Agentic AI is rapidly emerging as a game changer for banks facing growing pressure to become more agile, resilient, and customer focused. Going well beyond traditional automation, agentic AI has the potential to fundamentally reshape how work gets done, how decisions are made, and how value is delivered at scale.
In this episode of McKinsey Talks Operations, host Daphne Luchtenberg explores the opportunities agentic AI affords banks, including why some institutions are pulling ahead, why others are stuck in “pilot purgatory,” and what banking leaders can do to capture real impact from their agentic and AI programs. She is joined by David Deninzon, a McKinsey senior partner who specializes in banking operations, and Abhilash Sridharan, a McKinsey partner and service operations lead in Asia.
The following conversation has been edited for length and clarity.
Daphne Luchtenberg: Your company’s future success demands customer-focused, agile, resilient, and efficient operations. Today, we’re diving into a topic that represents a true paradigm shift: agentic AI in banking.
Agentic AI is not just another buzzword; it’s poised to become a critical differentiator for banks everywhere. Leading institutions are leveraging AI as a platform to redefine their workflows and business models, transforming how they operate and deliver value to their customers.
However, the story is not the same for all banks. Slow adopters face the very real danger of falling into what we call pilot purgatory: dabbling in narrow use cases without fully realizing the transformative potential of this technology.
In today’s episode, we discuss the opportunities and challenges that agentic AI presents and how banks can navigate this new landscape to stay ahead of the curve. Let’s launch straight in.
David, let me pose this question to you first. Why and where is agentic AI going to be so powerful?
David Deninzon: That’s a great question. Banks are one of the heaviest users of what we call service operations, or people delivering services to customers. And AI is well positioned to have a significant impact on how the delivery of tasks and services is done for customers.
Unlike traditional AI or machine learning, agentic AI enables banks to run not only deterministic workflows but also analysis and tasks that are less structured, more personalized, and that effectively happen only once. Therefore, AI is very well positioned to help banks.
On top of that, if you look at operations specifically, it’s the area, after technology and engineering, that is prime for AI and agentic AI to be of help. We estimate that, depending on the bank, between 50 and 60 percent of the FTEs [full-time equivalents] are in some way tied to operations. And AI’s potential to transform how work gets done and how work and services are delivered is tremendous.
Based on our 2025 global banking report, we have seen a number of institutions in the banking space using AI. It’s still early to call it a win, but there are very promising results on the impact around speed, cost, quality, and, ultimately, customer experience.
Daphne Luchtenberg: Abhilash, what is making this landscape more complex for banking leaders who are considering AI transformation, and why is it taking longer than expected? What are you hearing from clients?
Abhilash Sridharan: As is the case with any new technology, creating value from AI won’t be a cakewalk, and it’s going to take time. Successful organizations will need to rewire entire domains across operations, frontline distribution, technology, data science, and risk management, with AI at the core and supercharging the impact versus AI-led applications just being a hammer looking for a nail.
In fact, it’s interesting you’re asking this question because there is a paradox here. The impact of AI, gen AI, and agentic-AI-led applications in the sector has been mixed. Nearly 80 percent of financial institutions that we work with in Asia report using some version of AI-led applications. But a similar proportion globally reports no significant impact on their bottom line.
There are a few challenges to consider, especially for financial institutions. The first is that AI initiatives are driven within functional and business silos, with unclear linkage to financial value.
The second challenge is chasing impact from gen AI alone, which has inherent limitations. And it’s important to pair it with a full-stack lens, where you blend gen AI with agentic AI, traditional AI automation, and digital applications to maximize value capture across the different domains.
The third is deploying narrow use cases and point solutions. It is very easy to pick and choose the easiest possible areas for you to go after—for example, building a chatbot for customer care, building knowledge management applications for your employees, or building a fast credit memo writer for a subset of the businesses. And then you plateau right after that. That’s essentially where you stymie the impact from AI versus driving end-to-end transformation of domains, which are business backed and have AI at the center.
The fourth is building LLM [large language model] applications like traditional analytics models, which limit their scope to generating content and making decisions.
The last one is limited reuse of AI enabling capabilities, which results in poor ROI and slower scale. It is important that most financial institutions ‘Amazon-ize’ their ability to drive cross-cutting applications of AI-led capabilities across the bank. The retail bank, the SME [small and medium-size enterprise] bank, the midcorporate institution, and the institutional bank shouldn’t build stand-alone AI capabilities, because the ROI would be through the roof.
Subscribe to the McKinsey Talks Operations podcast
Daphne Luchtenberg: How should banks balance the short-term pressure to deliver tangible impact from AI initiatives with the need to invest in a longer-term operating model?
Abhilash Sridharan: A key point to consider is that it is critical for banks to have a telescopic lens while also solving for near-term impact in the next 12 to 18 months. And the reason I’m saying that is because the operations of the future are expected to change, and banks need to be prepared for it or risk being disrupted by a combination of fintechs and big techs.
The nature of work, including team structures, product, and workflows, are expected to change. Employees are expected to move from spending 80 percent of their time on coordination and rule-based execution, like collating information and writing credit memos, toward spending 80 percent of their time on engaging with customers, engaging with stakeholders, making key decisions, and focusing on innovation. For example, reviewing a credit assessment question that is being flagged by the credit memo agent. That’s the evolution that’s going to happen.
The second one is product. It’s expected to evolve from standard, one-time static outputs—which are prone to errors and subjectivity across checklists, approval notes, and memos—toward AI-led dossiers, with reasoning and self-updating insights.
The third is workflows. Across the globe, one of the biggest challenges we’re seeing in banks is that you have linear, siloed SLA-bound1 workflows, which increase the turnaround time in end-to-end processes. That is going to transform into agent-led orchestration that adapts to context and accelerates end-to-end flow, while escalating exceptions to humans in the loop.
The last one concerns team construction. This is expected to evolve from homogeneous human teams into diverse, global teams that comprise humans and agents working in sync with supervisors, who are developing the skills to manage across that.
It is critical that banks invest in and master this technology, despite all the complexities. This space is expected to evolve, and banks will have to get there eventually.
Daphne Luchtenberg: That is indeed an extremely complex arena of impact, with all of these interlocking areas. David, let me come back to you. What could a successful agentic AI implementation look like for banks, and what would its potential impact be?
David Deninzon: This reminds me of the time of automation and RPA [robotic process automation] that we went through about a decade ago. And our view is that, depending on the operations journey or process, we should expect north of 40, 60, or 70 percent of capacity creation coming from AI—but only if a few conditions happen.
The biggest risk we face is what we call the “fingers and toes problem”—creating models and capabilities that automate only a portion of what individual contributors do. This is basically why automation and RPA failed in the past; it was very difficult to automate all the capabilities of a frontline analyst.
In order to capture that impact, a few things need to happen. This has to be a priority for the top of the house. We have seen a number of our clients in the US where the CEOs are driving the agenda and holding their teams—including business, technology, operations, and risk—accountable for delivering on this.
No single function or person will be able to drive this across the enterprise. It has to be a group of committed leaders, with clear targets and accountability. As part of the impact, we expect people to improve their ability to do and scale their work, with a single person managing roughly 20 to 30 agents in some cases.
Think of it as an individual contributor working with a team of 20 to 30 colleagues to deliver an outcome. We also expect that a lot of the operations and back-office functions, like HR and financial planning and analysis, will get a significant boost from these capabilities. We’re seeing that happen faster today than in other areas.
The one area that our clients have been most interested in, at least in the US, given the regulatory complexities, has been around risk management as the first line of defense. Things like RCSAs [risk and control self-assessments], control management, documentation, and model reviews often involve a lot of data gathering and simple initial analysis. AI can help free up a lot of time on the front line to deliver value to customers, increase the number of cases and widgets that get reviewed, and improve the accuracy of those reviews. As a result, leaders in the organization can focus on identifying risks and making decisions to mitigate them rather than spending 80 or 90 percent of their time collecting data to be able to have those discussions.
Daphne Luchtenberg: What actions should leaders take to get started?
Abhilash Sridharan: That’s another great question. To drive AI-led or agentic-AI-led transformation, banks will need to unlock five enterprise-wide blocks that are critical for a successful operations transformation.
It is important for a bank to set a bold, bank-wide AI operations vision that is aligned with its strategy and has clear financial and operational outcomes to help deliver a competitive advantage.
The second is to prioritize high-value domains for end-to-end transformations that are specific to your bank and the region you operate in. For example, in specific Asian markets, the cost of labor could be cheaper. Therefore, to create full impact, banks need to provide a comprehensive offering across a number of areas, including operational efficiency, risk and fraud effectiveness, better customer experience, and better employee experience. And finally, this would have to be supported by specific milestones that link up to it.
Daphne Luchtenberg: When we kicked off this program, we talked about a paradigm shift facing the banking industry. You’ve outlined the landscape here, and it is indeed a significant transformation. Are we already seeing aspirations and vision clearly articulated at the senior levels of global banking? Are there any bold movers, or is there more that needs to be accelerated?
David Deninzon: I think it is very much happening. If you see the third-quarter earnings calls for the large US banks, AI has a front-and-center role. Most CEOs already have a view of AI’s impact. Many banks have a plan, and many others are already executing. There have been news articles about one of the trillionaire banks and its strategy and how it is driving change management, so this is very much happening.
There are two potential limitations. One, the technology is moving too fast. The vision is there, but the tools change very quickly, which is giving some institutions a little bit of a pause. The other limitation is a perceived lack of clarity on regulatory guidance. That could slow down or prevent some capabilities from being rolled out in full until everybody is comfortable that the capabilities work effectively and that the quality is sufficient, especially the customer-facing ones.
Daphne Luchtenberg: Abhilash, is that also what you’re seeing?
Abhilash Sridharan: If I take an Asia lens, I see a wide spectrum as far as this is concerned. On the one extreme, and I’m quoting this from the recent McKinsey interview with the CEO of DBS, Tan Su Shan, who called out that AI is “eating the world” and that DBS’s future is an AI-enabled organization “with a heart.” That’s one extreme.
And then we have other sets of organizations that have a natural proclivity to become fast followers, while focusing on three to four key domains where they can help establish conviction with the board or senior leadership. These companies first ensure that they have conviction, and then double down when they see the entire industry moving in that particular direction.
And then, of course, at the very back end, you have people who still are not sure if this technology will scale and who want to buy time. And some institutions have inherent challenges around tech talent, tech infrastructure, and so on. So it is a wide range that we end up seeing.
The other encouraging thing we see, especially in India, for example, is that [the Reserve Bank of India] has taken a forward-looking posture. They recently published a well-written AI report where they talk about specific areas, domains, and use cases where banks can effectively deploy AI to improve the customer experience for their respective client bases. So a combination of all of that makes me cautiously optimistic about this.
Daphne Luchtenberg: David, you mentioned earlier the important role the CEO plays in setting the vision. Talk a bit more about the collaboration that is needed between the CIO [chief information officer] and the COO to bring all these strands together.
David Deninzon: It’s a very interesting question, because it’s not only the CIO and the COO; other C-suite executives have to be involved. In our view, the main goal that the CIO has in all of this is to build the right infrastructure and enable the right capabilities for the organization to leverage—for example, stacking LLMs, creating ontology layers, and providing access to the different tools and capabilities for the businesses and functions to build upon.
The second critical role in all of this is the chief risk officer and the risk organization. As you know, banks need to go through model risk management. Every model used at a bank needs to be documented, reviewed, and approved by the model risk management committee. So having a clear operating model with risk—where the bank understands the risks from AI and can move quickly to approve different AI and agentic AI models coming through from operations and other areas—is going to be critical in all of this.
And then the role of the COO is to work with others by setting the agenda for the operations organization and figuring out which domains need to be transformed first. Where does it make sense to transform the domains, and what is the experience that could be or should be delivered through operations? Leveraging the tools that the CIO has put in place, leveraging the risk guardrails that the risk organization has put in place, working with HR to make sure that the right capability building exists to work with AI—it’s not a one-person task. The leadership of the bank has to work together to be able to deliver AI impact in operations.
Daphne Luchtenberg: Abhilash, what is your perspective on that?
Abhilash Sridharan: David summarized it very well. And from what we’re seeing in Asia, there are three components on this one. The first one is the combination of the COO and the CTO [chief technology officer] working in a two-in-a-box model, with the COO identifying the top ten processes across the domains they would pursue first to drive measurable operational impact. This is super critical.
For example, at one of the Asian banks, we recently took the bank-in-a-box model and broke the bank’s operations into about 600 processes and subprocesses, across a combination of businesses and value chains. And then the COO’s team identified the top ten processes that would give them the highest bang for their buck. Then the CTO or the CIO and their team essentially ensure that they have the right data, platform, ML [machine learning] pipeline, and governance in place to support it.
The second area where the CTO plays a crucial role is ensuring that, for mobile-first customers, the bank can translate use cases and their associated impact across all channels where it reaches customers, from onboarding and acquisition to servicing and renewal, as far as clients are concerned.
The third part involves governance risk and scaling, which is where the partnership with the risk compliance team becomes critical. And so the three of them, the tripartite combination, would have to come into place for this to be a success.
Daphne Luchtenberg: And that leads me to the last question, which is when you’ve got that handshake between the senior leaders, what kind of tactics and programs should they be putting in place to bring the whole workforce along with them? David, what have you seen so far that works quite nicely there?
David Deninzon: It is not necessarily different from your big transformation change management approach. Derek Waldron, from JPMorgan Chase, spoke with our colleague Kevin Buehler about this. JPMorgan has used a number of different levers to make sure that its employees are understanding and adopting AI.
So, for example, this includes lots of marketing to make sure that everybody is aware of the capabilities that exist and where to find them. Also, a set of loose metrics—not tracking individual usage but tracking usage of tools and capabilities—can reveal what is more helpful and where to continue developing versus where not to.
And finally, this involves providing a lot of training to both senior leaders and midlevel leaders to help them understand how to engage with AI and how to encourage their front lines to use it. Additionally, ensuring that the front lines themselves understand how AI can help them in their day-to-day and how to manage AI responsibly is very important in a regulated banking environment.
Abhilash Sridharan: I was recently having dinner with the COO and the CTO of a large Asian bank, and they shared a really memorable analogy. They said, “Look, AI is not the pilot replacing the crew. It is the new engine that makes the aircraft go farther and faster with the same team on board. And the sooner you can get your employees to realize the truth, you’re able to really get AI into the bloodstream of the bank.”
I thought that really captured the essence of how to democratize AI in a bank. We talk about this more in our recent report on AI-enabled operations transformation in Asia.2
Daphne Luchtenberg: Clearly a lot of shifting platforms, a lot of things happening. The door is open for banking leaders to see value from agentic AI. But to truly capture value, leaders need to move fast and cautiously at the same time.
