Customer care is changing rapidly as businesses continue to automate different touchpoints. Some organizations are on the way to automating as much as 70 percent of customer contact, with others playing catch-up. Daphne Luchtenberg is joined by Eric Buesing, a McKinsey partner and global head of the firm’s customer service operations group, and Gadi Shamia, CEO and cofounder of Replicant, a leading AI-powered contact-center automation platform. Together, they bring us the trends and context that are shaping customer care today and share their advice for leaders looking to get the most out of AI for customer experience.
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. I’m your host, Daphne Luchtenberg, and you’re listening to McKinsey Talks Operations, a podcast where the world’s C-suite leaders and McKinsey experts cut through the noise and uncover how to create a new operational reality.
Priorities for customer care professionals have converged over the past decade. Customer satisfaction remains the number one priority, but improving operations, implementing the right tech, and generating revenue are closing the gap. The pace of change is accelerating, and new forces are helping AI scale faster—raising the bar for both tech and human performance.
Here to talk about the state of customer care in 2025 are Eric Buesing, a McKinsey partner based in our Charlotte office and global head of our customer service operations group, and Gadi Shamia, the CEO and cofounder of Replicant, a leading contact-center automation platform powered by AI. Thank you for joining us.
Eric Buesing: Hi, Daphne. Thanks. I’m really excited for the discussion today.
Gadi Shamia: Thank you for having me. I’m excited about this conversation.
Daphne Luchtenberg: Eric, I’d like to start with you. What are you seeing today in the landscape of customer care?
Eric Buesing: Care, and even customer operations more broadly, has massive, massive potential to be disrupted. This is not a function that is a stranger to change, but I think something feels different, particularly in the past two years. We’re seeing new capabilities really come into focus. And these are coming in the forms of new tools that employees and customers alike are starting to take advantage of. I also think what’s different is that leaders are taking notice, too. This is more of a CEO agenda than it’s ever been before. And we recently surveyed and captured insights from 440 executives across industries. And it offered us, I think, one of the clearest pictures of how AI maturity is really starting to evolve and separate what we call leaders and laggards.
In our survey, leaders comprise the top 10 percent of the respondents, and laggards were at the bottom 30 percent. And what we really started to see, and what we’re learning, is that leaders are separating from the pack. They are building the AI muscle to scale. They’re pursuing gains in customer experience and efficiency and sales and growth. And they’re starting to see AI as the future of what I’ll call full-service care.
Subscribe to the McKinsey Talks Operations podcast
Opposite of that, laggards in some of the organizations that we’ve interacted with are more hesitant to let go of what I’ll call the human-only model.
Gadi Shamia: I truly agree with you that we are in this good-to-great moment. The ones today that automate, the ones that don’t automate, the ones that use AI, and the ones that don’t use AI look pretty similar because for both of them, the majority of customer service is still being done in a more traditional way. Give it a couple of years, and you’re going to start seeing pretty radical differences between the companies that still keep their customers on hold and the companies that automate, with every task or 80 percent of the tasks in the contact center being achieved in under 120 seconds.
Daphne Luchtenberg: Gadi, tell us first a little bit more about how the technology of conversational AI has actually evolved. What is possible today?
Gadi Shamia: The early AI agents were using AI to navigate a preexisting map. If the road was paved, we could take you down that road or even jump you to a different location, but you always had to land back on a paved road. The biggest difference today, and what opens up endless opportunities, is that modern AI agents don’t follow a fixed path or flow. Instead, they work toward a set of tasks and goals, guided by instructions and examples of good and bad behavior. That gives them a level of flexibility that is often better than what humans can achieve.
Daphne Luchtenberg: And Eric, with that evolution, how should we be starting to think about agentic in a customer care setting?
Eric Buesing: I think that the terms agentic and agentic AI are thrown out a lot, an almost catch-all for everything, like digital was 15 years ago. Personally, I like to think of agentic AI agents like LEGO blocks. Each one is a small, specialized AI worker that’s designed to do just one thing and do that one thing really well. It might be like fetching the right document, running quick math or metrics checks, scheduling a meeting with a technician, critiquing an output, or checking compliance.
On their own, they’re simple and focused. When you connect them together, just like when you put LEGOs together, they form really complex structures and workflows. And when they do form these structures, they operate very much like you would when interacting with another human. They can reason. They can prioritize. They can remember current and past interactions. And they do things faster, without wait times.
And as Gadi mentioned, they converse in a much more natural and conversational way. The AI can be more casual, more familiar. You might even term it playful in some of the right inquiry types. And we might be approaching a period where we start to see some interactions and inquiries where customers prefer interacting with AI rather than, say, a human agent.
Daphne Luchtenberg: I was getting quite excited about that coming into all of the calls that I have to make to retailers and others. I’m not really seeing that yet today, but if we go back to Gartner in 2023, they were saying there would be an 80 percent adoption of gen AI in customer care by the end of 2025.
What are we really seeing? Eric, I’d love to hear your thoughts first on that.
Eric Buesing: I think that was right in the height of the hype cycle, right? When generative AI was being introduced, it was exciting, and it still is. I’d say most organizations are doing something in AI. And the majority of what I’ve seen is that it’s focused on helping internal employees.
There was also, even more recently, an MIT report, I think it was in August, that created a lot of buzz. It was The GenAI Divide: State of AI in Business 2025. And I think, to summarize what it was saying, was that 95 percent of gen AI pilots were delivering little to no measurable impact to the P&L. And that created a lot of buzz. And I do think there’s something there. I don’t know if it’s 95 percent, because I certainly see organizations that are achieving impact to the P&L, but we’re seeing some use cases where internal employees are using it at scale. It could be call summarization or searching on behalf of the agent or the customer faster. So there are cases, but I’d say generally the pace of deployment and adoption by end customers has been slower.
But I do believe—and we saw this in our survey—that leaders are pushing ahead. They’re building that muscle, and they’re starting to see momentum in unleashing kind of a delightful AI experience to their customers. And I think we’re just going to see more and more of that.
Daphne Luchtenberg: Gadi, would you agree?
Gadi Shamia: Yes. To give you a little bit of perspective on being in this conversation for six years now, AI, for many companies in the past two years, was a shiny object, and IT wanted to try it out. They wanted to get their hands dirty. They wanted to try something. And they were deceived by the simplicity of building a demo with AI. And you see it not only in large companies; you see very busy people building an app on the side, like a passion app, and ship it, which is something they could not have done two, three, or four years ago without quitting their day jobs.
So it’s much easier to create a demo. It’s much easier to create a very basic solution. It’s as hard to create a product that will work at scale, that will stay updated, that will be resilient, and so on. It became pretty deceiving because building, say, your own CRM instead of using Salesforce was a very daunting, not glamorous task. Building a cute AI demo that does call summarizations or some voice automation, which the CEO can test, is actually pretty easy. But getting to the full solution is pretty hard.
I totally agree that the Gartner prediction is correct. Everybody’s trying AI, but very few people are successful.
Daphne Luchtenberg: So, Gadi, how do these companies, how will they get their customers comfortable with the agent and allowing the agent to do that work for them?
Gadi Shamia: I actually think it’s not about the customers. I think it’s about the companies. I think there’s not much that companies can do other than putting it out there and making sure they choose the right technology, and that, collectively, the high tide will raise all boats. But as consumers, when we call one good bot and another good bot, we need to erase 20 years of speaking with terrible IVRs [interactive voice response], and that wouldn’t happen overnight.
Now there are things we can do as the technology advisers of this company that have to do with conversation design, and maybe we can open up on this topic a little bit because it’s not discussed a lot.
There’s a whole art to it. We hired our first conversation designer in early 2020. We have a very large team, including PhDs, that does research and advises our customers around that. And this is actually how you do it. You deploy it. You deploy it well with flexible technology. You design it well so the UX is good. And it might not help with a person the moment they hear, “I’m an AI agent,” or when a core line says, “agent, agent, agent.” But over time, these people are going to get used to having more, better experiences and will give more agents a chance.
Eric Buesing: I agree with you, Gadi. I think the experience matters tremendously. The technology does work and will work even better in the future, but the experience needs to be surprising and delightful. The quality of the response, the latency of how long it takes to respond, the casualness of the free-flowing nature of that response—all of these things matter just as much as how effective and quick that agent can resolve the issue.
And where we’re heading with this theme of empathy is the question: Can an agentic AI voice agent be empathetic? And it’s probably how you define empathy, right? That feels like a very human-to-human word.
But if you think about it more of understanding and being able to recognize when a human needs a different type of interaction. Now maybe you could interpret this as empathy: It remembers what the customer asked last time. It remembers interactions that garnered the right responses, and it can change its pitch, tone, and speech that demonstrate that type of understanding. And I think another thing is that an agentic agent never gets frustrated. It can always remain on an even keel and calm, especially when we’re talking about sensitive topics. We’re going back to this idea of “Are there certain inquiries we’ll see leading from the front where people prefer to interact with AI?”
And I think we should expect this. I think we should expect this from the AI. And very transparently, I don’t think there’s enough thought put into that experience. There’s a lot of technical thought put into making it work and not enough on how do we make it, as I said, surprising and delightful, because that will cause the individual who might have said “representative” in the first five seconds to pause and listen and say, “Well, that’s different. Let me engage with it.” And if it continues to be interesting and continues to have that understanding, I think you start to see greater and greater adoption.
Daphne Luchtenberg: So we’ve kind of sketched out what “great” looks like, Eric and Gadi, and I’m getting quite excited about my next call with a customer care center. But we’ve obviously talked a lot in the past about pilot purgatory, particularly when it comes to any kind of digital transformation.
That is also a phenomenon in this evolution, right? Can you talk a little bit more about that, Eric?
Eric Buesing: This has to be a top-of-the-house priority. And it doesn’t mean it has to be the CEO, but it has to be a leader who is senior enough to have the influence so that these investments get the attention that they need. I think that’s one. I think another one is picking a few areas to really place bets versus trying AI everywhere.
I think another thing is starting to think about scaling and the measurement, the KPIs that you’ll use from the beginning. And that means investing in the change management, investing in the adoption from the start.
If it’s for employees, it’s helping them understand the importance of how this actually benefits them. And in some cases, putting in means by which to tie the AI to their performance. I think that’s another thing.
I also think getting out of pilot purgatory means both funding the initiative fully and having the right talent assigned to lead it, technically from both the business side and from the product side. And that talent, the people who you put against it, they need to believe in it, too. This can’t be an assigned project where you just go launch AI. You have to have leaders who are excited about it, who believe that this is the future, and they’re willing to bring together, as I mentioned, the leaders from business, from product, from tech, to collaborate together in order to get to an outcome.
Gadi Shamia: I would say the first thing I would ask a company is, do you need a pilot? And there are times when you do need a pilot and times when you don’t. For example, I will tell you about our first pilot, which was with DoorDash in 2019. DoorDash came to us and said, can you automate outbound food orders for us? And we said, we think we can.
But what they had, which was a game changer for us, was a metric. They said, if you can successfully automate 70 percent of these calls in 2019, this is the ROI we’re going to get. This is the impact on customer satisfaction for a variety of reasons. And this will be a very successful pilot for us.
So this pilot was excellent for two reasons. One, it had a clear goal: a 70 percent success rate. There was a way to measure its success. The second was that we didn’t know if we could do it, and we couldn’t prove to DoorDash that we could do it. It was absolutely an experiment, which is why it was a great idea for a pilot. We ran it for two weeks, we got to 74 percent—I still remember it—and then it took a year with procurement to get the deal done, and DoorDash is still a customer. So that’s a great example of when a pilot is needed.
Daphne Luchtenberg: And Gadi, what other things do organizations need to do or change to really make this bold move?
Gadi Shamia: Two things come to mind. We’re at a point where technology is not your limiting factor. The limiting factor is willingness to change, and the organizational changes that need to happen when AI gets to automate 50, 60 percent of your volume. The second is API. This is the most mundane thing. If companies had used all the time they spent, all the engineering hours, months, and years on trying to build AI solutions already built by vendors, and instead of that, had open APIs to transact on behalf of the customer, they would have been in a much better place.
Everything that we need to do, every agentic action, needs to be supported by an API or multiple APIs. They’re not hard to build, depending on the system you use, but they’re always deprioritized. And they are the limiting factors right now to get from the 2, 3, 5 percent automation some companies see to the 50, 60 percent.
So if I’m a CEO now and I want to drive my agenda, I’m actually talking with my CIO about APIs, and what’s the road map, and what’s the limitations on opening up every single user interface field that a human agent has access to that needs to be supported by an API. Otherwise, you’re going to be stuck very quickly with building FAQ bots that no one wants to interact with.
And the third, and maybe the most interesting one, is that people need to think outside the box because a lot of the conversations I hear are “we used to do it with humans, and we want to do it with AI agents right now.” And I would say that’s interesting, and that’s an easy one because you can map it one to one. But what about the things you don’t do with humans, because they’re just not scalable? If you’re a large moving company, you don’t have the capacity to check in with every customer 30 minutes into the move about customer satisfaction.
But what will be the impact on customer retention, on customer satisfaction, on decreasing postmove complaints, if you can exactly at the half-hour point when the move starts, place a call and have an intelligent conversation with someone about their moving experience. So don’t overthink what you used to do, because you made decisions through a limitation of the human model. You can now make new decisions without those limitations, and they will look different.
So maybe it goes to this third part, which is AI does not need to mimic what humans used to do, not only in the use cases, but even in how you handle the specific use cases that humans used to do. Take a step back and rethink what would you do when you have unlimited elastic and inexpensive capacity to create this change for your business? What are your goals, and how could you map these goals into the capability of a technology?
Daphne Luchtenberg: That is wonderful to hear. It sounds to me that you’re both saying there is no holding back anymore. This is now becoming a conversation at the highest levels of the organization. CEOs are really excited about this as a prospect as well. But if you look to the future, what kind of advice would you give now, Eric, to any business that has a customer front line and needs to think about innovating it? What should they be thinking about right now to move into the leader quadrant? Eric, first to you, and then I’ll finish with you, Gadi.
Eric Buesing: I’d say two things. One is that it is still easy to get enamored by a new, shiny object. And I would encourage leaders and organizations to have an independent view, their view of specifically where and how AI and agentic can both change the experience materially, but also show up in the P&L. What’s that path all the way through? I think that’s one.
And the second piece of advice I’d give is to be prepared to challenge the way things are done today in the organization. Challenge the norms. Be prepared to ask tough questions about risk tolerance. Be prepared to challenge model risk management. Be prepared to challenge through mindsets that the change isn’t possible, that customers won’t adopt this, or employees won’t use it.
I think we have to really ask ourselves—to get to a new model, we have to think differently—and how are we going to think differently to get there?
Daphne Luchtenberg: Gadi, what do you think?
Gadi Shamia: I completely agree with Eric on his points. The one I would maybe add to be more specific is when you think about your customer service operations in a world in which half your calls, maybe next year, 70 percent of your calls and end chats are fully automated, it’s a whole redesign. And the redesign is not only from a software perspective.
You may have used a specific contact-center solution or scheduling solution that met the requirements when every single conversation would have been handled by an agent, and now all of a sudden, when almost every conversation is handled by AI, it requires you to reassess every piece of software you have, and ask yourself, do I need the software anymore? I may not need this complex scheduling software anymore, because my scheduling task has now become much simpler.
The second question I would ask leaders is what do you want to do? How do you want to change your customer service when everything transactional is going to be automated? If you invest in it, it’s going to happen before you can even make an organizational change. So it’s time to think about it now. What do you want your human touch to look like?
I talked with a CEO of a large retailer, and they have a vision in which their human agents are actually more like customer success managers, more of the B2B model where a human agent actually does not speak with all the customers. Whenever a random person calls, they’re assigned to a group of 100, 150 top buyers of this brand who know a little bit about them, have the personal record, and can make some connections that AI can mimic but not really create.
So this discussion should happen now because this change is going to happen over a long period of time. You have to find those agents that can grow to be customer success managers. You have to train them. You have to promote them. You might have to hire people from the outside who are capable of doing that. And people changes take time, so I would encourage people to actually think about head count, the type of people you need to have in this contact center of 2027, because they will have to start investing in hiring or training—or both—these people today.
Daphne Luchtenberg: Really well said. So it seems clear that the organizations that will lead in the next few years are the ones that are going to be bold enough to rewire their customer care around AI—not just seeing AI as a tool, but really as a core capability and helping them drive growth and new value-add services.
You’ve been listening to McKinsey Talks Operations with me, Daphne Luchtenberg. If you like what you heard, subscribe and stay tuned. Another great episode starts now.
