For most banks, AI began as a productivity tool: a way to answer questions faster, draft documents, or flag anomalies in data. But the shift now underway is of a different order entirely. Agentic AI systems don’t just assist bankers; they act autonomously, execute multistep processes, and increasingly inherit the same access rights as the humans they work alongside. The question is no longer whether AI will transform banking, but which institutions will be fast enough—and organized enough—to lead that transformation rather than scramble to keep up.
In this video Explainer, McKinsey Senior Partners Andrea Del Miglio, Klemens Hjartar, and Saptarshi Ganguly draw on their work with banking leaders across Asia, Europe, and the United States to explain where AI is already delivering measurable results, the misconceptions still holding organizations back, how the role of every member of the C-suite must evolve, and what will separate the banks that are thriving in 2030 from those that aren’t.
This interview has been edited for length and clarity.
How is AI being used in banking today?
Andrea Del Miglio: It’s being used in many different ways, and it’s growing and becoming more exciting by the day. Especially last year, there was an inflection point with use cases that have become quite impactful.
We’re now seeing AI-powered call centers, voice bots, and chatbots handling the majority of customer interactions. We’re seeing credit and know-your-customer [KYC] processes—customer onboarding—that are largely automated through agentic AI. We’re seeing hyperpersonalized marketing campaigns. It is popping up everywhere.
Saptarshi Ganguly: AI adoption in banking is still in its early stages. Most institutions are running numerous pilots across retail, commercial, wealth, and investment banking—spanning productivity gains, revenue growth, and customer experience.
However, very few of these pilots have actually scaled. That’s due to fragmented deployments, point solutions being slapped onto processes, and insufficient attention to things like operating model, talent, risk, and governance. But now the shift is toward picking big value pools and transforming those holistically, front to back.
Klemens Hjartar: AI is everywhere. It’s a top three topic on any strategy in any bank worldwide—in Europe, the US, and Asia. The expectations are high, and they’re beginning to materialize. It is central to plans around innovation, productivity, expansion, talent, culture, and risk.
And we’re no longer just seeing 2,000 pilots flowering. We are beginning to see genuine consolidation into transformation efforts that are delivering real results. Real promises are being made. This is a major change factor in the banking industry.
What makes AI different from previous tech waves?
Klemens Hjartar: The biggest previous tech developments were the internet and mobile, which were primarily channel replacements. They added amazing new ways to reach customers, changed distribution, and opened up new services. Cloud underpinned all of that transformation.
AI comes on top of all of those things. It arrives at a moment when we already have compute everywhere and mobile capabilities. That’s part of why it’s moving so fast. And unlike previous waves, we don’t yet know exactly how this ends. The compute stack of generative AI is still not fully settled.
Andrea Del Miglio: Speed is the first thing that is really making a big difference. For previous technologies, it took many years before they were ready for prime time. With AI, I saw the first proofs of concept in European banking in 2023. By 2024, it was already full speed.
Looking for direct answers to other complex questions?
The second thing is impact. At first, the biggest impact was in cost. Then it moved quickly into risk, which is one of the hardest areas to open up to innovation, for obvious reasons. Then marketing and sales. And now we might be at the edge of a totally new paradigm: not an app, but a conversation. Conversational banking might be the next—or is likely to be—horizon for how we interact with our financial institutions.
Saptarshi Ganguly: Previous technology waves in banking solved discrete problems. Payments. ATMs. Digital banking. AI represents an absolute paradigm shift.
It is an opportunity to fundamentally rewire how banks operate—to refine distribution, customer engagement, risk management, product delivery, and even the need for physical infrastructure. This is not a tool change. It is a transformation of the entire front-to-back banking model.
Where will AI have the biggest impact in banking?
Klemens Hjartar: In corporate banking, the production of customer data and sales collateral, the creation of complex products—these are areas where AI tools are exceptionally capable. Giving structure to complex data and making something meaningful from it will enable a whole new generation of products.
In retail and private banking, we’re going to see the rise of the agentic adviser—someone who truly understands your needs, takes over many of the interactions, and fulfills them. And in the back office, the long tail of difficult operational processes that have historically been hard to automate—things like KYC and AML [anti–money laundering]—will now come into play. So this is really opening up a completely new era of support and automation.
Andrea Del Miglio: For tech companies using AI, productivity should at least double. That means we’re either going to do double the work or do the same work with fewer people and less money. Banks are only starting to experiment with AI.
There are exciting developments on the horizon. I see a future in which, for example, we have fully agentic development factories—systems that constantly refactor front-end code, making it better and faster overnight. What used to be a two-week sprint now has a 24-hour cycle.
Saptarshi Ganguly: AI will materially impact every single area of a bank—across cost, productivity, revenue growth, customer experience, capital, and the balance sheet. Right now, the most traction is in productivity and capacity creation.
What will happen next is a shift toward growth and hyperpersonalization. An AI-equipped relationship manager will be far better prepared to have meaningful, personalized conversations with clients. Over time, human involvement will decrease significantly, with humans in the loop at the outset to oversee the agentic systems, but with increasing autonomy over time.
What are banking leaders misunderstanding about AI?
Klemens Hjartar: I wouldn’t say there’s a misunderstanding, exactly, but the technology is changing fast. What you learned six months ago has, in some ways, already changed. The orchestration frameworks, the tooling around AI: it’s actually becoming simpler in 2026, because the models are getting more powerful.
That changes the return profile—it’s easier, not harder, than it used to be. We all need to revisit our assumptions about how long things take. With these technologies, complex tasks can be done not ten times faster, not 50 times faster, but 100 times faster. That’s genuinely difficult to absorb when you’ve spent 30 years in a relatively stable environment.
Andrea Del Miglio: I have two stories to tell. First, giving everyone Copilot is not going to yield value on its own. I spoke with two CEOs in the past two weeks who were frustrated—they’d spent significant money on tools that had low usage and zero impact. The lesson is that AI alone won’t change much without proper change management. And adoption has to come from the top.
Second, the terminology is still unsettled. When I talk about agentic AI, I mean complex multiagent systems that fully automate a process. But many of my clients still use that phrase to mean AI-assisted tools for humans.
I’ll add one more: You have to move from individual use cases to domain-level transformations. Use cases are great for proof of concept, but to capture full value, you need to rethink how an entire domain works.
Saptarshi Ganguly: First and foremost, an AI transformation is not a tech transformation. This is much bigger than just technology. In fact, the technology component—the agents, agentic layer—accounts for no more than 20 to 25 percent of the value.
The bulk of the value requires significant shifts across the operating model, data, talent, risk management, and governance. Above all, it requires shifts in the culture and mindset of the organization. Because, ultimately, it’s people at the center of all this change. That’s what banking leaders are still missing, as they’re distracted by the bright and shiny object.
What can banks change internally for AI to work at scale?
Saptarshi Ganguly: Leadership alignment is absolutely critical. This is a CEO-led transformation. The CEO sets the ambition. But every member of the management team has a role: The CFO makes room for investment, the COO takes a front-to-back view of processes to agentify, the CTO [chief technology officer] builds the platforms, the CHRO [chief human resources officer] manages the talent equation, and the CRO [chief risk officer] embeds risk management from the start, not as an afterthought. The whole thing stalls if even one line in that chain is weak.
Klemens Hjartar: Agentic access to data is the big liberator. We went through a phase where everyone had a chatbot sitting on the side of their applications—essentially a Q&A machine. What’s happening now is different: an agentic engine sitting on your desktop, with access to the same systems you have rights to use. It inherits your access rights. That opens up a completely different world.
But the change itself is 85 percent people and culture, 15 percent technology. That’s what I tell every CEO I speak with. We’re moving from an imperative world to a declarative world. The new skill is expressing outcomes, not processes: declaring what you want, setting the boundaries, and measuring results.
It’s kind of like being married, right? You need to be very clear on your intentions and your boundaries, and remember to talk. Those are the new skills of the agentic era.
Andrea Del Miglio: The banks that are leading in AI share a common trait: a clear strategy. Five things that will make a difference in the next three years. That focus prevents you from dispersing your talent, which is extremely scarce, across too many initiatives.
You also need someone senior who can unlock quickly. You will hit policies, cultural habits, and ways of working that simply don’t fit anymore. And you need to invest in a central platform early. A central, safe, compliant platform is what gives the rest of the bank permission to experiment.
What are the new risks, and how should they be managed?
Saptarshi Ganguly: New risks include cross-agent vulnerabilities, data leakage, and synthetic identity fraud. Traditional risk frameworks and human-led approaches are wholly insufficient to manage these. Risk teams need to evolve and get armed with the same agentic tools that rogue actors use to find vulnerabilities—and stay two or three steps ahead.
Regulators face a learning journey, too. The challenge is to mandate effective risk management without being so conservative that you stifle innovation. First, second, and third lines of defense inside banks—and the regulators outside them—need to be equipped with the skills and tools the business side already uses.
Klemens Hjartar: In many ways, I’m more concerned about the risks of the past than the risks of the future, at least imminently. The biggest immediate threat is in how we implemented legacy IT systems and the cyber vulnerabilities that remain in that infrastructure.
The new AI models arriving in 2026 have an enormously powerful capability to find vulnerabilities in those systems. Banks need to deploy these tools proactively—into their own tech estates—and be genuinely forward-leaning with their cyber teams. The processes of the past won’t protect us in the future. We need first-principles thinking with the new technology available.
Andrea Del Miglio: There are real questions about identity management for agents—how you identify an agent, how you control it. There is no market standard solution yet. But unlike humans, agents leave a full audit trail of every action they take. For the first time, you can trace back to the single prompt that generated a piece of code. That level of accountability has never existed before.
Of course, agents are also being used to launch cyberattacks—enabling much more sophisticated, faster-moving threats. What was once only possible for elite hackers is now more broadly accessible. The arms race between attackers and defenders with agentic tools is very much in play.
What does the future of banking work look like—and who are the people who do it?
Klemens Hjartar: People should be less worried about whether there will be jobs. You are more likely to be replaced by someone who knows AI than by AI itself. We should approach this with a positive, learning mindset—and be realistic about how much this will change our society.
We’ll see a plethora of new job categories that didn’t exist a few years ago. And a great deal of the tedium of today’s work will be automated away. I genuinely believe we are heading toward better and more exciting jobs. I’m an optimist.
Saptarshi Ganguly: Three things will happen to talent. Some roles will go away entirely—doers, checkers, hands-on coders. Some roles will evolve significantly: engineering talent, for example, will step into product management and systems thinking, using agentic tools to build software while applying their own deep expertise to embed it in complex legacy stacks.
And entirely new roles will emerge—AI trust and safety specialists, for instance. Banks need to be thinking about this talent equation now: how to upskill people whose roles are changing, how to plan for roles that are disappearing, and how to attract the scarce talent needed for roles that didn’t previously exist.
Andrea Del Miglio: Talent is truly of the essence. There’s a huge difference between having a very strong data science and engineering team and having a B-team. Banks will need to rethink constraints around location and compensation that no longer make sense when you’re competing for an extremely small global talent pool.
Who will win—and lose—by 2030?
Klemens Hjartar: Speed is the difference. When new technologies can move customers between institutions quickly, being six months late will be very expensive. We can statistically prove, from the internet and mobile revolutions, that being a fast follower was a bad strategy. It seemed reasonable on paper, but it wasn’t.
That will be even more true in this revolution. If you wait six months to implement new cyber practices after models have arrived, that is a very expensive adventure. If you’re not in agentic commerce when it takes off, you won’t learn as fast as your competitors. The fastest learner wins. That is the dividing line between winners and the rest in this new era.
Andrea Del Miglio: I believe that the majority of these capabilities will ultimately need to be commoditized. Every bank will need a base level of AI sophistication just to compete. The first movers and fast movers will have an advantage, but over the next three to four years, more institutions will follow.
If you are truly a laggard—if you fail to get on the train at all—you may simply be out of business. But I think it will mostly be a wave, where early movers gain, others catch up, and we arrive at a new normal. What that normal looks like, honestly, is very hard for me to picture right now. Things are moving that fast.
Saptarshi Ganguly: We will see a significant and growing strategic distance between two groups of banks: those that move early and fast, and those that follow. Our preliminary estimates show up to a six-percentage-point gap in return on equity between first movers and fast followers.
There is also the threat of disruption from fintech attackers targeting the most valuable parts of the banking value chain—potentially 8 to 12 percent of the total value pool disrupted over the next three to five years. For bank CEOs, this is the moment to set a bold ambition, identify the biggest value pools—on both the cost and revenue side—and put in the enablers to go after them. Otherwise, prepare to be disrupted yourself.
