Banks and credit unions have historically treated the contact center largely as a routing problem. A customer calls, the IVR identifies a broad category of need, the interaction moves through a decision tree, and if self-service fails, the customer eventually reaches someone who tries to determine what they were actually trying to accomplish.
Most of us know the experience. At some point, you stop participating in the menu and start repeatedly saying “operator” until the system finally gives up on you.
That infrastructure that platforms like Genesys and NiCE provide still matters because itE solves very real problems around routing, workforce management, escalation, recording, compliance and service operations.
The interesting question is not whether AI replaces that stack, but whether the IVR should continue to be the first system trying to understand the customer. Increasingly, I don’t think it should.
I’ve been having more conversations with large financial institutions and technology partners about where agentic AI can create meaningful value in banking, and contact centers keep surfacing. The reason is that there is now a better way to understand intent before traditional routing logic takes over.
Instead of beginning with “Press 1 for checking, press 2 for credit cards,” imagine beginning with a much simpler question: “How can I help?”
A conversational agent can understand that the customer is not simply calling about a credit card. They are calling because a payment appears twice, their available balance looks wrong, they are traveling tomorrow, and they are worried the card could be declined.
Naturally, context changes the experience. The existing contact-center stack can still support authentication, routing, queuing, escalation and connection to the appropriate resources. But instead of receiving a menu selection, it receives something much more useful: intent and context.
Deflection may be the wrong ambition
I keep hearing “call deflection” discussed as one of the most attractive contact-center metrics. The economics are obvious: a customer who successfully self-serves generally costs less to support than one who requires human assistance.
But avoiding a call and resolving a problem are not the same thing.
That distinction becomes much more important with AI. The industry is increasingly moving toward outcome-based containment, where success is determined by whether the customer accomplished what they needed to accomplish, not simply whether the issue was resolved without human intervention.
The agent’s job should not be to keep customers away from the contact center. Its job should be to find the best path to resolution. Sometimes that path will be completely automated. Sometimes it will require a specialist. And sometimes the best outcome will be moving the customer out of voice and into authenticated digital banking, where the task can actually be completed.
A pizza order recently reminded me why that distinction matters.
What happens when every support channel starts from zero: A pizza made the point better than I could
I’m going to put an unnamed national pizza provider on blast for a moment.
Last Sunday, I ordered pizza for my family during a game. More than an hour and a half later, there was still no pizza. The app’s tracker had been permanently frozen on “bake.” The store was not answering its phone. My increasingly hungry household was beginning to question my ability to perform the fairly basic responsibility of acquiring dinner.
So I tried another channel.
I called the company’s national customer-service number. Then I tried another automated experience, followed by a feedback form. What followed was effectively a one-sided argument between me and several electronic systems, none of which appeared to know what the others had already been told.
Eventually, I ordered another pizza.
The remarkable part is that this company did not lack digital tools. It had an app, a tracker, a phone number, automated assistants and complaint forms.
It had channels everywhere, but had continuity nowhere.
That is the part banks should pay attention to.
My problem was not that I desperately needed to speak to a human being. I just needed something in the ecosystem to understand: Marc ordered a pizza 90 minutes ago, the store has not completed the order, the tracker has stopped providing useful information, and he would now like to know whether dinner is actually coming.
Any channel could have solved that problem if the context had traveled with me.
Instead, every new channel effectively asked me to begin again.
Banks have the same opportunity, with much higher stakes
Now replace the missing pizza with an unfamiliar transaction.
A customer calls the bank because they see a charge they do not recognize. An intelligent agent understands what they are asking, identifies the transaction and determines that the next action requires authentication.
Rather than forcing the customer through another menu or beginning a completely separate process, the agent offers to continue inside the bank’s mobile app.
The customer opens the app and the transaction has already surfaced. The context from the conversation is preserved. The appropriate workflow is ready. They authenticate, complete the next step and move on.
Voice was simply where the customer chose to explain the problem. Digital banking became where the bank resolved it.
That is a very different experience from calling, explaining the issue, being told to open the app, navigating to the appropriate screen and effectively beginning again.
More importantly, the customer does not need to keep reintroducing themselves to their bank.
That is why I think the role of the agent is larger than building a smarter phone bot. The agent becomes connective tissue among channels.
Voice captures intent. The agent understands the task. The contact-center platform handles what it already does well. Authentication happens where it should. Digital banking becomes the place where appropriate journeys continue.
Notice how these systems are not competing with each other. Instead, the bank remembers what happened before the handoff.
Every conversation contains more intelligence than a disposition code
There is another benefit that may prove just as important.
Contact-center analytics have traditionally been useful but often retrospective. Calls are categorized, recordings are analyzed, dispositions are entered, and teams study broad trends in volume, handle time, escalation and service levels.
An agentic layer at the beginning of the interaction creates the potential to make that information much richer and more immediate.
There is a meaningful difference between knowing that someone called about a “credit card” and knowing that a customer is concerned about a duplicate transaction, is traveling internationally tomorrow and has already looked at card controls inside the mobile app.
That context can influence routing. It can determine whether someone should self-serve or reach a specialist. It can shape the digital experience presented after authentication. And if a human does ultimately receive the interaction, it can allow them to begin with an understanding of the issue rather than asking the customer to explain it again.
More importantly, when those signals are connected to the broader relationship, individual service events begin to tell a story. A customer searched for travel notifications, reviewed card controls, called about an unusual transaction and later updated travel settings. Separately, those are events. Together, they are intelligence.
That intelligence can help product teams understand where customers struggle, digital teams identify broken journeys and operations teams see where supposedly automated processes continue to generate calls.
The call stops being an isolated service event and becomes another signal in the relationship.
The contact-center platform still has a job
Whenever new technology arrives, there is a temptation to redraw the architecture with the new technology sitting in the middle of everything.
I do not think that is necessary here.
Banks have spent years building reliable contact-center infrastructure that solves complicated problems involving telephony, routing, workforce management, compliance, recording and escalation.
AI does not need to replace those capabilities to create value. The better near-term architecture may be additive.
Put an intelligent layer at the edge of the interaction that understands what the customer wants, gathers the appropriate context and determines the best next action. Then let the systems underneath it perform the functions they were built to perform.
That also avoids one of the mistakes I see frequently in conversations about agentic AI: trying to make the agent responsible for everything.
An agent does not need to become the IVR, the contact-center platform, the system of record and the digital banking application. It needs to understand the journey well enough to orchestrate them.
Sometimes the right answer is still a person
None of this means AI eliminates the contact center.
A balance inquiry may not require a person. A routine card-control request probably should not require fifteen minutes on hold. But fraud, hardship, complex disputes and emotionally charged situations can absolutely benefit from human judgment and empathy.
The goal should be to remove unnecessary human effort without removing human availability. The most intelligent system is not necessarily the one that automates everything. Sometimes it is the one that knows when not to.
A good agentic experience therefore needs to understand not only what it can resolve, but when it should stop trying and bring someone else into the conversation. That is another reason I prefer the idea of orchestration over replacement.
Maybe call deflection is too small a goal
I am getting increasingly less interested in the question, “How many calls can AI eliminate?” There is certainly economic value in reducing avoidable interactions, but banks have a much larger opportunity elsewhere.
Take the moment when a customer actively asks the bank for help. Understand why they are there. Resolve what can be resolved. Route intelligently when expertise is required. Preserve the context. Move appropriate journeys into authenticated digital banking. Then learn from that interaction so the next experience is better.
I would hesitate to even call that call deflection. What we’re actually talking about is journey orchestration.
And perhaps most importantly, it removes the burden from customers of understanding how the institution is organized internally. They should not need to know whether their problem belongs to cards, payments, fraud, servicing, digital or the contact center.
Stop forcing customers to learn your processes.
They already have a problem. Understanding the bank’s operating model should not become a second one. The customer should be able to tell the bank what they need, and the intelligence layer can determine where the work belongs.
I do not expect large financial institutions to wake up tomorrow and remove the IVR infrastructure they have spent years building. Nor should they.
But we are approaching a point where asking customers to translate their needs into a rigid routing tree increasingly feels unnecessary.
The IVR still has an important job. It just may not need to introduce itself anymore.
And somewhere, an unnamed pizza company has inadvertently provided me with a very good banking lesson: having every channel is not the same as having one continuous experience.
Had an intelligent layer understood my intent, known the status of my order and carried that context from the app to the store to customer service, this entire article might never have existed.
And for the record, I’m still waiting for someone to tell me what happened to that first pizza.
