How Small Financial Institutions Can Reset to Keep Pace with the AI Opportunity
By Nicole Volpe, Contributor at The Financial Brand
Simple Subscribe
Subscribe Now!
For much of the past three years, the business story around artificial intelligence has centered on experimentation: running pilots at the individual or team level and looking for quick productivity wins. More recently, however, a new and more demanding phase of AI adoption has been gathering steam, one in which organizations convert their AI learnings into durable use cases and sustainable advantages.
McKinsey calls this next step “rewiring to capture value.” Deloitte calls it moving “from ambition to activation” — and projects that the number of companies with ≥40% of AI projects in production will rise sharply in 2026.
Perhaps not surprisingly, smaller banks and credit unions have been slower to make the next-level AI transition. A 2026 survey by the American Bankers Association revealed that, even though banks that use AI report meaningful productivity gains, most are prioritizing low-risk applications that support human judgment. They continue to cite data readiness, governance, regulatory uncertainty and information security as challenges to adoption. Meanwhile, December guidance from NCUA identifies algorithmic opacity, fair lending concerns, data privacy and security, operational resilience, and model risk as barriers to responsible AI use.
Yet small financial institutions increasingly acknowledge that they can’t continue to dabble. Policy groups and regulators increasingly see inaction as a risk in itself. In ABA’s survey, community institutions cited AI as their most important issue to address, followed by cybersecurity and data privacy — but respondents said their greatest risk is doing nothing as they decide how to integrate AI into their business models.
For smaller institutions, moving AI into production requires a new operating framework and philosophy that suits their business model, while also borrowing from the agile cultures of many startups. It includes rethinking approaches to readiness and cost structure, and reassessing the role of their technology partners and what it takes to identify use cases worth investing in. Following are new guidelines that bank and credit union leaders can adopt as they get ready to make the leap.
To explore the topic, The Financial Brand spoke with Narmi Chief Technology Officer Nathan Gonzalez, whose career has spanned emerging technology companies and established financial technology providers.
Want to read more like this? Check out Narmi’s content portal on The Financial Brand: Be Where Banking is Going
1. Let Workflows Set the AI Agenda
Many business leaders still imagine AI’s end state as a universally useful, always-available human-like platform built on one of today’s frontier models. That assumption helps explain the default tendency to “platform-first thinking.” But for smaller banks and credit unions, the better place to begin is with their current workflows.
This means an institution must closely evaluate the specific processes it wants to upgrade or transform — for example account opening, loan-document review, or back-office exception handling. In each case, financial institution strategists should not focus on how to move the work onto an AI platform but on how to translate a specific workflow to an AI-based solution.
NIST’s AI Risk Management Framework, which many small institutions use as a guide, emphasizes defining and documenting the specific business purpose and context for an AI system before assessing risk or value. The tighter the workflow definition, the easier it becomes to define the data the model needs, the controls that must remain in place, the outcome it’s targeting, and how performance should be measured.
According to Gonzalez, this “workflow-first approach” helps keep AI investment grounded in operational value. If account opening requires repeated manual intervention, can AI help identify missing documentation or route exceptions more accurately? If call-center teams spend excessive time searching across siloed systems, can AI surface answers in-the-moment based on clues harvested via ambient listening? If fraud review requires comparing disparate data and information sources, can AI funnel those signals into a consolidated decision flow?
2. In the Long Run, Customization Can Reduce Costs and Risks
At first blush, smaller banks and credit unions may expect software customization to be the most expensive option. But in practice, generic AI tools that don’t suit an institution’s workflow are likely to lead to cost overruns and unforeseen workflow gaps later on. In fact, according to Gonzalez, the whole purpose of an AI agent is to be a custom solution — which is why, in his view, “generic AI solutions just don’t work well.”
An account-opening workflow, for example, is defined by each institution’s internal policies, which in turn impact data handling and decision points across sales, marketing, risk management, and compliance. All of these vary from one institution to another, not just in terms of settings and inputs but in terms of process: The points of differentiation are far more important than the points of commonality. A generic tool, even a generic banking tool, will ultimately fall short in meeting the institution’s needs.
To be sure, Gonzalez does not advise hiring multiple engineers to build large quantities of highly complex agents that the institution then must support with internal resources going forward. Instead, he recommends targeting solutions that are “80% productized and 20% custom,” with a partner able to “finish the hard parts of your workflow.”
3. Make Experimentation a Permanent Best Practice
Institutions that advance from experimentation to production do not leave experimentation behind. In fact, those adopting AI should give “start-up” attributes like experimentation and agility permanent and central places in their cultures.
As an organization advances along the adoption curve, AI makes it easier to test more ideas earlier, with less upfront commitment, and learn which ones deserve further investment. The idea is to test diversely and build selectively.
Interestingly, increasing AI adoption may also trigger a role reversal within your organization: In a competitively-oriented bank or credit union, marketing typically complains that technology is the bottleneck in its push to bring new products or processes to market. But, Gonzalez said, as AI-based systems take root, software teams find themselves able to move from concept to production far more quickly so the new bottleneck becomes the go-to-market operation — whether compliance review, or training, or customer communication.
To get ahead of this new bottleneck, Gonzalez suggests a forward-deployed engineering approach: where system architects and developers sit alongside client-facing teams to deeply learn workflows. For smaller institutions, that kind of close collaboration can help experimentation stay connected to implementation, while also deepening the organizations’ trust and fluency in what AI can do. Of course, smaller institutions may not have large IT staffs to begin with, but they can seek such support from their tech partners.
4. Put the Burden of Proof on Your Partners
Finally, small banks and credit unions should not abandon their fundamentally cautious approach to AI. But that caution should be productively channelled, in particular by demanding system proofs and validation from their vendors.
In practice, that may mean testing an AI process against historical institutional data, using prior outcome data as a source of truth to test an AI tool’s performance and ability to adhere to the institution’s standards. A similar standard can apply to handling of data that includes PII or other information with sensitive regulatory requirements.
Cost control is another component of the proof discussion. Is model routing an option? — matching each task to the right model, rather than using the most expensive model for every request. This can guard against the Trojan horse effect in which AI costs skyrocket once the system is implemented and usage takes hold.
The final proof point is data readiness. Gonzalez calls data “the enabler” of the AI future for community financial institutions, but says institutions still need to make that data usable through cleaning, organization and strategy. Without that foundation, even a capable AI tool will have limited value.
For small institutions eyeing enterprise AI deployment, data readiness means establishing a centralized, clean data repository that ensures the information AI agents require is accurate, safe, and accessible. Prioritizing continuous data governance and strict quality standards early on prevents the common pitfall of feeding automated systems flawed data, laying a secure foundation for scalable AI integration.
Where the Action Is
AI has often been discussed as a choice between cost reduction and growth: either a way to automate work and lower expenses, or a way to improve performance, move faster, and create new capabilities. That tension becomes sharper as organizations move from experiments to enterprise adoption. Menlo Ventures estimates that enterprise spending on generative AI rose from $11.5 billion in 2024 to $37 billion in 2025, with $19 billion going to AI applications alone. In the near term, even “cost-saving” AI requires new spending on software, workflows, infrastructure and operating capacity.
For smaller banks and credit unions, the lesson is not to follow the largest enterprise spenders into a broader AI arms race. It is to apply the best habits of smaller, more agile companies: stay close to the work, test quickly, demand evidence from partners and invest where AI can improve a high-impact operating process. The institutions that get the most from AI will be those that leverage their inherent constraints to focus on the use cases where integration can produce measurable value.
