Behind banking’s AI divide: What’s keeping leaders ahead

Banks are accelerating their use of AI, but evidence of its financial payoff is still scarce.

The 2026 Evident AI Index, which benchmarks AI maturity at 50 major banks, finds that average scores rose 26% from last year, while JPMorgan Chase, Capital One and Royal Bank of Canada retained the top three positions. Leading banks are embedding AI more deeply into core operations and investing in the people, infrastructure and controls needed to expand its use. Yet only about 1% of the more than 1,100 use cases with reported outcomes tracked since 2021 disclosed a direct financial result, underscoring how much banks still have to prove as adoption grows.

The FR editors spoke with Daniel Shackleford, managing director of banking at Evident, to discuss what the data reveals about the future of banking and AI.

Banks’ average scores on your index, which measures AI maturity across talent, innovation, leadership and transparency, rose 26% this year. What’s driving that progress, and why hasn’t it been enough to dislodge the same banks at the top?

This year marked an inflection point from experimentation to industrialization. Banks are investing in the talent, infrastructure, and governance needed to deploy AI at scale, while expanding the number and sophistication of use cases across their businesses. 

But the leaders aren’t standing still. Banks like JPMorganChase, Capital One, and RBC have spent years building proprietary capabilities that allow them to move faster, deploy more broadly, and increasingly demonstrate measurable returns. So while the entire industry is becoming more AI-mature, the advantages accumulated by early movers are proving difficult to replicate. The real question now is whether second movers can accelerate quickly enough to close that gap.

Last year, Evident told The FR that AI’s impact might not reach banks’ bottom lines until 2027. This year, only about 1% of tracked use cases with reported outcomes disclose a direct financial result. Does the evidence still support that timeline, and what should investors accept as proof that productivity gains are translating into financial returns?

The evidence broadly supports that timeline. Although some banks are already reporting regularly on aggregate AI ROI, we expect the financial impact to become increasingly visible through 2027 and beyond. We’ve seen 13 of the 50 banks (26%) report AI ROI figures, up from only eight (16%) this time last year.

Banks are already reporting significant productivity improvements, but translating time saved into lower operating costs or additional revenue takes longer. A 30% improvement in employee productivity doesn’t automatically translate into a 30% reduction in costs, particularly if the bank hasn’t redesigned the underlying workflow. 

Also, revenue generation is critical. Leaders are making clear that AI is not only a productivity play but also a significant growth lever. Ultimately, the strongest proof will come from banks that can connect individual AI deployments to realized financial value, net of implementation and operating costs.

The report identifies underwriting, fraud investigations and customer servicing as areas with substantial potential for AI. Where are you seeing the strongest evidence that banks are changing how those businesses operate, and what is keeping those changes from spreading beyond the leaders?

Customer service is arguably where we’re seeing the most ambitious approaches to business transformation in production. These range from retail banking chatbots that provide services beyond banking, like Capital One’s Chat Concierge, to AI avatar-based advisory services in wealth management, like CitySky, and investment banking, like UBS’s research AI clones. We’re also seeing much-improved natural language voice bots for customer service across multiple banks. However, although less visible, AI-enabled underwriting, agentic fraud investigations and, more broadly, agentic exception management are transforming these processes at their core.

Scaling these applications requires more than the tool itself. Banks need integrated data, redesigned workflows, robust controls and the organizational capacity to deploy AI consistently across the enterprise. A lack of these prerequisites often prevents the benefits from spreading from leaders to laggards.

As AI models become more interchangeable, your report argues that banks’ advantage increasingly lies in their proprietary data, integrations and controls. How should that change what banks build themselves versus buy from fintechs and technology providers?

Banks shouldn’t be trying to build everything themselves, particularly when increasingly powerful foundation models and productivity tools are available off the shelf. The real opportunity for differentiation lies in what banks build around those models: proprietary data assets, reusable infrastructure, deep integrations with core systems, and controls tailored to regulated financial workflows. These capabilities determine whether AI can move beyond isolated applications to become embedded in how the bank operates.We expect leading banks to increasingly adopt a hybrid approach, buying commoditized capabilities while retaining ownership of the architecture, orchestration and intellectual property that underpin their competitive advantage. In other words, the model itself is becoming less important than the bank’s ability to put it to work.

Eighty percent of the top 10 banks disclose controls for specific AI use cases, compared with 40% of the remaining banks. As AI agents take on more complex workflows, what evidence should bank leaders require before granting them greater authority, and how much does this disclosure gap tell us about banks’ readiness?

As AI agents move from supporting individual tasks to executing complex workflows, the challenge shifts from evaluating the accuracy of individual outputs to controlling how systems behave across multiple steps. 

Banks need evidence that agents can operate reliably within defined boundaries, including rigorous testing, continuous monitoring, clear escalation procedures, and the ability to intervene or shut down an agent when necessary. Critically, these controls need to be embedded directly into the technology rather than relying solely on human oversight after the fact. The disclosure gap is revealing because it suggests leading banks are further along in translating responsible AI principles into practical, use-case-specific safeguards. Disclosure isn’t definitive proof of readiness, but it is an important signal that banks are developing the governance capabilities needed to deploy more autonomous systems safely.