Agentic artificial intelligence (AI) could give customers a new way to interact with their banks. AI agents may help customers research products, compare options and complete authorized financial tasks. Banks must design experiences for both customers and their agents while maintaining clear permissions, customer control and human support.
Dubberke predicts that AI could give customers another way to discover banking products and complete selected tasks. This channel would complement branches, contact centers and digital services rather than replace them.
For example, a customer's AI agent could research offerings or interact directly with a bank, creating two connected experiences: one for the customer and another for the authorized agent.
Banking apps could become trusted control centers where customers review agent activity, approve sensitive actions and understand what their agents have done.
As AI handles routine research and preparation, branches could focus more on complex advice. Customers may arrive better informed, increasing the importance of employee judgment, reassurance and decision support.
Eryilmaz points to research suggesting that younger banking customers still value in-person financial advice, even when they can research products online. He sees an opportunity to combine that human expertise with AI-supported recommendations.
This approach reflects a model of “co-intelligence,” where AI helps employees tailor recommendations while preserving human expertise and empathy.
Banking employees will need stronger judgment, data literacy and the ability to challenge AI outputs. New capabilities in AI operations, model risk and agent supervision will also require practical learning across business, technical and supporting functions teams.
Agent-to-agent banking enables a customer's AI agent to interact with a bank's agent to request information, compare products or initiate approved tasks.
Over time, agents may negotiate rates, fees or service conditions within defined limits. Banks must establish who authorized each agent, what actions it can perform and when customer confirmation is required.
According to Dubberke, repeated negotiations could push prices toward minimum levels, potentially overlooking service quality and the broader customer relationship. Banks will need to balance competitive pricing with the value of the overall customer experience.
Clear safeguards include limited permissions, confirmation for sensitive actions, audit trails and dispute or reversal processes. Banks must also distinguish between customer mistakes, agent errors and process failures to establish accountability and appropriate remedies.
Customers need to understand when they are interacting with AI, what information it uses and what actions it can take. They should also know when human support is available, be able to choose how much authority to delegate to AI, and review important decisions before they become final. Trust also requires accurate product information, secure identity verification, reliable connections and consistent answers across banking channels.
Eryilmaz points out that even simple customer interactions require disciplined back-end operations. Banks must manage this complexity while keeping experiences understandable and giving customers meaningful control.
Banks should start with a controlled business process, a defined customer outcome and limited agent permissions. Risk, compliance and frontline employees should participate from the beginning.
Success measures should extend beyond productivity to include accuracy, customer effort, employee adoption, fairness, escalation rates and outcome quality.
Banks must also prepare for external customer agents by structuring product information, strengthening application programming interfaces (APIs), clarifying pricing and testing whether independent AI systems can interpret their offerings.
Customer communications, pricing and regulatory reporting require particular care, supported by clear governance and appropriate oversight.
Listen to Andy Schmidt, Mehmet Eryilmaz and Katharina Dubberke discuss AI as a banking channel, agent-to-agent interactions and trustworthy customer experiences in part two of CGI's From Transactions to Trust podcast.
- Chapter 1: How could AI become an additional banking channel?
-
Andy Schmidt:
Welcome back to From Transactions of Trust, the financial services podcast by CGI. My name is Andy Schmidt. I'm the Global Industry Lead for banking at CGI, and I'm your host for today. I'm joined by my colleagues Mehmet and Katharina for episode two of our discussion on the agentic bank. Today, in this episode, we're going to talk about the customer experience and how AI becomes a channel and how clients and agents will interact in that channel.
You may have seen that Katharina wrote a blog on this very topic recently, and so I'm thrilled to have her with us. Katharina, I'm going to ask the first question to you, and we're going to be focusing on three topics today: the AI channel, agent-to-agent banking, and experience and execution. Given the work that you've been doing recently in the blog that you just wrote, will AI become a new banking channel alongside branches, contact centers and other digital journeys?
Katharina Dubberke:
Hi, Andy. Yes, I think that there will be an additional channel regarding AI, but I also would not expect that it will replace every existing channel that we already know. AI will become an additional way for our customers to discover products, ask questions, complete selected financial tasks, stuff like this. But for simple needs. The customer, for example, will never again open the banking app if they just want to answer a single question. Their agent can do that and may interact directly with a bank on their behalf.
And that means for the banks that they will need to think about two experiences: the experience of the person and the experience of the agent acting for that person. And the banking app, don't get me wrong, the banking app will still remain important. But it will become less like a destination like we are used to it today, and more like a trusted control center where the customers can review, approve, and understand what their agent has done.
Andy Schmidt:
And what customer needs are better served by AI, and which still need people? Which still need human intervention?
Mehmet Eryilmaz:
CGI, we were recently at Sibos, and we had one of the leading UK banks who gave an example across many flows. I think flows that could be automated a bit faster, that they absolutely exist. But there are also these flows that are heavy human touch. And the example that the gentleman gave was the bereavement flow. And he said, I work with my team and I look at this flow, and when we're going to automate a lot of the processes, that's going to be one that will be extremely difficult. Do we even want to move that over to AI?
And I think that sets a stark example of how some processes are going to be fully automated versus those that will still need the human touch. I think AI is particularly powerful in helping customers navigate complexity. It can explain products in simple terms. It can provide personalized guidance any time of the day. But again, for many routine information-heavy interactions, AI could be faster. But I would not go so far as to say that every process will be AI-driven.
- Chapter 2: How could AI change bank branches and employee roles?
-
Andy Schmidt:
No, I agree. There are going to be processes where people are just coming in for information and need help to walk them through a process. Human-to-human interaction is what's going to be key.
Branch experience is the topic of our next question. How could the branch evolve when customers use AI to ask better questions?
Katharina Dubberke:
I think branches become less focused on routine transactions, and they will be more focused on complex conversations because an AI agent can handle basic comparison preparations. Customers may arrive with better questions and a clearer understanding of that option. And the role of the branch advisor could then shift towards interpretation, reassurance, or decision support. And we should not forget about empathy, because at least we are still talking to a human.
Mehmet Eryilmaz:
I agree with that. And from a UK perspective, there was a recent study looking at the new generation of banking customers. And the results were: they want to go to the branch. They wanted to get advice. And when we dig deeper, the response from the new generation was: we can research a lot of things online, but to be financially well-versed, we would love to get a live person's counsel at the branch.
So, for opening a new account, understanding how they could do savings, they want to visit the branch. I think to add on to what Katharina said, absolutely, I think the branch will become a hub for more targeted advisory. And I would put the future branch model as one of co-intelligence, where you have—and I'm borrowing the term from a professor from MIT, Professor Ethan Mollock, who had this book on co-intelligence—saying you have to have the tools that assist the branch frontline in coming up with recommendations that are tailored, hyper-personalized and that the customer cannot find online anymore.
- Chapter 3: How can banks establish authorization and accountability in agent-to-agent banking?
-
Andy Schmidt:
And I like that model, that co-intelligence concept, especially if you’re using AI to start the conversation and want to finish the conversation with the human whose empathy is going to be so key in helping make that important financial decision. Which brings us to the next topic.
Now we're talking about actually using AI first and closing the loop with a human-to-human conversation. But you know, Katharina, you just wrote about this: the rise of agents in banking and agents working for the client. Agent-to-agent banking is also here. And it's a world where banks and customers are deploying agents to negotiate with and work with one another. So, in this future, how do you think bank agents and customer agents will interact?
Katharina Dubberke:
Initially, the interaction will probably be quite structured. The customer agent will request information, compare products, or initiate a process. The bank agent will provide eligible options, explain conditions and perform approved actions that he or that the agent is allowed to do.
Over time, I think agents may negotiate within defined parameters such as rates, fees, repayment preferences or service conditions. So, the banks will need to know who is authorized to act, what the agent is allowed to request, and when the customer must confirm the action. The interaction needs identity, consent, auditability and clear accountability. In short, agent-to-agent banking requires rules of engagement. Even AI agents need to know when they are negotiating and when they are simply being very confident.
Andy Schmidt:
Understood. And that raises an interesting question. What happens when agents negotiate fees, rates, or conditions?
Mehmet Eryilmaz:
We were having a discussion with one of the UK banks recently on the agent/bank future and I think our fear is, from a bank's perspective, if it is just agents talking with one another, is there a scenario where we drive the price down to the absolute minimum? I think financial services institutions face a challenge when agents interact with one another. Where do we draw the lines? And is that still going to be economically feasible?
And also, in the future, where we might be able to develop new products very, very quickly, and that's repeatable, that could be done by every bank overnight, pricing becomes an issue. So, I think it will go back to what sets us apart as a financial institution. Do we have that trust with the customer, the longevity, and the AI agent interaction being a tool? But my overall interaction with my customer and the pricing models have to be bespoke and tied to the customer experience that I bring in its entirety.
Andy Schmidt:
And then with those negotiations, how do banks manage errors when an agent acts on the customer's behalf?
Katharina Dubberke:
Banks need to design for mistakes before they give agents meaningful authority. Before going live, you need to prove that you, for example, in your use case, have mentioned all the error scenarios that could occur. The same applies here for the AI agent. And that also means limiting permissions, requiring confirmations for sensitive actions, so maintaining a full audit trail and offering a clear reversal or dispute process.
And customers must be able to see what their agent did, why it did it, and what happens next. On the banking side, the banks also need a way to distinguish between a customer error, an agent error, and a bank process failure, for example. So those questions matter for the remediation and accountability, because afterwards, if you need to check, okay where this error really comes from, you need to check all the possible options that you have. When an AI agent is also in the loop, you need to also take care of this one.
Mehmet Eryilmaz:
And I think Andy, life is really messy, and there are lots of banking products, there's a lot of money movements. And what will happen is what we do day to day in banking, we manage that, banks manage that and appease the customer. Customer experience, customer loyalty come first. Now, how do we balance that with something that is going to be fully industrialized and automated? And I think especially as we scale the volume, we will need to make sure that customer experience is kept as a principle. And then maybe we walk into a future in a couple of years where things are a bit more automated, but I don't think we're there yet.
- Chapter 4: How can transparency and customer control build trust in banking AI?
-
Andy Schmidt:
I agree. Crawl, walk, run, and we'll get there soon enough. Which takes us to our next topic: experience and execution, making the future practical for customers, making it practical for employees, and keeping in mind at all times that this is a regulated industry and these are regulated institutions. So, with that in mind, how should banks design a customer experience that remains trustworthy, that keeps that trust that the bank has taken so long to build up?
Katharina Dubberke:
A trustworthy AI experience should contain three elements. It should be understandable, it should be controllable, and it should be consistent. The customers need to know when they are interacting with AI, what information it is using, and what it can do, and when a person is available. They should be able to review important decisions before they become final. And trust also depends on the less visible parts of the experience. So, accurate product information, reliable IPs, secure identity, and consistent answers across the channels. And if the customers may see a simple conversation behind it, the bank needs a disciplined operating model.
Good AI feels effortless to the customer at the end because the complexity has been handled already before within the bank in the responsibility backstage. So that's the key, which is, at the moment, the challenge of the banks.
Mehmet Eryilmaz:
I think there are going to be a cadre of successful banks through this. And I feel that is going to be those who automate with more transparency. They will provide clearer explanations around how AI being used, maintain that human escalation path. And I think that authority dial of how much I can delegate to AI systems, that needs to be a choice.
Andy Schmidt:
And with that in mind, what skills and roles change as AI handles more work? I mean, Katharina, you talked earlier about the branch becoming more a place for more specialized advice. How else will roles and skills change in the bank as AI continues to gain ground?
Katharina Dubberke:
I think the biggest shift, like we already said, regarding the branches, will be from processing information to supervising decisions and improving processes. So that means, in the end, the employee who is in the branch will need strong judgment, data literacy, and the ability to challenge AI output. They also need to be some kind of experts to make sure that the AI output is correct and that they can use it in their face-to-face meetings within the branch.
Domain experts will become even more important because someone must define what a good decision looks like for the bank and also recognize when the answer is correct or wrong. We will also see that roles will be more focused on AI operations, process design, model risk, agent supervision and customer advocacy. This is an opportunity to make the banking work more meaningful. There will definitely be a shift from repetitive tasks to more specialized tasks. If the AI removes the repetitive navigation and administration, people can spend more time solving cases that actually require their expertise.
Mehmet Eryilmaz:
With the role question, there’s also a challenge for business leaders, as well. I look at it in terms of a competitive advantage coming from simply giving AI tools to everyone. I think probably directionally, the advantages will come from building a more AI-capable workforce that understands where AI adds the greatest value, where human judgement matters, and where we have perhaps those roles that are able to use AI hands-on, but really has deep content and industry subject matter expertise as well.
We need to be able to couple that. I think banks, financial service institutions need to really think about how they're going to upskill the workforce in such a way that there are different learning pathways for the neurals for both business, technical, and back-office users where they use AI tools but also know how the business details work.
- Chapter 5: How can banks test agentic AI and measure customer outcomes?
-
Andy Schmidt:
I agree with you that specialized training, helping folks on the bank side leverage AI, be trained to use AI to go deeper, to provide them more specialized skills is going to be so vital, whether it's providing, as you mentioned, career paths for the employee, but also making it easier to give the customer experience that seamless, that value-added customer experience that they're looking for.
But you know, the other piece of it that's so important is we're looking at some of these models and how we're going to serve the client base is what should leaders do now to test the model or the models safely.
Katharina Dubberke:
I would suggest starting with a controlled business process. You have a clearly defined customer outcome and a limited set of permissions. Almost everybody is measuring productivity. It is also key to measure other topics like accuracy, customer effort, employee adoption, fairness, escalation rates, and quality of the final outcome. So, involve risk, compliance, and the front-line employees from the beginning.
At the same time, prepare the bank for the external agent. So, make the product information structured, strengthen your APIs, clarify pricing and test whether an independent AI system can understand the bank's offering. If not, customers will choose another bank because the API works as expected there, and they take into account what other customers are doing. And the first test should answer two questions: can the bank create value, and can it remain accountable? If it cannot answer the second question, it's not ready for the first.
Mehmet Eryilmaz:
I agree with that. I go back to organizations needing to decide which parts of the organization AI will lead and which parts we need to keep heavily human-in-the-loop. And it won't be enterprise-wide at the same speed. Some parts of the organization will move faster. What we’re seeing is that, on the engineering side, back-office side, and some of the product functions, things are moving quickly.
Customer-facing is more tempered. Customer communication, pricing, any type of regulatory reporting, especially in the UK, is still very, very hands-on. So, we need to be a bit more diligent on that. So definitely the guardrails, clear governance, and getting safety around that becomes a priority.
Andy Schmidt:
Well said. With that, we'll move to close the second episode on the agentic bank. Thank you, Mehmet and Katharina, for your time. Really appreciated the conversation and your thoughts on the topic.
And thank you again for joining us for the latest episode of From Transactions to Trust, a financial services podcast by CGI. And look forward to seeing you on our next episode.
Thank you.