The Agentic AI Challenge: Solve for Both Efficiency and Trust
By Nicole Volpe, Contributor at The Financial Brand
Simple Subscribe
Subscribe Now!
Agentic AI promises to open new horizons of efficiency and growth to even the smallest financial institutions, yet most don’t know where to begin and many lack the confidence to get started.
In a conversation at the 2026 Financial Brand Forum (“From Probability to Certainty: Predictive vs Deterministic AI Modeling for Financial Services”), James Dotter and Derek White unpacked the challenge, offering bank and credit union leaders strategic and tactical insights into the choices they must make in order to realize their agentic AI ambitions.
Dotter is Chief Business Officer at MX, a technology firm that specializes in helping financial institutions harness their data. White is the founder of Primitive, a startup focused on translating agentic AI into applications for regulated financial services. The discussion included first principles that can help strategists stay anchored in their mission and values as well as concrete guidance for those ready to take their first or next steps.
Want more insights like these? Check out MX’s content hub: Data in Action
Process Complexity, Human Amplification
Small financial institutions looking to implement AI often face a three-way bind: They operate under regulatory and risk expectations similar to those of large banks, and many of the same legacy-system and data-silo challenges, but with far less data engineering capacity and budget. As a result, though community banks and credit unions may envision many possible AI use cases, they often lack capacity to evaluate them.
Such institutions should narrow their focus, White said, targeting the internal workflows that agentic AI is classically built to handle: processes that are complex, cross-functional, and data-heavy, requiring multiple approvals and handoffs with extensive documentation requirements.
They should also tread carefully when it comes to customer-facing use cases, especially agents transacting on behalf of the consumer. Digital interfaces are an unnatural way for humans to communicate — and AI agents that communicate conversationally are closer to how trust actually develops than the rigid interfaces banks typically rely on. Deploying consumer-facing agents in an informational, conversational mode first, while limiting the AI agent’s permission to act, lets that trust develop before the stakes get higher.
At the same time, White argued that financial institutions should not confine their early AI efforts to back-office efficiencies, tempting as that may be for a resource-constrained organization. The bigger opportunity is growth: deploying agents that enable team members to identify opportunities, improve decisionmaking, and act faster on priorities.
“My hypothesis is that in the very near future, there will be more AI agents in banks than there are humans,” White said. He described working with a 700-employee community bank — no on-staff developers, all development outsourced — that already has three AI agents deployed. “They’re leaning in very hard on the opportunity of leveraging agents because their CEO and chairman has prioritized top-line growth.”
White offered the example of a high-performing commercial banker at a community institution: someone with strong local relationships and judgment, but limited time to work through the multistep process required to move deals forward. The opportunity for agentic AI, he said, is to reduce the coordination burden around that work by helping assemble data, documentation, and other inputs across a process that typically may stretch for weeks.
The aim is to expand the banker’s capacity so that she can spend more time guiding clients and closing business. “You don’t want to take that community banker away from their engagement with your customers,” Dotter said. “You want to amplify their engagement.”
Another growth-oriented application is marketing campaign planning and execution, which can require coordination across dozens of personas, multiple data inputs, and cross-functional sign-offs. Dotter argued that AI can help address the “analysis paralysis” that often results from that complexity by mining the data, separating signal from noise, and surfacing the strongest options. As with the commercial banker, the technology amplifies the marketer’s work, in this case by supporting faster, better-informed execution.
Risk and Control Foundations
When it comes to managing AI risk, data security and privacy are only part of the challenge. More fundamental is the gap between probabilistic and deterministic systems. Large language models and AI agents are probabilistic, while the systems banks and credit unions rely on to run their operations are deterministic. “There are certain things within finance you don’t want to have probabilities around,” Dotter said. “You don’t want a ‘maybe’ in an account balance.”
Establishing effective AI controls is fundamentally about defining how probabilistic tools can operate safely alongside deterministic banking systems. Large language models and agentic tools can be effective at pattern recognition and recommendation, but banking depends on ledgers and core systems — underwriting and fraud controls built for fixed outcomes and auditability. Is this transaction in compliance? Do these loans meet our risk criteria? How do I validate that?
To ensure the right risk and control foundations are in place, White recommends institutions focus on three areas: gateways, guardrails, and governance.
Gateways are the controlled entry points through which external model intelligence enters the institution and connects with internal systems and data, which needs to be kept secure. Institutions may need automation around data hygiene, for example, including monitoring for anomalies that could signal data-integrity problems.
Guardrails come next. This describes the controls that limit what the model can “see” and do, including restrictions on how customer information is exposed or shared. Once outside intelligence is allowed in, guardrails determine the conditions under which it can be used.
And finally, governance is the operating structure that encompasses gateways and guardrails, and describes the organization’s overall vision for AI risk control. It includes the people, processes, business lines, and approval paths that determine how AI is used and who is accountable for it.
Vision and Execution
Both speakers flagged data readiness as a prerequisite; the need for institutions to establish a strong data foundation before asking AI to do meaningful work. In practical terms, that means centralizing consumer account and transaction data, and enriching it such that it will yield useful insights. Other guidelines included:
Put security and data protection at the front of the process. White said: “The lead person when bringing AI into a bank, or into any organization, right now is the CISO, because the CISO critically must be able to define that the institution is going to be protected when customer data meets external data.”
Map AI workflows to the institution’s operating structure. White stressed that governance is tied to “people, organizations, P&Ls,” and that AI has to map to the bank’s risk and control environment, organizational structure, and approval processes. In other words, governance cannot sit apart from the institution’s actual lines of business and accountability.
Design human decision points into workflows from the start. It’s tempting to think of humans in the loop simply as a backstop, as if they are primarily fact-checkers for AI’s work. But the better frame is a higher-order one: distinguish early on the work that humans uniquely can optimize from the work that AI uniquely can optimize. “It’s essential to inject humans in the loop where you want them,” said White.
For community banks and credit unions, the last point may resonate most. When AI agents reach a point of mass adoption in banking, they will have changed the industry in fundamental ways that will be hard to unwind.
Invoking a world in which we will all be “banking with Anthropic or OpenAI or Apple or Google,” Dotter said: “We have a moral responsibility to ensure that the foundational models that have been built that deliver financial stability in our society persist into the future. And that they’re not replaced by AI, but that they partner with AI.”
