Let’s look at the next chapter of AI—agentic AI, and how it could unlock the next generation of operational excellence and productivity in service operations. We dig into the high stakes of continued investment in digital transformation and the potential payoffs. We discuss collaborations between the chief information officer (CIO) and COO, and how thoughtful talent strategies can set organizations up for continued digital success.
In this episode of McKinsey Talks Operations, McKinsey’s Christian Johnson talks to Malte Kosub, the cofounder and CEO of Parloa, an agentic AI platform that helps customer-centric enterprises build and manage millions of AI agents for customer support and communication. Joining them are Oana Cheta, a partner in McKinsey’s Chicago office who leads generative AI and agentic AI in service operations in North America, and Brian Blackader, a partner in McKinsey’s Düsseldorf office who spends much of his time working with service providers in the customer service space.
The following conversation has been edited for length and clarity.
Christian Johnson: Oana, let’s start with you. It’s two years after the launch of ChatGPT, and organizations are looking to implement generative AI and agentic AI in their operations. What are you hearing today from clients?
Oana Cheta: Thanks, CJ. I’m excited to discuss how we finally move past experimentation. More than two years in, companies are shifting from AI curiosity to AI accountability. I fully engage in agentic AI execution now. So, what’s top-of-mind for our clients as they move to the next phase? First, return on investment and business impact. Companies are under increasing pressure to demonstrate real, measurable returns from their AI initiatives and justify further investment.
Next, it’s not about automating tasks anymore. It’s about redesigning how work is done. This is not an efficiency play but rather a transformation play. So, deploying AI alone or generative AI alone is not enough. Companies must redesign their processes to integrate AI at the core of their operations. AI should write decision-making, workflows, outcomes, and more. It’s not just an add-on.
Then, of course, there’s a build, partner, and buy decision, and then finally we’re moving into the shift from generative AI to agentic AI. The future is not about managing AI but collaborating with AI agents that think, act, and optimize in real time.
Christian Johnson: Excellent, thank you, Oana. Malte, building on Oana’s comment, could you clarify for our audience the links between gen AI and agentic AI?
Malte Kosub: Happy to go deeper into that topic. Everything started with generative AI, which was about creating images, texts, and videos. In the beginning, companies, especially customer support functions, used that technology to run internal FAQs when customers were calling or reaching out.
The second step was creating answers to give the outside world, such as for customers through chatbots. What we’re seeing now is that we’re moving from a generative AI world into an agentic world. Agentic AI is more about a goal-driven approach, where the AI is capable of making decisions and taking actions on behalf of the human.
In this third phase, we see companies implementing agentic AI to get value from goal-driven and process-driven tasks and work. We are already seeing enterprises moving to a fourth phase, in which they are integrating agentic workflows into their dealings with the outside world, and agentic AI is helping customers to solve problems in real time.
Christian Johnson: So, how do companies prepare themselves for the two big transitions you have just described? First, to agentic AI internally, and then to linking agentic AI to the outside world?
Malte Kosub: What is important to get right is to build a process for how to deploy AI agents that actually take action on behalf of a customer. You need to think about how to orchestrate a lot of different AI agents at scale in an enterprise environment. You need to think about how to define the right compliance layers so that the guardrails and all security measures are in place. You need to think about how to simulate and evaluate the millions and millions of conversations in which your AI agents will have to guarantee the highest accuracy possible.
This is a completely new way of implementing AI into an enterprise, and there needs to be the right infrastructure in place to bring agentic AI to the masses. It can bring you significant value. It can transform your customer experience, but you need to start implementing the right processes and the right infrastructure to get there.
Christian Johnson: That’s a good segue, because one of the things we’ve wanted to give our audience is a sense of how these journeys are playing out in organizations today. Brian, what are you seeing with your clients? What are some of the cool and interesting things that they’re starting to do with gen AI and especially agentic AI?
Brian Blackader: It’s certainly an interesting time to be working in this space. If you speak to a hundred different clients, what you’ll find is that 99 of them will tell you, yes, we’re doing something.
The part that gets the most airtime is about copilots, chatbots, and voice bots. What you see there is that most clients that you talk to—95 percent, in fact, when we did a recent survey—were still stuck in pilot phase and only 5 percent were scaled. So it’s not easy. The ones that do it . . . it’s really quite impressive.
Things like copilot solutions and AI coaching, where agents are being fed the information to create a better experience with a customer in a quicker way, those actually have a bit more traction, with between 30 percent and 45 percent currently at scale, depending on the use case. So I think that we’re going to see copilots picking up. We’re going to see voice bots and chatbots improve as the technology improves. But certainly, where we’re seeing the most traction for our clients, where we’re seeing the impact at scale, is on some of the things that I would say are a little less obvious.
Christian Johnson: What are some good examples of what is less obvious, the things that maybe get us out of this traditional customer care space and more into other back-office or mid-office functions?
Brian Blackader: One example is coaching for the agents. We worked with a telco in North America and another one in Europe, where they have built algorithms that identify opportunities for coaching and for learning for call center and field service agents. The technology is able to intervene and say, “We see that there’s a particular type of call or a particular issue that you’re dealing with in the field and that you’re not doing so well on.”
We produced a customized learning journey for that employee and reviewed its effectiveness. Over a 12- to 15-week period, the algorithm helped by pushing various pieces of content to the agent, from a small video to a PDF—things that take two to three minutes to consume on a regular basis, a couple of times a week. If it’s effective, then we push similar content to people with a similar profile.
Christian Johnson: What sort of metrics are improving in these organizations?
Brian Blackader: We’re seeing improvements across the board, but the first and most obvious one where you can really tie a dollar value to it is a reduction in average handle time—how long it’s taking somebody to handle an inquiry. But more importantly, from a customer point of view, the first contact resolution is improving, meaning that people don’t need to call back to try and resolve their issue because they got it resolved the first time.
Christian Johnson: That raises a great question, because when we talk about implementation, one of the long-standing issues in many organizations is the collaboration that’s needed between the chief operating officer on the one side and the chief information officer on the other. So, Oana, over to you. How are you seeing organizations overcome this kind of long-standing barrier and learn to speak a common language?
Oana Cheta: Alignment between the CIO and COO is critical for success. Without it, we’ve seen companies risk misalignment that may undermine execution and ultimately dilute business value. We’ve seen that joint ownership of outcomes works. Successful companies ensure shared accountability between the two branches. They align business outcomes, tying KPIs and incentives to those goals. AI gets fully integrated into those operations, not just siloed to IT. There’s also a need to embed collaboration into decision-making. As I said, it’s not about technology. It’s how we create those feedback loops where operations in different functional areas like sales would guide AI decisions, ensuring it’s enhancing rather than disrupting daily operations.
The second thing that comes to mind is role modeling leadership. It’s leaders who champion collaboration between IT and operations that are driving true cultural change right now. Coauthored road maps and joint strategy sessions reinforce AI as a company-wide initiative. For instance, at a client that was in the process of defining its vision, the focus was not on where AI fits. It was, what should AI own as we execute on our vision?
To really think through and implement agentic AI, think about it as a team member. The push in this case was “What tasks can we delegate?” and “How do we measure success?” As AI agents evolve from assistance to fully owning workflows, with humans overseeing and stepping in only where necessary, the COO and CIO and their teams must collaborate closely to design a performance management framework that provides real-time feedback. The specific framework should support agility and adapt to the changing dynamics of a digital or hybrid workforce.
Christian Johnson: What does it mean pragmatically?
Oana Cheta: It will allow both teams to track performance gaps, refine or redefine processes, and ensure human oversight maintains the right balance with automation. By working together, the two offices will ensure a seamless integration of the AI operations while maintaining effective oversight and continuous improvement.
Christian Johnson: Fantastic. Thank you. Brian, how are you seeing this play out in some of your clients’ organizations?
Brian Blackader: The point where the job of the CIO ends and the COO’s begins is starting to blur. In the past, it was a little more clear that the infrastructure needs to work, the telephony system needs to work, the CRM [customer relationship management] needs to be there, and so on. That was the job of the CIO, and operations needed to deliver the experience. As new technology is delivering the experience, it’s erasing some of those traditional lines. What we see now is that the COO and CIO, or the folks in the role of technology and those in operations, really need to work together in teams to make things happen.
Christian Johnson: And when it doesn’t?
Brian Blackader: If you just make the implementation of agentic AI, or any other technology solution, purely a technology implementation project, it’s more or less bound to fail. From my point of view, it really needs to be both teams working together to make sure that it’s secure, to make sure that it integrates with all the other systems. That’s the job of the CIO, and the operations need to make sure that it delivers impact and that there’s an ROI on these things, because they’re not free. You need to make sure that at the end of the day, you’re able to have measurable results, whereas, if we go back two years ago, people were just trying to do something. Increasingly, the CFO is looking at and asking both the CIO and the COO, “OK, you’ve spent this money; what do you have to show for it?” And both teams are accountable for that.
Christian Johnson: What sorts of cultural or managerial metrics do companies need to rethink to enable this collaboration? Meaning, what’s on the backend, or on the human side of this, that helps this work more effectively?
Brian Blackader: I think that the answer is less about the difference between the CIO’s office and the COO’s office, but rather between operations and whoever owns the front end. What do I mean by “front end”? The website, the app, how customers enter the world of care before they speak to a human. What we often see in organizations is that those have two very different owners. Sometimes they talk to each other, but maybe they don’t. And if you really want to do this right, particularly regarding customer-facing solutions that are speaking to a customer, interacting with a customer, those two departments need to work together.
I would even argue that there needs to be one overall ownership of the entire journey, starting from where you entered and where you end up with a human in the end, which I think is different than how most organizations have thought of this in the past. There’s a digital group that owns the engagement when the human is not involved, even to the point where they might be dealing with a chatbot, and a different group that then deals with the human interaction. I think that line is being blurred, and it becomes very difficult to make it happen if you don’t have that joined up in some way.
Christian Johnson: And Malte, for those that are getting it right, those organizations where the CIO and COO organizations are collaborating increasingly effectively, how does agentic AI itself play a role in the next evolution of this technology?
Malte Kosub: Great question. Because we believe that the customer experience will change more drastically than it has ever changed before, every home page, every app, every customer touchpoint will look different in the next three to five years. Every touchpoint will become conversational, and there may be a personal AI agent talking to customers at each one.
Imagine you’re calling an airline. Right now, you would probably wait in line for 15 minutes, then talk to a different person every single time you called. But now imagine you’re calling an airline, and after one second, your personal AI agent picks up and says, “Hey, Christian, how are you? Have you decided if you want to upgrade to business class for your next flight from New York to San Francisco? If not, no worries, just call me back in 20 minutes.” And in 20 minutes, this same personal AI agent would answer the phone again, and after one second, the AI agent with the same context will say, “Hey, Christian, have you decided? I can help you with upgrading your flight.”
Christian Johnson: But let’s say I’m in a meeting and I can’t call.
Malte Kosub: If you don’t want to call, you could also take out your smartphone and write a message on WhatsApp or iMessage to have a chat about upgrading your flight. And if you don’t want to chat or call, maybe on Monday next week your personal AI agent might call you and say, “Hey, Christian, there are just two seats available in business class. Do you want to have one of those? Otherwise, I’m going to give those seats to someone else.”
It’s a completely different relationship between the customer and the company. If an airline has 100 million customers, it will have 100 million personal AI agents. And those personal AI agents are guiding customers along the entire customer journey and not just doing customer support. They’re doing sales marketing. They are building a relationship. So that’s a paradigm shift in how customer experience is being done, and every enterprise now needs to start thinking about how to implement AI agents across the different touchpoints.
Christian Johnson: That is obviously a compelling vision. It does raise some really profound questions about talent and what sort of talent needs we are going to have. What sort of talent are organizations, indeed, whole societies, going to need? So, Oana, could we get some perspective from you on some of the talent solutions that you’re seeing and some of the talent questions that organizations are grappling with?
Oana Cheta: Sure. It’s interesting because for a long time, we’ve been talking about the need for technical talent, from engineers to data scientists, cybersecurity experts, and so on. There’s been less attention on truly how organizations should think holistically about their talent revolution. Right now, we’re moving to finally start talking about that and less about the technical specialists. There’s also more discussion about how companies can rethink their entire workforce structure, how they allocate roles, what skills are needed, how teams are getting optimized to work alongside AI and the agents, and how to manage that blended human-digital workforce.
The real challenge turns out to be a question of how we adapt the operational talent to complement AI. That’s what I want to leave you with.
Christian Johnson: What are some of the models that seem to be working?
Oana Cheta: There are different ways that successful companies are approaching this. From reskilling and upskilling, which includes everything from building literacy across the workforce to reframing roles. Not just eliminating repetitive routine tasks, but moving the focus toward relationship management, strategic thinking, complex problem-solving. Thinking about how you blend the customer care experience support with sales.
Christian Johnson: Companies have some experience with that, no?
Oana Cheta: There’s more of a proliferation of not just cross-functional collaboration when you’re looking at how you actually think through dynamic, data-informed decision-making, but also how do you merge functions that in the past have traditionally been siloed? And there’s a more integrated approach to how this type of customer service model may work with sales in the future, where they will all be integrated using AI tools to drive customer satisfaction, retention, and engagement in real time.
Thinking beyond technical skills and thinking about new roles, such as customer journey architecture and more of the human-in-the-loop roles, overseeing a lot of what AI does, the focus is to evolve that team structure, to upgrade those skill sets, and think through the new roles that would help unlock the full potential of AI and move us from short-term to long-term value.
Christian Johnson: Brian, we’ve heard a great vision of the future, which I think is really inspiring for our audience. What are some of the critical things that companies need to get right first to make this vision a reality?
Brian Blackader: I would put these hurdles into three different groups or buckets. The first is about security and regulation: ensuring that things are secure and safe, and that you’re not going to fall foul of the regulators by doing something wrong. And it turns out that most enterprise clients are quite conservative in this regard. They don’t want the regulator on their backs. That’s the first bucket.
The second is about data and systems, particularly in the enterprise segment. Call center agents and people working in customer care have to deal with a wide variety of systems and data sets to do their jobs. We often see that an agent might need to access up to seven or ten systems. The data is often not great and can be fragmented and spread across systems. That’s why, when you call a lot of places, they don’t know who you are, even though they should.
Christian Johnson: It’s very frustrating from a customer perspective.
Brian Blackader: The third category, which I think is less obvious for those who don’t spend a lot of time in contact centers but will be obvious to anybody who has worked in one, is that there’s actually a lot that an agent does that isn’t documented anywhere. I think there’s probably a belief that most contact center agents work through and read a script. That is not the case for most of what contact center agents do. In fact, we recently did a small piece of research where we found that almost 70 percent of respondents working in the contact center told us that at least 25 percent of what they do on a daily basis isn’t documented anywhere. It’s more like folklore that’s passed on through the generations of workers, and from the person that has sat in the chair before them. So, as long as we’re not able to overcome that, then it’s going to be difficult for agentic AI to provide an answer. That said, the answers exist in call transcripts. They exist in chat transcripts. Being able to unlock that as a data source will be super powerful in overcoming this final challenge.
Christian Johnson: Excellent. It sounds, at least to my layperson’s ears, that to some degree, agentic AI would be a great technology to extract some of that information in a useful way. Am I correct in assuming that?
Brian Blackader: I think that there’s absolutely an opportunity for the technology to be able to overcome some of the challenges that exist today. Evidence isn’t there yet that it’s going to happen, but this space is moving so quickly. I think it’s only a matter of time before somebody cracks that.
Christian Johnson: Excellent. We’ve had a great vision of what gen AI and especially agentic AI can now do for organizations. Before we go, I’d love to get from each of you a thought about where companies should begin in their journey right now.
Oana Cheta: I would say design for automation and disruption in the long run, but focus on the building blocks in the short run.
Malte Kosub: We always say start small, integrate the first systems, learn quick, and think big, because this is an opportunity for every enterprise out there to make big leaps forward in their customer experience.
Brian Blackader: I think that’s spot on. Thinking big, having a vision is important, but being able to start with something that you can show tangible results from is also quite important. Related to that, believing that technology is the solution is probably the wrong way to start. It’s part of it. Change management, changing people’s behaviors, and really understanding how the operations work are equally as important. So I would encourage people to not believe somebody who tells you that whatever piece of technology they’re trying to sell you is going to solve all of your problems. That’s not likely to be the case. It’s going to require a bit more effort than that.
Christian Johnson: Excellent. Thank you, Brian, Malte, Oana. I would like to thank all of our guests today for joining us to explore the breakthrough potential of agentic AI, most importantly in productivity, but also in service quality, responsiveness, and complete engagement. That is extraordinary value for today’s leaders to create.
