall the bank expects its LLM Suite and other AI capabilities to create $2 billion in value.

“CIOs should understand that the cost of this stuff is collapsing — the API cost for OpenAI services has come way down,” Abbott said.

Accenture tracked a 74% annual cost decline between GPT-3 and GPT-3.5 Turbo from December 2021 to December 2024. GPT-4’s annualized cost came down 58% between March 2023 and the end of last year, the firm found.

Hyperscale cloud providers have continued to pour tens of billions of dollars into infrastructure to support model training and inference. Abbott compared the economics of a data center gold rush to the 19th-century U.S. transcontinental railroad boom.

“It costs an enormous amount of money to build the first model and train it, just as it takes a lot of capital to lay a railway track,” Abbott said. “But once you got the track laid, moving a ton of coal a mile is marginally cheap.”

Abbott pointed to the January release of DeepSeek-R1, an open-source model family with greatly reduced usage fees compared to ChatGPT, as a harbinger of cost reductions to come.

“It’s way too early to tell if large banks will use DeepSeek, but it is open and freely available so they can analyze it and understand everything in there,” Abbott said.

Sponsored content

By Glia

Artificial intelligence (AI) is rapidly reshaping the financial services landscape, promising increased efficiency, cost savings, and enhanced customer experiences. However, despite AI's potential, many financial institutions struggle to implement it effectively. The road to AI adoption is fraught with challenges, from resource constraints to regulatory concerns. Understanding these obstacles and how to address them is crucial for banks and credit unions aiming to harness AI's full potential.

Below, we outline the most common concerns financial institutions face when implementing AI and provide actionable strategies to overcome them.

1. Implement AI With Speed and Ease

Financial institutions often perceive AI deployment as a resource-intensive endeavor, requiring significant time, budget, and technical expertise. Concerns about complex integrations and long implementation cycles can create hesitation.

How to Overcome It: Institutions should adopt AI solutions that prioritize seamless integration with existing systems. Low-code or no-code AI platforms can reduce the technical burden while allowing organizations to implement AI incrementally, focusing first on high-impact use cases such as automating routine customer inquiries or enhancing fraud detection. Partnering with vendors that offer turnkey AI solutions can also expedite adoption while minimizing disruption.

2. Tap Into AI’s Power Without Taking on Risk

High-trust industries such as banking and insurance are bound by strict regulatory requirements. Many organizations fear that AI-driven automation could expose sensitive data or fail to meet compliance standards, leading to audits, fines, or reputational damage.

How to Overcome It: AI solutions should be purpose-built for compliance, ensuring data privacy and security measures align with industry regulations. Financial institutions should prioritize AI models that are explainable and auditable, allowing them to track decision-making processes and mitigate risks. Additionally, forming cross-functional teams involving compliance officers, IT leaders, and AI strategists can ensure that AI implementations align with regulatory expectations from the outset.

3. Deploy AI to Solve Your Everyday Challenges

While AI is a hot topic in financial services, many decision-makers struggle to see its tangible impact beyond the buzzwords. Without clear use cases, AI adoption remains a theoretical exercise rather than an operational transformation.

How to Overcome It: Institutions should focus on real-world AI applications that directly address customer and operational pain points. AI-driven chatbots for customer service, intelligent call routing for contact centers, and predictive analytics for fraud prevention are proven use cases that deliver immediate value. Case studies, pilot programs, and data-driven insights can help stakeholders visualize AI’s potential and accelerate buy-in.

4. Use AI to Enhance Human Service, Not Replace It

In high-touch industries like banking, customer relationships are paramount. Many institutions fear that AI will replace human employees, eroding trust and diminishing the quality of service.

How to Overcome It: Rather than replacing employees, AI should be positioned as a tool that enhances human capabilities. AI-powered assistants can reduce agents’ workloads by handling repetitive tasks, allowing staff to focus on higher-value interactions that require empathy and problem-solving. Providing AI training programs for employees can also ease concerns and enable staff to leverage AI effectively in their roles.

5. Create an AI Strategy Built for Today and Tomorrow

Many financial institutions recognize AI’s importance but lack a clear roadmap for implementation. Without a defined strategy, AI initiatives often remain fragmented or underutilized, creating competitive disadvantages.

How to Overcome It: Developing an AI strategy aligned with business objectives is crucial. This involves setting measurable goals, identifying priority use cases, and ensuring executive sponsorship. Institutions should also consider forming AI governance committees to oversee implementation, monitor performance, and adapt strategies as AI capabilities evolve. Regular benchmarking against industry peers can also provide insights into best practices and emerging trends.

AI Is No Longer an Option—It's a Necessity

While the challenges of AI adoption are real, they are not insurmountable. By addressing implementation barriers, ensuring compliance, focusing on practical applications, empowering employees, and developing a clear AI strategy, financial institutions can successfully integrate AI into their operations. AI is not just the future of banking—it is the present. Institutions that proactively navigate these challenges will be best positioned to deliver efficient, secure, and customer-centric banking experiences in the years to come.

More information at glia.com

Klarna’s workforce shrunk by 21% in 2023 and 2024 as the buy now, pay later behemoth increasingly turned to artificial intelligence to support its operations, the company disclosed in March when it filed to sell public shares. As of the end of 2024, 2023 and 2022, Klarna had 3,422, 4,352 and 5,527 full-time employees, respectively, the filing said.

“The reduction in the number of full-time employees resulted from our strategic decision to reduce our overall headcount and drive operational efficiency by leveraging AI in our business and focusing on what really matters to our mission,” the company said in the filing. “We expect the number of employees to continue to decrease in future periods.”

Klarna turned profitable in 2024 with $21 million in net income on a 24% jump in revenue to $2.8 billion, the buy now, pay later payments provider said in the filing.

Klarna had 93 million active users and 675,000 merchants in 26 countries, as of the end of last year, according to the filing. The payments company filed a confidential registration in November with the Securities and Exchange Commission for review ahead of its initial public offering on the New York Stock Exchange, where it plans to trade under the symbol KLAR.

The business was founded 20 years ago in Stockholm by Sebastian Siemiatkowski, Niklas Adalberth and Victor Jacobsson, becoming one of the first BNPL companies. Today, Klarna is one of the most prominent names in a sizable field of BNPL rivals including Affirm, Afterpay and Sezzle.

Its competitors also include banks in the payments realm, such as card issuers like JPMorgan Chase and Citi, as well as neobanks such as Revolut. The traditional card network companies are also competitors, including Visa, Mastercard and American Express. Digital payment firms Block and PayPal are also rivals.

The company disclosed in the filing that it reorganized last May in a share-for-share exchange to redomicile its parent company in the U.K., from Sweden, registering its principal office at a London address. 

Nonetheless, the company’s operations remain predominantly situated in mainland Europe, with the largest offices, by square footage in Stockholm, Berlin and Giessen, Germany, respectively. Its U.S. headquarters in Columbus, Ohio, is relatively small at 15,602 square feet, compared to 244,308 in Stockholm.

That seems to be reflected in the workforce, as well, with at least 2,130 employees, or 62% as of the end of last year, based in Sweden and Germany. (The company didn’t disclose exactly how many workers are in each region, but it said that many were associated with European labor unions or worker councils in those countries.)

Klarna pointed to AI as key to its business, saying that 96% of employees, as of last August, used the generative form of artificial intelligence in some way in their work. 

“We believe that society will experience vast changes that will be powered by AI,” the company said in the filing. “Accordingly, we are utilizing AI to transform commerce. We harness AI for the benefit of consumers, merchants and our employees. AI makes us more productive and efficient.”

The installment payments market has grown steadily in recent years – especially among younger consumers – as a form of credit for many people without access to traditional credit card products. The BNPL model offers interest-free financing, or, typically, interest rates that are lower than traditional credit.

The company said its network gives consumers and merchants access to payment and advertising software across Europe, North America, Australia and New Zealand.

Klarna seeks to raise at least $1 billion in the IPO and is targeting a market valuation for the company of more than $15 billion, Bloomberg reported March 6.

The company’s profit in 2024 compared with a $244 million loss in 2023, on sales of $2.3 billion.

Klarna touted its recent investments to scale its global revenue growth, specifically in the U.S., but it also noted the growth stateside contributed to net losses. Klarna said it “reached an inflection point” in 2023, which led to significant operating margin improvements.

The company’s gross merchandise value rose 14% to $105 billion last year, according to the filing. 

“Our GMV has consistently grown faster than the broader market,” Klarna said. “Our flexible and personalized products, trusted consumer brand, global distribution and proprietary scalable infrastructure are the foundations enabling us to become our consumers’ everyday spending and saving partner, available everywhere and for everything.”

Klarna and JPMorgan Chase announced a partnership in February to extend Klarna’s BNPL offerings to the bank’s business clients.

As of 2023, the consumer retail and travel spending in Klarna’s markets was $18 trillion, with half of that in the U.S., the company said in the filing, citing a third-party consulting firm’s market opportunity study. Klarna says its addressable market is the $450 billion payments revenue opportunity connected with that spending.

While Klarna earns most of its revenue from transaction fees it charges merchants, it also derives revenue from consumer fees, as well as from advertising services sold to merchants. The BNPL provider taps its trove of consumer data to let merchant clients target their ads.

Klarna said its advertising revenue increased from $13 million in 2020 to $180 million in 2024, and it believes it is “uniquely positioned” to address the $475 billion global digital advertising market opportunity, excluding China.

“Additionally, we are well positioned to build a leading presence in strategic adjacencies such as retail banking services, given our trusted relationships with consumers, our experience and our existing banking services in select regions, where we held $9.5 billion of consumer funds,” at the end of 2024, Klarna said in the filing.

Varo Bank is pursuing faster growth by integrating machine-learning models into various aspects of the customer life cycle, said incoming CEO Gavin Michael.

Varo’s strategy includes leveraging artificial intelligence and ML in underwriting for the neobank’s low-dollar loan offerings, fraud prevention and detection, and in data-driven customer experiences with enhanced personalization. He also sees the bank using it in app curation to help customers better manage their finances and enhanced customer service. 

AI and ML can “help make sure the great experience that our customers are seeing on the way in and through our app is reflected in every touch point they have — simple, human, personalized experiences that happen across all of our channels,” Michael told Banking Dive.

Michael, the former CEO of Bakkt, was instrumental in taking the cryptocurrency exchange public in 2021. He joined Varo in November as a leader, according to his LinkedIn profile. Michael’s prior stints include head of technology for Citi’s global consumer bank, head of digital for consumer and community banking at JPMorgan Chase, chief technology innovation officer at Accenture, and IT director for Lloyds’ retail-banking business. 

It was at Lloyds, in 2009, that he met Colin Walsh — the outgoing CEO of Varo, who has led the company for the past 10 years — and the two kept in touch over the years.

“When the opportunity came along to bring Gavin into Varo, I was very excited about that, and got the board excited about it, and started to put in motion what I would say is a very well planned out succession,” Walsh said, adding that Michael will help put the company in “turbocharge” mode. 

Michael’s expertise in software development will help him drive commercial success — a skillset that Varo needs now, Walsh noted. Varo was not profitable in 2024, though its revenue grew 22%, lending rose 45% with a 38% reduction in losses, he said. Varo recently secured $55 million in funding, Walsh said. 

The transition of power should be completed by April 30, Walsh said. He will remain active on the bank’s board as founder and a major shareholder and will engage in public speaking for the company. 

Michael stressed that Varo will remain focused on customer value and financial inclusion. To further those efforts,  the San Francisco-based company will practice more careful management of its expenses, align technology development with commercial goals, and implement a multi-quarter roadmap with clear priorities. 

“The next stage is scaling and just taking the business to the next level, achieving profitability, and then thinking about part of why we brought Gavin in is — he's taken a company public as well,” Walsh said. “So we'll see. We will keep all options on the table.”

Technology boosting customer experience

Varo’s “secret weapon” is its charter allowing it to offer fee-free banking and lending products, Michael said.

Varo focuses on building products that precisely match its target customer needs, like high-interest savings accounts, deposit products, evolving cash back offerings, free banking access through a partnership with drugstore chain CVS, secured credit cards and low-dollar loans.

Varo’s products are typically created in-house, and Michael said this suite serves Varo’s target market well. That group includes low-to-middle income customers, financial beginners and those with thin or no credit files. 

The lender has undergone a major upgrade of its app, moving from iOS, Android and web platforms into a single React Native app. Walsh underscored Michael’s contribution to the change and said Varo will begin to see the benefits in 2025 with a clearer path toward profitability.

Rather than expanding to new product categories, Varo plans to refine its existing offerings, Michael said. While the lender is open to building, buying, or partnering as customer needs evolve, the current focus is on scaling the company’s existing products.

“And it's the charter that enables how our business operates. And that's something that is the secret weapon that allows us to be able to serve everyday Americans,” Michael said.

Customer acquisition, crypto

Last year, Varo’s customer acquisition cost fell 31%. Walsh credited the company’s new chief marketing officer, Hussein Kalaoui, for improving efficiency through disciplined marketing. Varo has about 5 million customers, he noted.

Walsh said there’s potential to bring back customers who left Varo, by leveraging customized product offerings. Michael also emphasized the importance of customer acquisition and retention in Varo's growth strategy.

Although he has a background in crypto and aims to leverage his experience in technology and banking, Michael underscored that Varo will not rush into new technologies unless they provide meaningful impact.

Walsh noted that he sees digital finance opening up new avenues for financial inclusion.

“Gavin has been very much a practitioner in the space, having run Bakkt. So, I think it's a matter of, how do some of these things intersect at some point, as we think about our technologies and our product offerings over time?” he said.

As the financial sector moves forward with AI plans, banks added governance and ethical use expertise to their workforces in 2024, according to research by Evident Insights. Out of 50 large banks surveyed, 41 had dedicated AI governance professionals at the end of the year, up from 31 in 2023, the industry research firm said in March.

Banks are standing up AI risk and governance groups to further ensure safe and responsible artificial intelligence implementation. More than half of the banks tracked by Evident Insights have established AI oversight teams. Executive-level AI governance positions exist at 33 of the banks surveyed, up from 15 in 2023, the analysis found.

Leading banks are conducting responsible AI research in-house to help guide adoption. JPMorgan Chase has published the most research on the topic in the past three years, a period that saw the overall number of responsible AI research papers by banks surveyed increase by 136%.

Banking emerged as a nexus of generative AI experimentation and adoption in 2024, with some of the sector's largest players deploying large language model-powered technologies.

JPMorgan rolled out the LLM Suite AI assistant to 140,000 employees in September, and Morgan Stanley leaned on OpenAI to launch its AskResearchGPT chatbot internally the following month.

The technology gained additional traction in the industry this year, as Citi equipped 30,000 of its developers with generative AI coding tools, and BNY expanded its partnership with OpenAI in an ongoing effort to build out its Microsoft Copilot-enabled Eliza generative AI platform.

Generative AI is poised for a breakthrough in banking, according to a January report from IBM’s Institute for Business Value. While tactical adoption was the norm in 2024, with three-quarters of banks IBM surveyed in the early phases of deployment, maturity is on the horizon.

“We are seeing a significant shift in how generative AI is being deployed across the banking industry as institutions shift from broad experimentation to a strategic enterprise approach that prioritizes targeted applications,” Shanker Ramamurthy, IBM Consulting’s global managing director of banking and financial markets, said in the report.

Evident Insights’ research highlights the role leadership, governance and risk controls will play in adoption strategies. Responsible AI talent, a category that includes governance, risk and ethics specialists, comprises a small fraction of the financial sector’s AI workforce, numbering 247 individuals across 50 banks in Q4 2024.

JPMorgan — which led the industry in AI adoption in 2024, according to Evident Insights’ AI Index — had 20 professionals in governance roles, nearly triple the average. The bank filed a patent application for a model bias risk assessment tool, as part of its responsible AI research initiatives, Evident Insights noted in the March report.

During an earnings call in 2024, JPMorgan CFO Jeremy Barnum said the company prioritized a “pragmatic and disciplined” approach to adoption.

Stash’s Money Coach, an AI-powered financial advising tool, has seen a favorable impact on its financials, with the company seeing deposits surpass its withdrawals.

The New York City-based fintech has sought to help consumers build better financial habits by advising them on long-term investing ideas. Every month, Stash helps 1.3 million subscribers and has nearly $4.3 billion in assets under management, Ed Robinson, co-founder and co-CEO of Stash said in an interview with Banking Dive.

The company has shifted from “burning a lot of cash” to growing revenue and being free cash flow positive – for the first time in the company’s history – while returning to growth.  

“We don't see [Money Coach] as the savior of the world of AI. We see it as a tool that can help drive really efficient and smart decisioning for our customers within our ecosystem,” Robinson said.

Robinson co-founded Stash with Brandon Krieg in 2015. Robinson and Krieg returned in September as co-CEOs after then-chief executive, Liza Landsman, stepped down.

About 86% of Stash customers are first-time or beginner investors, many of whom have never had access to personalized financial advice.

Now, nearly one in four users takes a step toward diversifying their portfolio, buying or depositing after connecting with Money Coach within 15 minutes – up 116% since September, Robinson said, highlighting Stash customers make an average deposit of $45 per week. 

Many people want to invest but lack proper advice on taking the first step, Sarika Doshi, senior vice president and head of marketing at Stash said, adding that customers tend to feel overwhelmed by the number of financial products in the market.

“It is often the case that the lack of advice actually is the biggest hurdle to getting started,” Doshi told Banking Dive.

AI will continue to play an important role in Stash’s operation, Robinson said.

“It’s going to be touching the whole business, all across from the way we message our customers…the way that customers reach out to us, the way they get advice, the way we personalize the advice, the way we interact as leaders to our employees, everything,” he said. 

Editor’s note: This interview has been edited for clarity and brevity.

BANKING DIVE: How do you differentiate yourself from rivals?

ED ROBINSON: Our biggest moat is that we are a registered investment adviser. [We do business] with fiduciary and so we do not try to get our customers to trade in and out. We provide recommendations, and we centrally provide financial education. That's one pillar. The second pillar is the target of the market we're going for. We aren’t looking for the big whales. We’re not looking for higher earners. We’re looking for everyday Americans. We’re looking to help them make the right investment decisions. We’re not like a Robinhood. We’re not like a world fund where we need people who have some money to trade actively. We’re a subscription business model. So, what we want to focus on is driving a really solid product through advice and education. 

There’s some concern that AI-generated results aren’t always accurate. What do you make of that?

We’re always training and updating the way the models are working, but the way we look at it is we’ve got a very, very highly trained and highly efficient model that’s looking at customers’ accounts and...[at their] risk profile.

It analyzes your portfolios, and then it helps make recommendations inside the Stash app, which is gated, which people pay a subscription for, and customers get an elevated and very customer-centric experience.

So, it’s not as simple as just a chatbot anymore. Chatbot is so done. When we look at the questions that our customers are looking for and what they’re actually asking for, the No. 1 question is: What is my next best action? We provide that in the app. “You should diversify, should turn on Auto-Stash.”

The second biggest question is: How am I doing? That’s what our customers are looking for, and the way that we’re analyzing that data is looking against customers based on segment, based on cohorts, based on risk profiles, based on ZIP codes, to actually give people advice, like, “Hey, you’re actually doing really well,” or “actually you’re too exposed to, say, Tesla right now, think about these three or four investments that could help you get diversified.” These are ways that we’re going to utilize it. 

How do you see the fintech space evolving, and what is Stash’s role in this evolution?

There are tons of things happening right now with X getting into banking, accounts and trading. You’ve got Square, you’ve got the Robinhoods and the trading brokerages. You’ve got the old incumbents. There are still over 5,500 financial institutions in the U.S. I think it’s going to become much more of a digital play and there’s going to be a lot more consolidation over the years, because the product offerings and the servicing through the technology that’s been built over the last few years are only speeding up.

You want products like Stash that provide customer-centric advice and recommendations to help because it’s getting more and more complex. Do you buy crypto? Do you buy international? Do you buy bonds? How much should you be investing? When should you be investing? This is where financial advice needs to come down to everyday Americans so they can make the right decisions and make sure they set themselves up correctly.

Citi armed 30,000 developers with generative AI coding tools as part of an ongoing modernization push, CEO Jane Fraser said, during the bank’s fourth-quarter earnings call in January.

The bank spent $11.8 billion on technology in 2024, investing in digital innovation, product development, client experience and cybersecurity, CFO Mark Mason said. Citi invested an additional $2.9 billion in transformation initiatives, including infrastructure, platforms, applications and data, according to Mason.

“The investments we're making to modernize our infrastructure, streamline processes and automate controls are changing how we run the bank,” Fraser said, pointing to a cloud-based risk analytics solution and two generative AI productivity enhancement platforms the bank deployed to 143,000 colleagues in 2024.

Citi has steadily beat the modernization drum for the last three years, retiring 2,000 legacy applications in the process.

The bank’s tech spend surged north of $11 billion in 2022 and reached $12 billion in 2023 before a 2% decline in 2024. Investments in transformation increased 4% to $2.9 billion in 2023 and remained at that level in 2024.

“We all want transformation to get done quickly and we want it to get done right,” Fraser said.

The push comes in the wake of what Fraser characterized as decades of historic underinvestment in Citi’s infrastructure and risk and control systems. The urgency was triggered by 2020 enforcement actions from the Federal Reserve and Office of the Comptroller of the Currency, tied to poor data quality management, followed by $135.6 million in penalties levied by regulators in 2024.

“While there are areas that are more advanced, there are others where we still have a lot more work to do, particularly around data and regulatory reporting as last summer's regulatory actions reinforced,” Fraser said. “We have reviewed the entire data program and made changes to its governance and structure as well as increased the level of investment.”

Technology lies at the heart of banking as the industry’s biggest players leverage massive data estates to fuel generative AI initiatives and broader digital transformation.

Data is an area where Citi has more work to do, Fraser acknowledged, noting the bank increased investments last year to “expand the scope and accelerate some of the work to satisfy our regulators' expectations.”

As remediations progressed, Citi consolidated four activity risk management platforms down to a single platform, launched a regulatory reporting platform and deployed an AI-assisted data management tool. The bank tapped Google Cloud for AI and analytics in October and rolled out the Stylus document intelligence and Assist virtual assistant AI tools to its workforce in December.

Despite elevated tech spend, Citi reported net income of $2.9 billion for the fourth quarter, compared to a net loss of $1.8 billion during the same period last year. Net income for the year went up nearly 40% to $12.7 billion.

“We are continuing to invest in order to achieve what we need to get done here and also ensure the businesses have what they need to drive sustainable growth going forward,” Mason said.

If the past 10 years of mobile banking has been a race to add more features, Frost Bank’s Jimmy Stead expects the next decade could lead to greater simplicity in banking apps, thanks to artificial intelligence. 

“The end result of focusing on features for 10 years: A lot of apps have become a lot more complicated to use. There’s so much in there,” said Stead, Frost’s chief consumer banking and technology officer. 

Looking out “on the next couple horizons,” Stead expects a shift from that mindset to “making these apps smarter, and better, and more customized. And it’s really AI that is going to allow you to do that in a meaningful way.” 

The San Antonio-based lender, which has about $51 billion in assets, debuted its mobile banking app 10 years ago. Since then, the bank has endeavored to keep the app simple and intuitive for its customers. “I think that’s helped us over the years,” said Stead, who was part of the team that brought online banking to Frost and created its mobile app.

All industries are talking about or pursuing AI as an efficiency tool, yet “the thing that's less talked about is just how impactful it can be on the customer and helping them,” Stead said. 

In mobile banking apps today, “if you need to get something done, you’ve got to follow a really defined path that that company has defined for you,” he said. “But we’re entering a time where you can really, in your own words, tell the app what you need to do, and it’ll walk you through that. It’s more on your terms.”

Personalization and customization in mobile banking apps will become more meaningful with AI tooling in place, Stead said, although it’s likely to take some time before becoming widespread in the industry. 

Broadly, banking is expected to be significantly affected by generative AI. AI is only just beginning to change mobile banking, consulting firm Forrester said in 2024, as top banks are expected to begin using AI to anticipate customer needs and provide personalized guidance. 

Some bank customers have reservations around AI use in their financial lives, but those in the industry with an eye on banking innovation are closely watching agentic AI technology, in which AI systems act independently and perform tasks autonomously without human oversight or guidance. In time, that’s expected to shift the digital banking mindset, from “do it yourself” to “do it for me.” 

Frost, for its part, is exploring AI use in different areas of the company, but remains particularly cautious about its use in customer assistance, Stead said.

“We’ve absolutely avoided fronting conversations with a customer using a bot. It’s something that I don’t think that we’ll ever do,” he said. 

The bank is using AI, however, to aid employees in assisting bank customers more quickly and efficiently, by giving them more knowledge at their fingertips. “It’s not a substitute, it’s not in the middle of our conversations with customers,” Stead added.

For Frost, it’s about “shrinking the barrier for a customer to interact with us,” he said. 

The first version of the bank’s app featured a customer service call button on the log-in page “and every page thereafter,” to make it easy for customers to reach an employee, he said. It’s evolved from there: Customers who now log into Frost’s mobile app pass their authentication along, so customer service staffers are immediately aware of who they’re talking to and customers don’t have to jump more hoops, he said.

Staying in touch with customers has provided Frost with key opportunities to improve, too. Frost builds its own digital banking software in-house – something that Stead said sets the lender apart from its peers, most of whom rely on large vendors – so when customers complained about a navigation change Frost made to its app, the bank was able to design an improvement within a few hours of receiving those complaints, and implement a fix within three days. 

For the bank’s digital employees, “the expectation is, they’re talking to customers,” he said.

On the brick-and-mortar front, Frost has been expanding its branch network in large, fast-growing markets such as Austin, Dallas and Houston in recent years. Stead contends the Texas lender has a headstart compared to the likes of PNC and Huntington, which have more recently announced plans to increase branch presence in those markets.  

“Obviously, we don’t have the purse strings of the really big banks,” he said. But “we’re really close to our customers, and we know what drives them and what they need, and can respond really quickly to that.”

Competitive pressure around incorporating generative artificial intelligence could heighten risks in financial services, Federal Reserve Gov. Michael Barr warned in February.

“Nonbanks may be more nimble and risk-forward in incorporating genAI into their operations, which may push intermediation to less-regulated, less transparent corners of the financial sector,” Barr, the Fed’s recent vice chair for supervision, said during remarks at a Council on Foreign Relations event in New York City.  

That competitive pressure “may push all institutions, including regulated institutions, to take a more aggressive approach to genAI adoption,” increasing governance, alignment, and financial risks around AI, he said.

Barr called for monitoring of how generative AI’s introduction alters banking. “The very attributes that make genAI attractive — the speed, automaticity and ability to optimize financial strategies — also present risk,” including increasing market volatility and stoking asset bubbles and crashes, Barr said.

Banking has been labeled one of the industries poised to be most affected by the technology, and many larger banks are eager to find the best uses for it and demonstrate return on investment. Generative AI, already embedded within much of the technology that banks buy, is being used by lenders to handle tasks such as call summarization at call centers. Banks are also pursuing uses such as drafting scripts for marketing purposes, or finding red flags in mortgage applications. 

But fintechs and nonbanks face fewer regulatory constraints than lenders, and some fintechs such as Bunq and MoneyLion have moved more quickly to incorporate generative AI in customer-facing functions.

To harness generative AI’s benefits and reduce its risks, lenders and Federal Reserve banks “should consider investing sufficient resources in understanding genAI technology, incorporating it into their workflows where appropriate, and training staff on how to use the technology responsibly and effectively,” Barr said.

Regulators, for their part, should approach the fast-moving AI landscape with “agility and flexibility,” he said.

Barr also stressed the importance of AI governance. 

“We should make sure that we think about GenAI as enhancing, not replacing, humans, and set up best practices and cultural norms to that end,” he said. 

Every bank should understand the technology’s limitations, consider where and when it belongs in processes, and recognize humans’ best position to remain in the loop, Barr added.

Attention, too, should be paid to data quality, Barr said, to ensure generative AI’s uses don’t perpetuate or amplify biases, or make incorrect inferences. Regulatory frameworks for understanding model risk “may need to be updated to address the complexity and challenges of explaining AI methods and the difficulty of assessing data quality,” he said.

“AI is kind of overhyped right now, in the short term – I think people are expecting more from it last year, this year, next year – but I think probably underappreciated, in the long term, in terms of the kinds of changes that it could cause in our society,” Barr said following his prepared remarks.

U.S. Bank’s innovation team was scouting for a particular trend at January’s annual Consumer Electronics Show in Las Vegas: agentic artificial intelligence.

That’s when AI systems act independently and perform tasks autonomously without human oversight or guidance – the sort of “do-it-for-me” functions that might inform customer experience of the future.

“We know it's coming, and we know that these technologies are going to be out there; we just need to find the right ones,” Don Relyea, the Minneapolis-based super-regional’s chief innovation officer, said during a recent interview.  

Relyea declined to say when he expects the bank might adopt agentic AI to assist customers, or comment on innovations the lender is planning to launch. But CES, historically, has proved fruitful for the bank: The team saw emerging voice technology at the show about a decade ago, took the concept into research and development, and rolled out its voice assistant in its mobile app in 2020. That timeline enabled the bank to get an early jump on integrating that assistant with Siri, Google and Alexa, and expanding into Spanish in 2022. 

Relyea predicts a “huge wave” of innovation over 2025. 

“Folks really are putting a lot of money into AI. And agentic AI, in particular, is really easy to weave into various customer experiences,” he said. “You’re just going to see so much innovation happening that, for us, figuring out the ‘how’ is not as important anymore as figuring out the ‘what,’ from a customer experience standpoint.”

Editor’s note: This interview has been edited for clarity and brevity. 

BANKING DIVE: How does the bank approach CES, and what struck you this year? 

DON RELYEA: We go to CES every year with a goal of looking at key future tech trends that we know are going to impact banking moving forward. A lot of these are things everybody’s talking about, like AI. The metaverse, smart things – when do we need to be prepared to bring financial services to those spaces? How are other companies leveraging agentic AI? 

We're really focused on our customer experience, so when we see companies that are a little bit more mature in the customer experience design space doing something with agentic AI, those things stand out. 

We saw some really interesting things where people had thought about their customer experience and combined it with agentic AI. Many of those are lifestyle-type companies – your Panasonics, your Samsungs, LGs – all thinking now about “how can we leverage this new technology to be more proactive for our customers, more predictive, and initiate conversations with our customers”?

In the past, we haven’t seen as many pure financial services companies showing up there. This year, though, was a little bit different. There were several foreign banks showing new types of innovations, some of them very AI-powered. Industrial Bank of Korea had some interesting demos. Some Chinese banks were demoing virtual tellers and bankers, other banks were doing things like leveraging agentic AI to research and do predictive analysis on your venture portfolio.

The virtual bankers can do all kinds of stuff for you, and you can have natural conversations with them. These felt much less weird than they have in the past. It felt like I was talking to someone relatively naturally. So they're getting better, and that technology is becoming more ubiquitous. 

What theme stood out this year? 

We saw a lot of technology that is going to really usher in the age of “do it for me.” When I look at what our customers’ expectations are of U.S. Bank, we want to have the best digital experience, and a lot of that right now is “do it for yourself.” I want to be able to do everything on my mobile phone, my tablet, online banking; don't want to have to go to the branch unless I actually want to talk to a financial adviser.

At some point, automation is going to come into play, even in financial services, once customers become more comfortable ceding control. So we're looking for what the iPhone-like defining moment is, which shifts the customer expectation bar from “do it yourself” to “do it for me.”

How do you expect we’ll get to a place of consumer comfort with the “do it for me” mindset? 

I think you’re going to see a lot of things change in the marketplace, around customer experiences – not in the banking space, but in others – where you don’t need quite that level of security and you’re going to get a lot of convenience. When that happens, when people begin to cede some of that control to automation in those areas, they’re going to begin to expect it in other areas, maybe with things like bill pay. 

We’re able to poll consumer sentiment and not put stuff out into market before people are ready for it. There’s a really fine line between being proactive and helpful, versus being proactive and creepy. And we don’t want to cross the creepy line. And, obviously, customers will have to want to opt in to these types of things.  

What’s the team download like, after attending CES? 

We have a backlog of things we’re exploring that are tied to strategic business priorities with the bank. When we come back from something like this, we’ll look at [and ask] is something more ready for prime time than we thought it was a year ago? We have a list of trends that we’re tracking market signals on, to know when those technologies are going to be impactful for the banking space. So we take a look at those and reshuffle them in priority. We had a lot of foundation futures around agentic AI that we thought were probably several years out, until ChatGPT came out, and that really changed our time horizon on that.

How does your team balance the drive to innovate with risk management and trust considerations that are crucial for banks?

We have a healthy relationship with our risk partners. They join my staff meetings every week, and we’ve got partners in digital risk that are active in anything we’re exploring. We’re one of the longest-running innovation teams in banking, at about 17 years now. We’ve got a lot of safe test-and-learn environments, and then we work and partner with our risk and control teams to de-risk proof of concepts before they go into production. 

There’s definitely a healthy tension, and we don’t get to just release stuff quickly. We’re often looking at things many years before they come out into market. But there is a robust risk and control process that they go through. They’re there to help us solve things. We’re even … beginning to bring them into ideation sessions. Having their point of view is useful to getting to a workable solution sooner.