Why Most Companies Still Aren’t Winning with AI

Welcome to the Future Chain Podcast. If you are a supply chain procurement or maybe a logistics executive right now, well, you're living through quite a moment, aren't you? Just unprecedented technological hype.

Oh, absolutely. It feels like every conference, every vendor pitch, you know, even board updates. Yeah, it's all AI, AI, AI. The promise that it's going to fundamentally transform everything.

It is. And look, Some of that excitement is definitely warranted. There's real potential there.

There is. But the challenge we often hit is separating the actual on-the-ground deployment from, well, let's call it wishful thinking.

Right.

So our mission today is really to try and cut through some of that noise. We've been looking at the latest research, specifically the McKinsey State of AI in 2025 survey, to answer one core practical question.

Where does—

When you look at what's actually happening, where does supply chain management, SCM, stand right now in this global AI race.

Okay. And the initial data, the big picture, tells us the ambition is sky-high, right?

It really is. Nearly 9 out of 10 organizations, that's 88% globally, say they're using AI regularly in at least one business function.

So, AI isn't optional anymore. It's, well, it's becoming pretty widespread.

That visibility is key, definitely. But here's the catch, and this is the bit every executive trying to scale AI needs to focus on.

Okay.

Despite this widespread initial use, most organizations are frankly stuck. Nearly 2/3 globally haven't actually begun scaling AI across the enterprise.

2/3? Wow.

Yeah, they're still firmly locked in that experimentation phase, or maybe piloting. The survey calls it the stubborn growing pain, which I think is quite apt.

Stubborn growing pain. I like that. It means we know the test case works, maybe in one warehouse or one process, but we just can't seem to integrate it everywhere.

Precisely.

Okay, let's unpack this specifically for our audience. Let's focus on SCM and inventory management. How far along is our function in that initial adoption compared to, say, other parts of the business?

Well, perhaps predictably, functions like IT and knowledge management are leading the pack. A lot of that's driven by GenAI use cases.

Makes sense.

But if we pull out SCM and inventory management specifically, 17% of all the respondents report regular AI use in this function.

17%. So where does that put us? Middle of the road?

Yeah, squarely in the middle, I'd say. It shows consistent, steady adoption. We're not, you know, leading the absolute charge, but we're definitely not lagging behind either.

Hmm, 17% adoption. That feels like a decent baseline, probably for things like predictive analytics, maybe some basic demand forecasting tools.

Exactly, those foundational applications.

But the future? I mean, the really dynamic stuff in supply chain automation, complex inventory optimization across multiple sites, or automated logistics routing that seems tied to something more sophisticated— AI agents.

Ah, yes.

These systems that can actually plan and execute multiple steps, kind of run a process end-to-end autonomously. What's the appetite for those? Is there interest?

Oh, the curiosity is palpable. You can really feel it in the data. We're seeing a massive spike in interest. 52% of organizations overall are experimenting with AI agents in some capacity.

62% experimenting. Okay.

People get the potential. They see this is a huge step up from just using, say, a single AI model for one task like forecasting. This is about true complex automation.

But, and here's the crucial question for us. If we translate that general curiosity specifically to the supply chain, does the reality match that ambition? Are we seeing agents deployed in SCM?

Unfortunately, no, not really. This is what we're calling the SCM reality gap, something we really need to address.

Okay.

When you look specifically at scaling AI agents in supply chain and inventory management, meaning they're fully deployed maybe across multiple business units or the whole enterprise, Hmm. The number is, well, it's shockingly small.

How small?

Only 5%. Just 5% of organizations report they are scaling or have fully scaled AI agents in SEM.

5%. Wow. So what are the others doing? That's 62% experimenting.

Well, a big chunk of those experimenting are in SEM, but they're stuck. 39% combined are still just experimenting or piloting agents within SEM. They haven't made the leap.

5% scaling. Okay, that really hammers home that stubborn growing pain point that reveals this massive gap between You know, testing the waters and actually diving in deep.

Exactly.

It's like you said, having one fancy self-driving forklift in one corner of one warehouse, it proves the concept works, sure, but it doesn't run the enterprise, does it?

That's a perfect analogy. It really is. But okay, even if we're mostly stuck in pilot mode, the question then becomes, are these initial SCM AI uses actually delivering value, real tangible value?

Yes, they absolutely are. And the returns seem quite immediate, primarily focused on efficiency, which makes sense for SCM.

Right.

If you look at cost benefits, often the first focus for procurement and logistics departments, 51% of respondents using AI in SCM report seeing some level of cost decrease from those activities.

51%. That's significant.

It is. It positions SCM quite high in efficiency gains, actually right alongside manufacturing operations, which reported 56%.

That makes perfect sense. AI can instantly optimize that obvious low-hanging fruit, can't it? Things like transport routes, consolidating loads, maybe fine-tuning inventory buffers to cut holding costs.

Exactly. So, it's a proven cost-cutter in the supply chain, even in these early stages.

Okay, but is it only a cost-cutter? For executives thinking about market share, about growth, what about the top line? Is it helping generate revenue?

Yes, revenue benefits are also showing up, which is interesting. It demonstrates that SCM AI can add value beyond just saving money.

How much?

57% of those using AI in SCM report seeing revenue increases.

57%. Okay, so slightly more are seeing revenue bumps than cost cuts.

It suggests AI is driving value through maybe improved product availability, perhaps better forecasting accuracy, leading to fewer lost sales, maybe even faster fulfillment times. Hmm.

Interesting. But I imagine that revenue impact from SCM is still probably overshadowed by functions closer to the customer, like marketing and sales?

It is, yes. The overall revenue impact reported from SCM activities is still, let's say, limited compared to marketing and sales. They report substantially higher revenue increases. They've just got a more direct line to the customer's wallet, I suppose.

Right. So, the takeaway there is, for SCM to deliver that truly transformative enterprise-level value, we can't just keep chipping away at costs and hoping for a few extra percentage points on revenue. We need something bigger.

Exactly. We need transformation. And this is where it gets really critical, especially for you, the executive listening, thinking about that bigger enterprise transformation. If efficiency is just the low-hanging fruit, how does an organization actually make the jump? How do you get from a successful pilot saving 5% in one DC to, you know, enterprise-level value?

I think we have to stop looking at the average organization for the answer. We need to study the outliers, what the survey calls the AI high performers.

The AI high performers. Okay, who are they?

This is the elite group, about 6% of companies in the survey, who are seeing significant value, And this is key— they can attribute 5% or more of their EBIT, their earnings before interest and taxes, directly to AI use.

5% or more of EBIT from AI? That's serious impact.

It is. They are the blueprint for how transformation is actually done.

So, what's the fundamental difference in their approach? Is it just that they're throwing more money at AI? More capital?

It's not just about capital, no. It seems to be more about a difference in their Strategic ambition, their goals.

How so?

When you look at the objectives these high performers set for AI, they simply refuse to stop at efficiency. I mean, 84% of them do aim for efficiency. That's pretty standard, similar to other companies at 82%.

Okay.

But the real differentiator is that these high performers also prioritize aggressive growth. 80% of them target growth versus only 50% for everyone else. And innovation, 79% target innovation, again, versus only 50% for the rest.

Ah, I see. So they're not just seeing AI as like a back-office optimization tool. They view it as an engine, an engine for creating entirely new supply chain capabilities that can actually drive market growth.

Precisely. They're using AI as a pathway to fundamentally new ways of operating, maybe even new business models. And that ambitious goal, it translates directly back to that AI agent question we were discussing for supply chain management.

How do you mean?

Well, if your goal is genuine innovation and achieving that 5%+ EBIT impact, you simply must be scaling those more transformative agentic systems. You can't get there just by tweaking existing processes.

Okay, right. Let's bring it back then to that frankly quite staggering 5% SCM agent scaling rate we heard earlier for the average company. How much faster are these high performers moving on deploying agents in the supply chain?

Okay, this, this is the key statistic. I think every logistics and supply chain executive really needs to internalize this number. When we look at the scaling of AI agents in SCM among this elite high-performer group, 31% of them are scaling AI agents in supply chain and inventory management.

31% compared to just 5% for everyone else.

31 versus 5. Yes.

Wait, that difference is, it's immense. That's more than 6 times higher. Is that really right?

It is absolutely right. That gap represents a more than 6-fold acceleration, a 6x faster rate of scaling agents in SCM. Among the most successful, most profitable companies using AI.

6 times faster.

So the message is pretty clear. If your organization's goal is to move beyond just pilot success and actually achieve that significant 5% EBIT impact level, the data strongly suggests you need to be 6 times more aggressive, 6 times faster in embedding automated agents into your core supply chain workflows than your average peer is today.

That That really defines urgency, doesn't it? Okay, so what enables that 6x acceleration? It can't just be flipping a switch and hoping the pilot magically works everywhere. What are they doing differently?

It seems to boil down largely to 2 things: organizational commitment and, crucially, fundamental redesign. And this redesign part is where most organizations seem to fall down.

Redesign?

Yes. High performers are nearly 3 times, 2.8 times to be precise, as likely to fundamentally redesign their existing workflows when they deploy AI.

2.8 times more likely.

They aren't just trying to automate a broken process or an old, manually intensive way of doing things. They're rebuilding the process around the AI's capabilities.

Okay, can you give us maybe a concrete SCM example? What does that fundamental redesign look like compared to just basic automation?

Sure. Think about, say, purchase order approvals. A non-high performer, an average company, might try to use AI simply to speed up the old, Maybe paper-based or multi-step email approval process, shaving off a few minutes, maybe an hour.

Yeah.

That's efficiency.

Right. Automating the existing steps.

Exactly. A high performer, however, uses the AI agent to fundamentally redesign that flow. Maybe they eliminate the need for manual approval entirely for certain categories. The AI agent might initiate the PO based on dynamic demand signals it's predicting. It could optimize the supplier selection based on real-time factors like commodity prices, lead times, transport availability, and then execute the entire transaction autonomously, perhaps only flagging exceptions for human review.

I see. So, it's not automating the old steps; it's creating a whole new agent-driven process.

Precisely. That's workflow redesign driven by the agent, and that's a big reason why they achieved that step change in impact, that EBIT difference.

And presumably, none of that kind of radical redesign happens without the executive suite, the leadership, providing air cover for that level of operational upheaval. It sounds disruptive.

Exactly right. It absolutely requires visible, active commitment from the top. The data links high performance strongly to leadership.

How strongly?

High performers are 3 times more likely to strongly agree that their senior leaders demonstrate true ownership and commitment to AI initiatives.

3 times more likely?

Yes. This isn't just signing off on a budget. It means really championing AI across the organization, securing the resources needed for those complex workflow redesigns we just talked about, and importantly, role modeling its use themselves.

Okay, so let's address the elephant in the room then— the human cost of this acceleration. If we're talking about a 6x speed increase in scaling autonomous agentic AI, what does this realistically mean for the SCM workforce? Are supply chain teams facing massive headcount reductions?

Well, the immediate impact looking back over the past year is still somewhat mixed, actually.

Oh.

Looking specifically at supply chain and inventory management headcount changes in the last year, 17% of respondents said they observed a decrease in workforce size due to AI, but 12% actually observed an increase.

Okay, so a net contraction overall, 17 versus 12, but not, not a dramatic cliff edge yet.

Not yet. It's been more of a gradual shift so far.

What about the expectation for the next year? Does that suggest the contraction is going to accelerate?

Significantly, yes. Yeah, that's where the outlook changes quite a bit. Looking ahead to the next 12 months, 32% of respondents expect a decrease in SEM headcount due to AI.

32% expect a decrease.

While only 11% expect an increase.

Okay, so the gap widens considerably from a net negative 5% observed last year to a net negative 21% expected next year.

Exactly. This confirms that executives, certainly in the near term, primarily see AI as a tool for efficiency and rationalization in SCM, and that is projected to lead to a steeper net contraction in traditional roles.

Hmm. And that expected contraction, it obviously necessitates a crucial talent trade-off, doesn't it? You might eliminate certain manual roles, but you absolutely need new skills to build, manage, and oversee the AI itself. Where are organizations focusing their hiring efforts to support these agents?

The focus is overwhelmingly on the technical side, building the infrastructure and ensuring reliability. Organizations are hiring for AI-specific roles, with software engineers and data engineers being the most in demand right now.

Makes sense.

And crucially, the survey highlights that AI success requires robust data readiness, having clean, accessible data, And also MLOps expertise.

Okay. For our executive listeners who might not be deep in the weeds of software engineering, what exactly is MLOps? And why is that expertise suddenly so critical for the supply chain?

Right. Good question. MLOps basically stands for Machine Learning Operations. Think of it like DevOps, but specifically for AI models.

Okay.

Simply put, these are the engineers and the processes needed to industrialize the AI code. They build the operational backbone that ensures the AI model running your logistics planning or your procurement system doesn't just break down at 3 AM when market conditions suddenly shift.

Ah, so it's about reliability and maintenance at scale.

Exactly. It's about creating enterprise-grade monitoring, management, updates, and overall reliability for the AI systems themselves. Without strong MLOps capabilities, that 31% scaling rate we saw among high performers, it's basically impossible to achieve safely and sustainably.

Right. And given that these agents are, by definition, planning and executing tasks autonomously, that introduces significant risk, surely. How are these high performers mitigating the chance of an AI making a really expensive mistake in, say, procurement or shipping?

Yeah, that's a critical point. And this is where the human element remains not just important but absolutely required, especially for high-risk decisions.

So, humans aren't completely out of the picture?

Not at all. High performers clearly understand the risks that come with rapid scaling of autonomous systems. One of the highest prevalence practices among that top 6% is having very clearly defined processes to determine how and when the outputs from AI models need human validation or oversight.

The human in the loop.

Precisely. They've established a robust human-in-the-loop imperative. It's essential to ensure accuracy, maintain compliance, and frankly, prevent potentially catastrophic mistakes.

And are organizations generally getting better at managing AI risks overall beyond just the high performers?

Yes, it seems so. The data suggests organizations are becoming more sophisticated about risk mitigation across the board. Overall, they're actively mitigating about 4 different types of AI risks on average today.

And how does that compare to before?

Well, that's up from just 2 risks on average back in 2022. So awareness and action have doubled in just a couple of years.

Interesting. What's the biggest risk people are worried about?

Inaccuracy, basically. The AI just being wrong remains the most commonly reported negative consequence from using AI, cited by 30%. And perhaps reassuringly, it's also the risk that the most organizations, 54%, are actively working to mitigate.

Accuracy is paramount, especially in SCM where errors can be costly. Okay, that brings us towards wrapping up. If you are a supply chain, procurement, or logistics executive listening to this, And you're grappling with your AI strategy right now, what are the absolute non-negotiable takeaways? Maybe the top 3 things you need to be thinking about tomorrow morning.

Okay. 3 things. First, I'd say acknowledge the current reality. While AI is delivering clear incremental efficiency gains, remember that 51% reporting cost decreases in SCM— true— transformative enterprise-level value is still trapped in the piloting phase for most organizations. That 5% agent scaling rate in SCM is stark proof of that stagnation.

Okay, acknowledge the gap between pilots and real scaling. What's second?

Second, the blueprint for breaking out of that stagnation is clear. You need to look to the AI high-performers, emulate their approach, and that means setting much bolder goals. Prioritize innovation and growth, not just cost efficiency, as the primary objectives for your SCM AI projects. Aim higher.

Right, think bigger than just cost savings. And third?

And third, that high ambition has to be matched by a real willingness to act decisively on the data. That means aggressively scale AI agents where they offer transformative potential, focusing hard on building the necessary infrastructure and envelopes capabilities for reliability, and crucially, embrace that 2.8x higher likelihood of needing to fundamentally redesign your existing procurement logistics workflows from the ground up, integrating the AI properly rather than just layering it on top.

So it comes back to that redesign point. Be prepared for real change. Okay, let me try and synthesize that into a final thought for our listeners.

Go for it.

If achieving that significant greater than 5% EBIT impact requires your organization to be, as the data shows, 6 times more aggressive, 6 times faster in scaling AI agents within your supply chain than your peers are today, the question for you is simply, how quickly does your current pilot need to transition from just experimentation into irreversible transformative action? It feels like the time for just testing might be over.

I think that sums it up perfectly. The window for cautious experimentation might be closing for those who want to lead. We look forward to seeing how the future chain evolves over the next year based on these trends.

Indeed. That's it for this look at the state of AI in supply chain management based on the latest McKinsey data. We'll talk to you next time.