Five Steps to Smarter AI Decisions at Your Bank

Adopting AI can feel like a leap of faith—especially for small and mid-sized banks that may not have the resources or expertise of larger institutions. But according to ProSight’s Managing Director of Advanced Risk Services Ed DeMarco and Lori Calhoun, chief risk officer at Dollar Bank, it doesn’t have to be. In a recent discussion, they outlined five foundational steps that can help financial institutions vet AI use cases with confidence and avoid costly surprises. The guidance is based on lessons learned from creating the ProSight AI Use Case Template—a member-driven resource designed to bring structure and accountability to the process. 

• Involve the Right Stakeholders Early: Before you even consider a new AI tool, map out who should be at the table. According to DeMarco, the project lead “should really consider ... who the internal stakeholders will be,” and make sure that data owners and colleagues from IT, legal, and marketing have a seat at the table. Calhoun added that Dollar Bank builds cross-functional teams to evaluate each use case from the outset. “An ounce of prevention is certainly worth a lot,” she noted. 

• Document Your Assumptions: An AI tool may be deployed by different users for different circumstances. In order to determine whether deployment met expectations, it is important to identify expected outcomes including human resource needs. “AI is designed to create efficiency,” DeMarco said, “but it doesn’t mitigate the need for human talent,” especially in areas where human judgment is needed. Calhoun emphasized the need to clarify which AI features are relevant—don’t “assume that all of the AI capability that is embedded in something will be used.” 

• Identify and Rethink Risk: The usual suspects—legal, compliance, operational risk—still apply, but AI demands a sharper lens. “Even though we’re accustomed to dealing with legal risk … you have to look at that differently for AI,” said Calhoun. That means thinking carefully about bias, privacy, and how liability is handled in contracts and outputs. 

• Don’t Skip Vendor Due Diligence: “Really doing that due diligence and understanding … what is the input, what’s the output?” is essential, DeMarco said. Calhoun added that sometimes the AI provider isn’t the same as the primary vendor—so banks must “flesh out exactly who is providing the AI component, who has the liability.” 

• Define Success—and Set Limits: Don’t let success be subjective. Tie KPIs and KRIs to your original assumptions and monitor accordingly. “Use something like the use case template to have guardrails,” said DeMarco. And don’t forget, as Calhoun pointed out: AI approval for one department doesn’t mean another can adopt it by default. Each use case must go through its own risk review. 

Bottom line: AI can unlock big value—but only if the guardrails go up first. One resource developed to support this process is the ProSight AI Use Case Template, which helps institutions think through assumptions, risks, and responsibilities up front. Used thoughtfully, it can support better decisions, stronger oversight, and fewer surprises.