Powering supply chain with agentic AI

Christian Johnson: Your company's future success demands customer focused, agile, resilient and efficient operations. I'm your host, Christian Johnson, 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. So over the last several decades, supply chain leaders have made significant investments in technology, particularly in planning and insight generation. And yet executing on those insights remains a fundamental gap in many organizations today. Now advances in artificial intelligence are looking to close the gap, notably with the potential of agentic AI, that is AI based agents to orchestrate decisions across planning, logistics and operations. But what steps should companies take to get the most out of their investments in agentic AI? I'm delighted to be joined by two of my colleagues today. Uh, first, Rahul Shahani is a partner in McKinsey's operations practice based in New York. He leads the AI and tech enablement for manufacturing and supply chain. Rahul also collaborates with the World Economic Forum on the Global Lighthouse network of advanced manufacturers. Kapil Dev Bansal, also in New York, is an associate partner and global lead for agentic AI in manufacturing and supply chain. His uh, research has developed the Task Goal Orchestrator framework for autonomous supply chain planning. Based on what we've seen over the last few years. Supply chain was an early adopter of some forms of what we now call AI. What are some of the lessons learned so far on what has worked and what hasn't?

Rahul Shahani: So supply chain, you're absolutely right. C.J. was an early adopter and has invested heavily in foundational tech in this space around the erp, planning tools, analytics capabilities, control towers, and the list goes on. And a lot of those tools were focused around keeping and creating visibility across complex global supply chains. So we now have been able to have visibility and through that insight to drive complex decisions. But execution of the supply chain still remains manual, fragmented, slow and quite frankly bureaucratic. And so decisions still rely uh, on emails, spreadsheets, phone calls and really handoffs. And that tends to slow us down from when I know what to do to actually doing it. Organizations ah, uh, today are starting to experiment with execution AI or agentic AI as we've, as we will talk about. And what we're seeing is that while everybody is excited about it, we're starting to get stuck in pilot purgatory when

Christian Johnson: it comes to agenda AI and so pilot purgatory. So that's a phrase I've certainly heard before, but could we explain it for our listeners Here absolutely.

Rahul Shahani: Pilot purgatory comes uh, out of our research with the World Economic Forum on the Global Lighthouse Network. The findings that we've had and have been true for over the last decade is every time we see new tech come out, many institutions gravitate towards it very quickly, deploy the tech, but then struggle to see meaningful impact from the tech and scale the tech. And that's really what we're seeing here as well is many organizations have started agentic AI pilots have very quickly been able to show proof of concepts and proof it's working, but have not been able to change their core business processes to bake it into the ways of working and then drive value from it.

Christian Johnson: We've heard a lot about the promises of AI, uh, the various forms that we've seen over the years. How can agentic AI based agents uh, help overcome some of the issues we've seen?

Kapil Dev Bansal: Agentic AI is a technology that is focused a lot on enabling the execution like Rahul explained. And AI stacks are a lot more modular, reusable, repeatable. That can help scale up which was not possible with the prior technologies. For example, we break down Agent Ki ecosystem in three different types of agents. First is a task agent which is doing something, a very simple task for me. Second is a goal agent which is doing the very well defined sequence of tasks uh, for executing the supply chain. And third is the orchestration agent where the sequence of steps is not very well defined, but agent has to think and define what it needs to do and then it go and even does that part. So that whole system of doing thinking and repeating it over and over the time, that's something that was not possible in the prior technologies. And that's why agents are a lot more impactful. Some of the client examples where we have done this work, uh, they have shown that AI agents can reduce the cost of goods sold by 4 to 7%, a lot of productivity gains by around 20 to 50%. And then the decision cycles, right, like they are shrinking from hours and days and weeks to almost like minutes, um, and seconds. So once fully implemented and adopted, agentic AI has full potential to fundamentally reshape the supply chain. And the key takeaway is the value of AgentIQ AI isn't theoretical, but it isn't automatic either. There has to be a lot of purpose and thinking on how do we shape this journey.

Christian Johnson: And I think that's one of the things we really want to help our listeners understand is that uh, agentic isn't an automatic set and forget success mode. Uh, there are things you need to do uh, to make it work properly. So what are the sort of failure modes that you see right now with some of the agentic efforts?

Rahul Shahani: Absolutely. So what we've seen so far with traditional AI and emerging with agentic AI efforts is that they've traditionally been siloed. And so that means that they've taken a very narrow focus and lean into use case thinking. And what that does is it limits our ability to, to only focus on the task based agents. As Kapil said, we're not able to orchestrate and drive towards common goals which is what the full potential of uh efforts uh really unlocks. So as uh, we see organizations take a more holistic and thorough approach to deploying AI into their core business processes, we start to see a significant uplift in AI and the conversation shifts from what is possible to AI to, to how do I use AI and how do I trust AI to take over core capabilities within my workflows?

Christian Johnson: What are some of the differences that you see between organizations that are still at a very early stage and the ones that are really starting to see some results from their AI investments?

Rahul Shahani: 90% of organizations report using AI but only 7% have scaled. And the ones that have scaled really are looking to AI to take over accountability over certain processes or rather its individuals in those organizations who take over accountability for work done by AI. And to do that they really require the AI to be accurate, explainable and auditable which is no different than other employees in their teams and organizations. And that becomes the unlock for how do you drive real value from agentic AI? It's not just about a point solution or a technology. It's about a evolution of our business process to support AI at its center.

Christian Johnson: I'd love to give our listeners an example of the sort of from to of uh, going from this as you say use case mentality where it's quite fragmented to uh, something where you really allowing AI systems to, to take over accountability for certain types of workflows. What are some good examples of that in supply chain?

Kapil Dev Bansal: So maybe let's talk about the order management. Traditional order management can be one of the slowest, most manual, intensive and the error prone processes. Why? Because it has to be done across systems. You need one system to check your inventory, second system to check the customer order and third system to see like if you have a supplier, a logistics supplier available. So this ecosystem across different systems is very manual even though systems are efficient in their own environment. But they don't talk to each other.

Christian Johnson: So if you Say these systems aren't talking to one another. How does a company get real end to end value from their investments in AI?

Kapil Dev Bansal: In an agentic order management, uh, ATPCTP check is done by agents across different systems. And agents can also check the transportation feasibility and cost to serve validation in some cases. Cases like even if you don't have a TMS to calculate your cost to serve, agents can do it via spreadsheets or they can build the custom calculations. And then you need to have like trade off conversations around inventory, cost and service that agents can run not just based on the feelings but based on the real value. And then they can do the autonomous execution within defined thresholds. So what we did was when we did the agentic order management for one of our clients, we saw the cycle time reduced from earlier like 20 to 120 minutes. It came down to like one to two minutes. And then obviously uh, the improved customer experience made the customers happy. And then there was a lot of operational efficiency uh, because of the reduced manual work value is demonstrable when agentic AI is orchestrated across systems and it stitches the systems and informations um, across everything.

Christian Johnson: Okay, so that's a really remarkable change here because you're going from a process that previously took 20 minutes to maybe two hours and you're taking it down to minutes, maybe even seconds in some instances, correct?

Kapil Dev Bansal: Yeah, absolutely.

Christian Johnson: So what are some of the things that supply chain leaders need to do in order to enable this shift from focusing just on the AI technology to realize transformation across their systems?

Rahul Shahani: I think this uh, largely harkens back to many of the traditional levers in transformation. Right at the first, at the core of this is the what, what does the business led reimagination of uh, the operation look like? And this is clear aspirations tied to business outcomes. Right. What do I need to change to change the trajectory of my performance then? What are the domains that you're prioritizing? So a domain level transformation that focuses uh, on end to end workflow orchestration versus individual pilots. So taking a capability, putting in the infrastructure and then having a series of agents tackling the tasks, the goals and the orchestration of them to really transform the work. And lastly it's leaders who are owning the entire end to end learning journey of this process. So that's the what. In parallel to this you also need the how, which is this is where you start getting a little bit more agentix specific because here you need the core execution enablers which is the end to end process redesign that takes the capabilities of Agent based AI at its center, which includes the data products, the underlying tools and infrastructure. The idea around atomic agents, which are agents that are built to perform tasks that are modular and repeatable so that you can rapidly scale up the agents, a governance model to monitor and performance manage these agents and continuously train and evolve them the workforce. That means the workers who are playing new roles, which would be agent supervisors, as I like to call them, agentic tamers, right? They are overseeing a whole new set of capabilities and skills that we've never had to do before. And then lastly, it also requires responsible adoption of AI with guardrails and human in the loop controls as we build the accuracy of AI, the explainability of AI and the auditability of AI in our system.

Christian Johnson: I love the phrase agentic tamer. So tell me Raoul, what does agentic tamer mean for somebody like picture somebody like me working in supply chain somewhere? How do I become an agentic tamer?

Rahul Shahani: An agentic tamer really is a new skill set, a new activity, a new task that many of our workers will need to upskill themselves on, which is how do I manage a group of agents that are performing a task in my business? How do I become a leader of a series of agents? Which means I need to oversee the agents to make sure they're performing their work accurately, they are not seeing bias in the execution that they're doing. And if they are, how do I train the agent to be able to get back to a nominal correct, uh, outcome or a nominal expected outcome within the guardrails of execution? So an agentic tamer really becomes a new muscle that we all get stronger at in giving, say, how do I drive the right outcomes from the fleet of agents that I am taking accountability over?

Christian Johnson: I think that's a great summary of the what and the how for companies in thinking about AI across their supply chains. What's the best way now for companies to make the adoption feel achievable for them? Because what you've described so far, this is a lot to undertake, right? So how can this feel like this is something that companies can achieve realistically and get lasting value as opposed to a bunch of pilots that don't really end up meaning very much?

Rahul Shahani: It's a great question. The most important place is where to start and which really goes back to where the business value sits. And it's on domains that are trapped business value that are ripe for tackling through agentic AI. This largely means domains that are very information flow heavy as that is the sweet spot for agentic AI. So I would start with a high value information flow, heavy business domain. Uh, this could be quality, it could be procurement, it could be planning. It is where we drive a lot of information flow. Pair that domain with clear metrics for cost, service, resilience, agility. What are the outcomes we want to drive towards? Invest in data readiness. And interestingly, data readiness in an agentic world is a lot less rigid than data readiness we've seen in the past because agentic AI is capable of ingesting fuzzy data, uh, as well as monitoring and maintaining and improving data quality and performance, which is very different from what we've seen in the past. And lastly, it's about workforce enablement. Embedded from day one is how do I train and upskill my workforce to become agent handlers out the gate?

Christian Johnson: So I'd like to interject an additional question here. I love where we are in the conversation right now, but what I would love to give, and I'd like to pause to just give you guys, uh, a second to think about this is, is I'd like to be able to give the listeners an example of what it means to go from the kind of use case mentality through to the full orchestration here. And what I'd love to focus on is if there are any examples you can think of where clients were really surprised by where the value lay. They thought that initially it was going to be X. And that was essentially thinking from very much, uh, a use case basis and into really understanding where the value was.

Rahul Shahani: Let me paint the picture of an order coming in. Traditionally when an order comes in, many times clients think the value sits in saying, do I know what I have in stock? Where should I dispatch this from and where should I route this? And so the idea is, if I have this in the control tower or system, I'm able to answer that question very quickly and process the order. The reality is where this, where this process tends to break down is in the handovers between, uh, between organizations and functions. So when the order comes in, it is the person looking at the dashboard saying, I have 200 units in stock in warehouse A. What I'm going to confirm. So I'm going to call up the supervisor at warehouse A to confirm that they actually have that in stock. Then once the supervisor confirms, I'm going to see can I ship it. And now I'm not sure if the truck can come in and pick it up. So I'm going to call X vendor or send an email and wait for 24 hours for the response on whether I can fulfill this order effectively or not. The idea being agentic is it's built around the master data that allows us to answer a lot of these questions autonomously or automatically trigger that question to the right stakeholders and through that orchestration, speed up these cycles from hours or days to minutes. That unlock happens because what we're really tackling here is simplifying organizational friction that happens day to day and that speeds up everyday processes and cycles.

Kapil Dev Bansal: Just echoing Rahul's point, a lot of clients thought the value is in doing a lot of prediction, a lot of analysis. But clients realized that with all the bi, uh, all the technology, they already have a lot of insights and they just don't know which insights to work on, which ones to prioritize, which one to act on. And the surprise was that agents helps to prioritize the available insights and helps execute on that. That's where the real unlock is, because it is less about what can you know more about supply chain existing tools. Bi like all the tools provide enough insights. But then agentic is the execution layer, which is a real, uh, pleasant surprise for the clients who are doing this work.

Christian Johnson: What would you like to say to people out there? A lot of listeners on their call may be concerned about their employees. Some of the employees listening may be concern about their own futures, about what do they see in a world in which AI is now taking over some of the tasks that have taken up a lot of their days, day to day.

Rahul Shahani: When we think about agent take, what the role of the individual is evolving to is from a doer or controller of the task to a conductor or an overseer of the task. That means that it's allowing us to move a lot faster as individuals because we have the agents doing the execution activity, but it allows us to oversee, to say, are we making the right decisions? Is the agent performing reliably and repeatably in its activities and effectively moving ourselves into performance management of uh, these small atomic micro agents that are driving the execution of the tasks. That does mean that as end users or, or overseers of agents, we do need to build new skills, which is to think about how to program agents, how to tune agents and how to performance manage agents, all of which are relatively new skills in the workforce that we will rapidly have to pick up. Lastly, agents don't remove human judgment, they relocate it. Because in many cases humans are still in the loop. Auditing many processes, especially the more complex processes, is where the humans come in. What the agents are doing is allowing us to spend more of our time really being the expert versus being the executor. And that is really how we tend to upskill our teams into really so focusing on the more complex problems.

Christian Johnson: So what is some of the real value, the types of value that we are seeing agentic AI uh create?

Kapil Dev Bansal: We have seen agentic unlocking real value for the clients where we have done the transformations. It can reduce the cost of goods sold by 4 to 7%, it can improve the productivity by 20 to 50% and it can shrink the decision cycles from weeks to hours and in some cases even to seconds. Once fully implemented and adopted, agent TKI has the potential to fundamentally reshape the supply chain. It is the most disruptive technology that has been discovered in the last two decades when it comes to concerns the supply chain. But the value is not automatic. There has to be like serious work that has to be done in order to get to the value. The value is not theoretical, but it requires a lot of purposeful thinking. And how do we go and get that value?

Christian Johnson: So what we've been hearing today is that with agentic AI, supply chain won't just plan better, it will act continuously, autonomously and with human judgment where it matters most. If you want to stay ahead on how AI is transforming operations, head over to mckinsietalksoperations.com linked in our show notes to explore our upcoming events and see what's coming next. You've been listening to McKinsey Talks operations with me, Christian Johnson. If you like what you've heard, subscribe and stay tuned. Another great episode. Episode starts now.