Ask any logistics or supply chain leader whether artificial intelligence can improve forecasting, routing or fulfillment, and you'll almost always get a yes. But if you ask whether their organization is actually ready to run AI at scale, confidence drops off.

That gap is about the foundation underneath the technology. And the challenge has only become more visible with agentic AI. For an agentic system to be able to optimize a shipment lane or catch an exception before it turns into a delay, it needs trusted data, consistent workflows and clearly defined business rules to act on. In many supply chain organizations today, that groundwork just isn't there yet. It's simply the legacy of how these networks were built in the first place.

Most supply chain leaders already feel this weight building. Systems got added region by region, asset by asset, usually to solve whatever problem was in front of someone at the time rather than to fit into some larger, coherent operating model. Shipment visibility lives in one platform, asset tracking in another, supplier data somewhere else entirely, and exceptions get worked out by hand because that's faster than fixing what's actually broken. None of this is new. What's changed is the cost of leaving it alone.

New research from Genpact and HFS Research puts a number behind that cost. The study surveyed more than 2,000 senior executives across 16 industries and 14 business functions and found that logistics and supply chain organizations are carrying more "enterprise debt" than the cross-industry average — the buildup of fragmented systems, siloed regional operations, and manual, exception-heavy workflows. For a sector that runs on coordinating hundreds of handoffs across a global network, it’s almost the expected outcome.

What is alarming, though, is how much of the sector's AI upside is already being constrained by that debt. Logistics and supply chain organizations are currently leaving a large share of value on the table, and it stays out of reach when teams choose status quo instead of the hard, unglamorous work at the foundations. While deploying a new model becomes news, redesigning a core supply chain process is a thankless, invisible grind.

It's tempting to read this as an argument for pumping the brakes on AI but, if anything, the data points the other way. Many organizations are layering AI on top of infrastructure that was never built to carry it, and then wondering why the results are inconsistent. An AI model trained on fragmented shipment data is likely to make inconsistent decisions. An agent dropped into a poorly designed exception-handling process can accelerate existing problems faster than a person would.

So the real question is whether the network can support AI once it's running. Ask that honestly, and you tend to land on the same handful of gaps every time: data that isn’t reconciled across systems, workflows that live in someone's head instead of a defined process, and regional operations that don’t act as part of the same network.

Where Supply Chain Leaders Should Start

None of this gets solved overnight, and trying to fix everything at once usually means nothing gets fully fixed at all. A more useful starting point is narrower than that.

Unify the data AI depends on. Shipment, asset, and supplier information that was never designed to be unified needs to start being so. That means deciding what "one version of the truth" looks like, and building toward it on purpose, rather than hoping it emerges on its own.

Fix the exception-heavy workflows before you add AI on top of them. If a process today depends on someone catching and correcting problems by hand, bolting AI onto that process usually just changes who — or what — is doing the correcting. It doesn't solve the actual issue. Workflows need to be redesigned so that humans and agents are each doing what they are best suited for: agents handling volume and routine exceptions, humans validating and making judgment calls on the hardest cases. Every transaction processed and every exception resolved makes the system smarter over time, so fixing the foundation is what enables intelligence to compound.

Be honest about where the debt is worst. For most logistics organizations, that's process and technology, which are not necessarily the areas getting the most attention or budget right now. Putting the first dollars where the drag is heaviest tends to produce results faster than spreading them evenly. Faster, visible results make it a lot easier to justify the next round of investment.

Supply chain leaders don't need convincing that AI has real potential. What's less understood is how much of that potential is already limited by decisions made years ago, from systems stitched together regionally to workflows built for a different scale of operation entirely. Fixing that isn't something you have to finish before AI can add any value. But ignoring it means accepting a ceiling on what AI can deliver, no matter how good the model is.

The organizations that pull ahead will be the ones that took an honest look at what their supply chain networks are built on — and did something about it.

Mantej Singh Randhawa is Global Growth Leader – Supply Chain at Genpact.