the credit union connection logo white

The AI governance paradox: More autonomy requires more context

Crystal Miceli_SVP of Brand and Product Marketing_Glia

Crystal Miceli, SVP of Brand and Product Marketing, Glia

For financial institutions, AI governance has largely become synonymous with AI limitation: limiting what AI can see, so it doesn’t improperly expose sensitive data or take unauthorized actions.

But as credit unions begin deploying agentic AI, they need to recognize an ironic twist: good governance of AI agents requires expanding, not just limiting, what those agents can see and access.

Because if you’re going to authorize an AI agent to execute banking actions on behalf of members, you’d better be sure it has all the context needed for a complete understanding of the situation.

That irony –  avoiding AI risk by expanding AI access – changes how credit unions need to think about connecting AI to their technology stacks.

Missing context was always a problem, but agentic AI raises the stakes

We already know what missing context feels like for members: They call their credit union with an issue, get transferred to another department and have to repeat everything. The next day they walk into a branch to finish the transaction, and they have to start all over again.

Financial institutions spent the last two decades working to eliminate these frustrating experiences and deliver the now-clichĂ©d “seamless omnichannel member experience.” That progress has improved member experiences, with meaningful impacts on member retention and loyalty.

But at the end of the day, we’re still talking about an emotional outcome: annoying members with unnecessary friction.  

AI raises the stakes. If an AI agent has authority to act but lacks context, it could execute against incomplete inferences. That’s much worse than annoying — it could produce real negative financial outcomes.

Imagine a member calls to replace a debit card after a suspected fraud incident. An AI agent can initiate a replacement, but if it’s not aware the member previously reported the account itself as compromised, it will simply replace the card without addressing the larger issue (and allow the fraud to persist).

That’s how we get to the ironic shift in AI governance:  If you’re going to give AI agents authority to act, you need to be sure they have the full context to determine and execute the right action.  

Most credit unions have the same context gaps

Credit unions have made tremendous progress connecting data silos, helping teams see fuller pictures of members so they can deliver seamless omnichannel experiences. But there’s still one disconnect: structured and unstructured data typically live in separate systems that don’t connect. 

Core banking systems house the structured data: account numbers, transaction histories, credit limits and so on. They know the “what” of banking interactions. The “why” behind those interactions lives within unstructured data (the conversations that members have through chat, email, text, etc).  

Neither of these data streams is enough on its own: a core banking system can tell you that a member cleared out their checking account yesterday, but a member calling or chatting to report fraud gives you the information you need to understand what to do next.

That’s why, whether it’s a human team member or an AI agent, the starting point for efficient, effective and safe actions should be a current and complete picture of everything the member has done and said.

The branch remains a huge context blind spot

The best example of this context gap is the branch itself. More than a decade after branches were widely declared “dead,” they remain one of the most important touchpoints in the member experience. The branch is where the richest member conversations happen.  

Historically, it’s also been the biggest blind spot for credit unions. Those rich member conversations live only in the memory of the frontline team members that were part of them. They aren’t preserved in any form of persistent institutional knowledge.

A member might spend 30 minutes talking to a branch employee about what they’d like to do (making a large transfer, shifting some money into other savings products to get higher return, talking through loan options). When they walk out, the only thing the credit union sees is the transaction: the transfer, the new certificate or the auto loan with the three-year term. All the valuable nuance on the “why” is lost.

For a human employee handling the member’s next interaction, that missing context is a disadvantage. For an AI agent authorized to act on behalf of the member, it could be a dangerous blind spot.

Credit unions can’t give AI full banking context with generic controls

This is where things get tricky for credit unions: Giving AI agents full context means giving them full access to extraordinarily sensitive information. Account data, transaction histories, personal financial circumstances and private conversations may all be relevant to completing a particular workflow.  

You can’t just connect a general-purpose AI tool to everything in your tech stack and hope for the best.  

AI governance still requires strict control over what an agent can access and do with information, including what actions it can perform, what information must be verified before executing an action, where human approval is still necessary, and what all gets recorded for monitoring, auditing and potential investigation.

And the reality is that building those controls from scratch is very expensive and complex, and therefore largely impractical for all but the largest financial institutions.

Credit unions need agentic AI built for banking

As the broader AI market matures, AI platforms built specifically for banking are dramatically lowering the barrier to entry for credit unions and other community financial institutions. These purpose-built banking AI platforms are tackling many of the biggest AI governance challenges head-on, and doing it at a cost that CUs can afford.  

For example, instead of leaving credit unions to manually integrate their many structured and unstructured data streams, top banking platforms are creating a new kind of relationship management system that automatically brings together structured core banking data and unstructured interaction histories. That’s the idea behind Glia Relationship Management: giving all actors within a credit union — human and AI — a unified understanding of each member across voice and digital channels.

The glaring branch blind spot is also closing. New tools like Glia Branch capture the valuable context from branch conversations, automatically incorporating it into the broader member profile.  

And to address the security and compliance needs that come with giving AI agents this deep contextual understanding, banking-specific platforms incorporate the pre-built security controls, workflow requirements and human oversight capabilities that credit unions need to safely and reliably handle highly sensitive data and control agentic actions.

Agentic AI will level the playing field for credit unions

Credit unions and other community financial institutions have always tried to keep pace with the biggest banks on tech-driven experience. Agentic AI is a once-in-a-generation opportunity to not just catch up, but lean into community banking’s longstanding differentiator: AI agents enable credit unions to scale their deeply personalized member experiences.

But context is the key to capturing this opportunity. A deeper understanding of members is the foundation for deeply personalized member experiences. This remains true in the age of agentic AI: Agents need full context to bring that same understanding to agent-driven experiences and, eventually, autonomous actions.

On the surface, this can feel like a contradiction: Addressing concerns about AI risk starts with giving AI agents more access, not less.

But this paradox demonstrates the importance of choosing the right AI platform: When you have confidence that your AI platform gives you the controls you need to protect sensitive member information and transactions, granting access and authorizing actions creates opportunity instead of risk.  

About The Author

Crystal Miceli is the SVP of Brand and Product Marketing at Glia, where she defines the narrative for the leading platform for intelligent banking interactions. At Glia, Crystal bridges the gap between sophisticated AI and human-centric service, ensuring financial institutions can make customer care more efficient without losing the personal touch. A veteran leader with over 30 years of experience across engineering and marketing at giants like Cisco, Broadcom, and Talkdesk, she specializes in translating complex tech into high-impact business value. Crystal holds a BS in computer science from Tulane and an MBA from LSU. When she isn’t scaling the Glia brand, she is likely navigating the skies as a certified private pilot or volunteering with the Junior League of New Orleans.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top