Artificial intelligence isn’t just another efficiency play for financial institutions; it represents a fundamental shift in how banks manage costs, attract and retain customers, and stay ahead of the competition. In this episode, McKinsey Senior Partner Ido Segev and Editorial Director Roberta Fusaro discuss the speed with which AI is advancing, what’s holding banks back from scaling it, and the potential advantages for early movers. This conversation comes ahead of McKinsey’s upcoming 2026 Global Banking Annual Review—preview it here—which takes a deeper look at the unprecedented pace at which AI is remaking the industry.
The McKinsey Podcast is cohosted by Lucia Rahilly and Roberta Fusaro.
The following transcript has been edited for clarity and length.
AI questions abound
Roberta Fusaro: Every year, McKinsey releases its Global Banking Annual Review. To no one’s surprise, one of the biggest themes this year is AI and its impact on banking. Ido, what are you telling clients about the speed at which AI is emerging?
Ido Segev: It’s important. While clients are going after it tactically and strategically, they are asking a few questions: How quickly is it coming? What is the potential impact? How much value would it deliver to my own P&L? How might it impact the industry overall, and how might revenue and profit distributions be affected?
The third strategic question that they are asking is, what does it mean for us? How quickly should we actually pursue this? A lot of clients are experimenting, but should they move more quickly? Should they scale more quickly? Or should they sit on the sidelines and wait to see how it matures?
Roberta Fusaro: What are the considerations for those banks that are starting to adopt AI?
Ido Segev: The first is how quickly the technology is maturing. It’s maturing and evolving rapidly. If you compare, for example, the different LLM-enabled tools—ChatGPT, Gemini, and Anthropic—you will see how they keep surpassing each other. Every three to six months, a new version comes out, and it’s smarter and more intelligent than the one before it.
The second driver of adoption and pace is how quickly customers will trust AI enough to delegate the tasks they normally do themselves to agents. We already see many customers going to agents for advice. But at what point will customers allow agents to move money from one account to another? At what point would customers ask AI to apply for a credit card on their behalf because they found a great opportunity?
The third driver of adoption is how quickly a bank can deploy AI at scale across the relevant parts of the organization. That means deploying agents as virtual employees and moving toward a ratio of 20 virtual employees to one human employee.
Moreover, particularly in this industry, the regulator’s stance will evolve toward AI. What we see there is that the regulator is obviously more conservative when it comes to how banks make credit decisions, such as whom to lend to and at what price. But when it comes to how the bank runs its operations, say, a call center, there’s less regulatory risk associated with that.
Roberta Fusaro: You mentioned one of the adoption factors being customers and trust. What can banks do to change their relationship with customers and build trust that enables agents to do their banking for them?
Ido Segev: Consumers will be a bit more hesitant to share more information or let agents take actions on their behalf if there isn’t a clear value proposition. A clear value proposition could be “I can do things much faster,” or “I can do things much more cheaply,” or “I can get better rates.”
If you take examples from the digital age, people were talking about privacy and sharing location information. Well, guess what? Everybody’s using navigation apps now. And with those navigation apps, we know where we are at every second. But people don’t mind using them because they get value from them.
Roberta Fusaro: Which freaks me out entirely.
Ido Segev: I know it freaks you out. But there’s value in the exchange, right? To accelerate adoption, the question is, What can the bank do to provide more value than the customers would get otherwise? Trust is going to be driven first by attackers, like challengers or fintechs, who will provide those solutions and add value at the same time with banks, maybe even before.
To accelerate adoption, the question is, What can the bank do to provide more value than the customers would get otherwise?
You could imagine how a nonbank fintech could come in and say, “Roberta, I can optimize your deposits.” You probably have a lot of deposits that sit at a low interest rate. I have an agent that would look at that and not only would tell you, “Roberta, you should move your money from Bank X to Bank Y because Bank Y has better rates” but could also move the money for you. “We could do that daily, weekly, or monthly on your behalf.” You would say, “I can see the value of that and can agree to do that if I have an optimized agent. Let me give it a try.”
So it’s a real use case that could have a meaningful impact on overall banking revenue pools. This could happen tomorrow. And it’s likely that a few early adopters will be excited by this use case and adopt it. Then, banks will say, “I need to adopt the same tools because I don’t want to lose that money. I don’t want to lose our customer.”
Solving the AI scaling problem
Roberta Fusaro: You also mentioned one factor for greater adoption: banks using AI at scale. There’s a lot of research about how many AI pilots get stuck in purgatory, and people can’t quite figure out how to move them beyond a particular group or use case. What do you think needs to happen for banks to solve this AI-at-scale problem?
Ido Segev: The first is focus. There are so many places where you can apply AI and agentic. You could say, “I can see opportunities here and there; the opportunity could be in my front office or back office, in commercial lending, or in marketing.” You need to focus. What two or three areas can immediately provide disproportionate value? You must also consider where agentic and AI can help you reduce costs while accelerating revenue growth.
Why is that important? Because creating a perception that AI is all about cost drives fear in the organization and creates stagnation, and if you can say, “This is something that could allow us to be more efficient and productive, but also accelerate our growth,” that provides a platform for further acceleration or scale-up.
The second is that it’s not a technology thing. This is a one-team, one-goal matter. It’s business and technology, as well as HR, which is a big contributor here in terms of virtual and real employees, how they work together, and how you manage performance.
The next big thing you need to do is upskill while you drive the transformation. There are early adopters and they vibe code at home. And there are people who are like, "I'm not sure.” So, everybody needs to be upskilled.
Fourth, and this is an important one, is running a proof of value and releasing capacity. For example, if you redesign a process in your call center and your representatives become more productive, that does not immediately translate into capturing value. Capacity is not value. Right now, that capacity benefits dogs the most, because their owners have more time to take them out for walks.
I hear that a lot from my clients, and it makes sense because you need a plan for monetizing that capacity. You monetize that capacity in two ways: You either put it to work to drive high productivity, or you take it out. From the outset, making sure you set up a redesign of a domain or a workflow and say, “I’m going to generate this much capacity and here’s what I’m going to do with it,” is a small but very meaningful step.
Roberta Fusaro: That makes sense.
Ido Segev: I think safety and governance, particularly in a regulated industry, and understanding how you can stay within guardrails that you are comfortable with as an organization, is something that accelerates pace. They say Formula 1 cars can drive very, very quickly because they have amazing brakes. That should be your mindset: Have a braking system that allows you to accelerate.
Finally, we’re suggesting you look at data, metadata, and contextual data as a multiplier, as something that only you can do. These would make your bank’s agents smarter and more skillful than other agents that you can “buy off the street.”
Roberta Fusaro: This point of differentiation seems to be at the core of a lot of this, even with your interactions with customers. You want to have a strong value proposition and give them something they can’t get now, or can’t get from someone else.
Want to subscribe to The McKinsey Podcast?
Ido Segev: Exactly. Therefore, how do you leverage the data and the history that you have with your customer? How do you put it to work to make every interaction much more valuable?
Roberta Fusaro: Is adopting AI harder for a particular cohort of banks than for others? Or do all these challenges apply equally across the board?
Ido Segev: It’s a question of who stands to win. Is it just the large banks with the big pockets who are going to win? Or could there be, say, super-regional midcaps or community banks that could be winners in their categories as well? This question is particularly relevant to the US market, where we have so many banks.
We looked at this question in our Global Banking Annual Report. Our view is that in each of those categories, there could be a set of winners—not many, but some. For example, in the global bank and the G7s, there could be one or two, maybe three big winners. The super-regional banks may have five or six winners. In regionals, maybe 10 to 15 winners. And those winners are the ones that will adopt this technology faster, overcome the scale issues we talked about, start seeing impact on their bottom line, and use that impact and extra profitability to reinvest in growth, ultimately capturing share from the laggards.
Who will come out on top?
Roberta Fusaro: When we talk about these winners, can you explain how their approach differs from some other competitors?
Ido Segev: It was around this time last year that agentic became more real. I say that because we’re still in the early innings here. There isn’t one bank or a few banks that we could say, “They figured it out, and they are now running at full speed.” Even some of the banks that we hear about in the news still have a ways to go. If I double down on some of the actions that have been taken and some of the impact we’re seeing, there is one bank we work with that has been on a journey to deploy agentic in its call center.
For example, in the call center, they provide an AI assistant to reps that provides information about the caller and recommends answers to reduce average handle time, which is a critical factor. In some situations, using agents decreased the average handle time of calls from four or five minutes to around one minute. That’s a meaningful reduction. Companies are now saying, “There are certain types of calls that are actually better handled, or can be handled more effectively, by an agentic voice.”
That’s meaningful impact. So that’s one example we see with clients. Another area is in the operational risk, “KYC,” or “know your customer.” That’s an area of the bank where they spend a lot of money and invest heavily in personnel. We see banks saying, “How can we make the process much more productive and efficient so we can run it with maybe 60 to 70 percent of the people, maybe even 50 percent of the people on staff right now?”
We see banks rethinking their workflows and where they apply the AI, and asking themselves, “Where do we still want to keep the human in the loop?” Because I don’t think anybody believes we’re in a place where we can do this entirely without people. What will be the role of the human in the loop?
The third example is in the front office, in sales. Imagine a banker pursuing opportunities with new clients. The AI can integrate all the information about that potential client. What is the latest information you can gather about them? What could be their needs? An agent can synthesize other interactions that we’ve had from this side and create a brief for you. For example, if you knew that the CFO of a company is a very outgoing person, you can create an AI simulator that will allow you to simulate a conversation with that person. Then, when you go to the actual conversation, you have already role-played it, so you feel more confident about your interaction with the human.
Roberta Fusaro: That’s amazing, the role-playing. I hadn’t even thought of that as a potential application. I’m going to build my own agent so that I can practice the podcast against various personality types.
Ido Segev: I had an agent role-playing you to prepare for this podcast.
Roberta Fusaro: Ido, as companies deploy AI to improve the customer experience and boost productivity, what cost factors should they think about? What did the report show?
Ido Segev: In our report, we planned out a few scenarios based on the level of customer adoption and the end level of bank adoption. It’s conceptual, but we used a two-by-two matrix with one axis representing customer adoption. On one end of that Y axis, we had customers using it just for advice. At the other end of that axis, we had agentic AI as the access point to a bank. So no branches, no call center, no mobile app. It’s a bit extreme.
On the X axis, we had bank adoption with one end representing “This is just a cool pilot” and, just to be completely extreme, the other end being zero touch with hardly any humans. We then spoke with 50 or 60 banking CEOs globally, investors, academics, researchers in this space, and fintechs. We posed the question, “What is the most likely scenario?” And the most likely scenario was one where customer adoption goes beyond just the advice. Customers are increasingly getting comfortable with agents acting on their behalf. But it doesn’t fully replace all other channels.
From a bank adoption standpoint, we reached a point where people, through those conversations and our insights, believe there will be 20 to 30 agents per bank employee. I think it’s important because then it’s under that scenario that you understand the potential impact.
Under this scenario, for example, you can play it out and say, “We’re talking about, say, a 25 percent reduction in bank overall cost.” By the way, this reduction in cost is also accounting for an increase in IT spend. A lot of that cost benefit is coming from the middle and back offices. Some of that impact comes from the front office, making our salespeople and marketing departments more productive. Let’s translate that to dollar terms. Our math suggests that a 20 to 25 percent cost improvement translates to about $250 million to $500 million for every $100 billion in assets.
So if you have $100 billion in assets, the impact to the bottom line is $250 million to $500 million. If you’re $200 billion, it’s $500 million to $1 billion to the bottom line. It’s quite substantive.
That’s the impact on costs, and therefore on the incremental profit. We expect increased profitability to be competed out and transferred back to customers—just natural competition. Attackers come in and set lower prices, there’s pricing compression, and then all that margin basically ends up going to the customer. That’s what we saw also in the digital transformation age, where banks have developed digital apps and more consumers, and not just consumers, but also users of the apps. Banks are not massively more profitable now than they were 20 years ago. It’s going to be outcompeted.
The last thing I would say about the dynamics is that urgency matters. If you were able to get ahead of that curve and achieve $250 million to $500 million of extra profitability per $100 billion in assets early, you will be ready when pricing starts to compress and profits start moving. You’ll be able to reinvest in capturing share from the ones who are not ready.
Roberta Fusaro: So you’re always staying one step, or two steps, or 500 million steps ahead.
Ido Segev: Exactly. I think it’s particularly important because, down the road, when those agents are optimizing your deposits or credit card loans, the overall profit for the industry might actually shrink.
By the way, we reviewed the report and estimated that overall profit pools compress by 9 to 10 percent. It’s quite meaningful. That translates to a 1 to 2 percent reduction in ROE for the average bank. If you say, “We are now in a dynamic where over time there’s going to be compression of overall profit pools,” that’s number one. Number two, the gained COE [cost of equity] dynamics here is one that, over time, extra profitability will be outcompeted. And the combination of the two basically implies, “Moving quickly is advisable and could be advantageous.”
Heft is out, and precision is in
Roberta Fusaro: It’s a great piece of advice. I would also urge people to go back and review the full report, as it contains a lot of great data. Speaking of that, what were some of the other big trends or findings beyond AI?
Ido Segev: The tag line for the report this year was “Precision over heft.” We tried to understand the impact of scale, measured by the number of people and the size of the bank by assets, and asked, “Are bigger banks necessarily more efficient, or do they return higher returns to shareholders?”
What we found is that in some places, certainly in the US, that’s not the case. Scale does not correlate with better performance. And this is where we say, “Heft might not be the key thing.” We believe that what matters more is precision.
“Heft might not be the key thing.” We believe that what matters more is precision.
In the context of M&A, for example, it’s not necessarily about how I buy the biggest bank that I can afford, to size up. It’s about finding opportunities to buy banks that are more complementary to me, that have talent that I need, and that have a technology stack that could be advantageous to me.
That’s precision in the context of M&A. In the context of capital deployment, which is critical for banks, how can I become super precise and categorize different loan types? How do I think about the right way to define those loans to optimize my risk-rated assets?
That’s the precision there. In the context of customers, you can identify the segments that you believe you can win with and tailor your approach to how you reach them, whether it’s through digital marketing or through your bankers, almost on a one-to-one level. So, that is some illustration of what we mean by precision and why we think that matters more than heft.
