Why “AI everything” is the wrong starting point for banks

Almost every financial institution leader I talk to feels the same pressure. AI is in every headline, every conference session, and every board meeting, and the message coming through is that an institution needs an AI strategy immediately or it will be left behind. The natural response for many teams is to look for a platform-wide transformation that modernizes everything at once..That instinct is understandable, and it is also where many good intentions stall. S&P reported that more than half of financial firms reported active AI initiatives in 2025, up from 40 percent the previous year, yet only 59 percent of deployed AI initiatives are generating measurable business value. When AI is framed as an “everything” decision, it becomes too large to start, too risky to govern, and too abstract to measure. The institutions that are seeing meaningful returns took a more focused approach. They ignored the noise, picked the two or three workflows where the work stacks up, and applied AI there first.The evidence backs this upAccenture found that only 8 percent of banks qualify as AI front-runners, or ones actually seeing strong returns, even as adoption increases. A Wolf & Company survey of community banks put it more bluntly. Every institution surveyed has adopted AI in some form, yet a large share of that adoption is limited to ad hoc or vendor-embedded tools rather than anything deliberate. The gap is not between banks that use AI and banks that do not. It is between banks with a focused plan and banks chasing the concept.The good news for community banks is that focus is a strength they already have. These banks know their customers, their markets, and their operations in detail. That same discipline applies perfectly to AI. Success does not come from becoming an AI bank. It comes from finding where the work slows down and applying AI to specific tasks where it can make a measurable difference.Start where the work slows downFor most institutions, the greatest opportunities are in onboarding and origination, especially within business banking. Opening complex business accounts, spreading borrower financials, and preparing credit memos can consume significant amounts of skilled staff time. Underwriters and bank staff can spend as much as 60-70 percent of their day gathering and assembling data rather than exercising judgment, and a single business loan file can run hundreds of pages. Manual spreading alone can take a day or more before any analysis begins. None of that work requires the professional judgment institutions actually pay for. It is mechanical, repetitive, and exactly what AI does well.This is the practical use case for AI and a much easier one to act on. Banks can start with document intelligence that classifies submitted documents, extracts relevant information, and validates the data for completeness and consistency. That information can then help prepopulate applications and assemble business ownership and financial information. From there, institutions can expand into automated financial spreading, AI-assisted credit memo preparation, and BSA/KYC narratives. Each capability addresses a defined bottleneck, creates a measurable result and keeps bank employees firmly in control of final decisions.Governance makes it defensibleKeeping that human in control is the governance backbone that holds all of this up. Examiners increasingly expect audit-trail and source-citation discipline as a baseline requirement. Every extracted figure should trace back to its source document, exceptions and employee overrides should be documented, and every AI-generated output should be reviewable from beginning to end. That discipline is achievable when AI is scoped to bounded, well-documented tasks, and nearly impossible when it is scoped to everything at once.The fraud case for moving nowThere is a second reason to move deliberately and now: fraud. The same tools that let institutions work faster are also available to bad actors, who now use AI to generate documents convincing enough to pass human review. That risk compounds in business banking, where a single relationship can involve several owners and multiple business entities, each of which must be verified. The most effective defense is layered. Institutions need to treat every document as one signal among many, verify each individual and business, validate ownership and identifying information, and use integrated identity, fraud, and risk controls to identify inconsistencies or unusual activity. A sudden surge from a handful of applications to thousands, for example, is the type of pattern technology can identify quickly and escalate for review.Expand with purposeAs institutions gain experience, the same, focused approach can extend to explainable risk scoring, policy-aligned decisioning, and natural-language access to portfolio and operational data. The principle remains the same: apply AI to a defined problem, produce a result employees can understand and review, and keep the technology aligned with the institution’s policies and risk appetite.Build in modules, not one leapWhat ties all this together is that these capabilities do not have to be adopted as a single leap. The most sustainable path treats AI as a set of modules that plug into the workflows an institution already runs. A bank can begin with document intelligence in account opening, prove the value, and expand into automated spreading, credit memos, and eventually more ambitious capabilities. That measured expansion is how an institution builds real organizational muscle around AI, rather than buying a grand vision that sits unused.The most effective approach keeps AI close to the onboarding and lending work banks already do every day, built into the process rather than bolted on beside it as a separate initiative. It is an approach we have taken, and the institutions that succeed with it tend to be the ones that found a specific bottleneck, removed it, and moved on to the next rather than the ones with the boldest AI mission statement.The next obvious stepWhere AI in banking is headed will keep changing, and no institution should pretend otherwise. What will not change is the value of starting with a specific problem. Find the two or three places where skilled employees spend their days on repetitive work that technology can do faster and more consistently. Establish governance before deployment, not after. Give your staff their time back so they can spend it on customers and judgment. Do that, and the larger AI strategy everyone keeps asking about will not feel like a leap. It will feel like the next obvious step.Glenn Bolstad is CEO of Vikar Technologies, a banking technology company that unifies the full client lifecycle for banks and credit unions on one platform, spanning account opening and deposits, commercial and SMB lending, treasury management, and wealth management.