Something extraordinary is happening inside the world’s banks, and most customers will never see it. Artificial intelligence systems are beginning to act, not just advise. They are investigating suspicious transactions, flagging refinancing opportunities, and engaging customers, all without a human initiating the request. Welcome to the era of agentic AI.
Unlike generative AI tools that respond when asked, agentic AI takes initiative. It interprets objectives, coordinates actions across systems, and operates within defined guardrails 24 hours a day, with no fatigue or waiting. For financial institutions, this represents a shift as fundamental as the introduction of ATMs or online banking. Possibly more so.
A major new research report, Agentic AI: Powering the Self-Driving Bank, sponsored by OpenText and authored by Jim Marous of the Digital Banking Report, surveyed financial institutions worldwide to reveal exactly where the industry stands, and where it is at risk of falling behind.
The biggest risk with agentic AI in banking is not moving too fast. It is mistaking pilots for progress.
Get the agentic AI banking report
The pilot trap: Busy but not moving
The headline from the research sounds encouraging: 96% of banking executives report their institutions are engaged with agentic AI in some form. But the real story lives beneath that number.
This means roughly four in five institutions are still somewhere between curiosity and testing. The report’s most sobering observation is that piloting has become a comfort zone, a state that looks and feels like progress, but doesn’t actually change anything for customers or operations.
This “pilot trap” is not unique to AI. The report notes that 93% of banks have fallen short in their broader digital transformation ambitions, the same pattern of enthusiasm followed by stalled execution that has repeated across every major technology cycle in financial services.
MIT Technology Review Insights found a parallel distribution: in their 2025 survey of 250 banking executives, 52% reported piloting agentic AI, while only 16% had reached deployment, corroborating the Digital Banking Report’s findings across independent research streams.
What agentic AI actually means for banking
For more than a decade, banks have used automation to reduce manual work at the task level: RPA for back-office processes, predictive models for credit and fraud, chatbots for routine service. These systems followed rules. They operated within narrow, predefined lanes.
Agentic AI is structurally different. Rather than waiting for instructions, an agentic system can interpret an objective, break it into steps, access multiple internal and external systems, and take action, all within pre-approved guardrails. The practical examples are striking:
- A fraud monitoring agent that detects an anomalous pattern, escalates to the right team, and initiates a protective hold, without waiting for a human review queue
- A lending agent that spots a refinancing opportunity, evaluates a borrower’s eligibility, assembles the recommendation, and initiates customer outreach, automatically
- A commercial banking agent that reprioritizes relationship manager workloads in real time as new signals emerge across hundreds of client accounts
As Citibank has described it, a “Do It For Me” economy is beginning to emerge, where agentic AI manages increasing complexity so that both customers and employees can focus on what matters most.
Where banks are (and aren’t) focusing
The research reveals a pattern of defensiveness in how banks are prioritizing their agentic AI investments. Institutions are starting where guardrails already exist and ROI is easiest to quantify.
Business priorities for agentic AI deployment
The pattern is telling. Banks are overwhelmingly applying agentic AI to protect and optimize, not to grow or empower. Revenue enhancement ranks near the bottom at 22%. Employee enablement, which underpins nearly every customer-facing improvement, sits last at just 14%.
There is a sharp internal contradiction here. Customer experience ranks fourth in priority at 45%, yet the employees who deliver that experience are barely being empowered by AI. In most banks, these two imperatives cannot be separated. The institutions that connect agentic AI to both sides of the experience equation will extract far more value than those treating them as separate initiatives.
McKinsey estimates that institutions leading in AI deployment could gain a compounding profitability advantage over slower-moving peers. IDC research suggests leading firms are achieving returns roughly three times higher than laggards on AI investments. The window for catching up, the report warns, is actively narrowing.
Three barriers that are holding the industry back
When asked to name their biggest challenges in scaling agentic AI, banking executives consistently pointed to three interconnected obstacles — none of which are about the technology itself.
Perhaps the most unsettling finding: while 56% cite regulatory uncertainty as their top barrier, the same survey reveals that 52% of institutions are not confident their current governance systems could support compliant agentic AI deployment. The industry is pointing at regulation as the problem while quietly acknowledging it isn’t ready for what regulators will eventually demand.
The data integration picture is equally sobering. Only 9% of respondents say they can “very effectively” connect content, communication, and transactional data to drive real-time decisions. Another 14% say “somewhat effectively.” That leaves 77% of the industry without the integrated data foundation that agentic AI fundamentally requires.
The road from pilot to production
The report’s conclusion is clear: strategic alignment is the single strongest predictor of whether agentic AI investments move beyond piloting. Deloitte’s research reinforces the point that technology creates value when entire workflows are redesigned, not when AI tools are layered onto legacy processes.
That kind of redesign requires board-level commitment, cross-functional authority, and governance frameworks built for autonomous decision-making, not compliance checkboxes, but genuine operating infrastructure. The report highlights DBS Bank’s PURE framework as a model: requiring all AI systems to be Purposeful, Unsurprising, Respectful, and Easy to explain, with real-time monitoring and automated kill switches for high-risk applications.
The institutions that will define competitive banking over the next decade are those investing now in API-driven architecture, unified data platforms, and governance frameworks designed for autonomous AI, not those waiting for regulatory clarity to arrive before building readiness.
Watch the report insights video
What this means for your institution
The report’s findings draw a stark line between two types of financial institutions. On one side: those with board-approved agentic AI strategies, modern data infrastructure, and governance frameworks that can support autonomous decision-making. On the other: those generating increasingly elaborate activity reports from perpetual pilots, while the competitive gap quietly widens.
The economics are unambiguous. KPMG projects $3 trillion in corporate productivity gains from agentic AI. McKinsey estimates it could reduce the banking cost base by 15–20%. IDC research suggests measurable returns are achievable within 13 months for institutions with the right foundation in place.
Yet the survey found that 57% of the banking industry expects only modest or minimal financial returns within three years, and 26% haven’t even defined what success would look like. The institutions reporting low expectations are, in many cases, simply the ones not yet measuring results.
The question every banking leader should be asking is not “should we be doing this?” The industry answered that with 96% engagement. The real question is: are we building the foundation to actually deploy it, or are we mistaking the comfort of pilots for the competitive advantage of progress?
Get the agentic AI banking report
