Industry forecasts about agentic AI adoption in supply chain have converged on a similar shape: a small fraction of enterprises running agentic capabilities today, a large majority expected to by the end of the decade. The exact numbers vary by source, but the direction is not in dispute.

‍

What's less discussed is what actually needs to be true inside an enterprise for that jump to happen, and it has almost nothing to do with the AI itself.

‍

‍

The Wrong Question

"When will we adopt agentic AI in our supply chain?" is the wrong framing, because it implies adoption is primarily a technology decision, something that happens when a vendor ships the right feature or a budget gets approved. The more accurate framing is closer to: "What would our supply chain data and processes need to look like for an autonomous agent to act on them safely?"

‍

That second question exposes the real gap. Most supply chain functions today run on a mix of ERP data, spreadsheet-based planning overlays, email-based supplier communication, and tribal knowledge held by a handful of experienced planners. An AI agent asked to act autonomously in this environment isn't blocked by a lack of intelligence. It's blocked by a lack of trustworthy, structured, real-time data to act on.

‍

‍

From Recommends to Executes

There's a meaningful distinction between AI that recommends and AI that executes, and most supply chain AI deployed today sits on the recommends side of that line: a forecasting model suggests a reorder quantity, a planner reviews and approves it. That's valuable, but it's not agentic in the sense driving the adoption forecasts. Agentic means the system initiates the purchase order, reroutes the shipment, or renegotiates the delivery window without a human in the loop for every instance.

‍

Crossing that line requires answering a question most supply chain organisations haven't fully answered internally: who is accountable when the agent is wrong? Not philosophically, operationally. If an agent reroutes a shipment based on a disruption forecast and the rerouting turns out to be unnecessary and costly, is that a system failure, a data failure, or an acceptable cost of operating at the speed autonomous action requires? Enterprises that haven't answered this before deploying tend to answer it defensively after the first expensive mistake, usually by rolling back autonomy rather than fixing the underlying accountability model.

‍

‍

The Data Foundation Problem

Underneath the governance question is a more mundane one: data quality. An agent making autonomous decisions about inventory, routing, or supplier switching needs consistent, current data about supplier lead times, in-transit inventory, demand signals, and cost. In a lot of enterprises, this data exists but is scattered across systems that don't reconcile with each other in real time. A planner navigating this manually can apply judgment to paper over the inconsistencies. An agent, by design, does exactly what the data tells it, which means inconsistent data doesn't get quietly corrected. It gets acted on.

‍

This is why the enterprises furthest along on agentic supply chain adoption didn't start with the agent. They started by fixing the data foundation the agent would need, often over a year or more before the agent was ever deployed, because they'd learned that skipping this step just moves the failure downstream and makes it more expensive.

‍

‍

What the Adoption Curve Actually Measures

Read this way, the adoption forecast isn't really measuring AI capability, which has arguably been ready for autonomous supply chain action for some time. It's measuring how long it takes enterprises to do the unglamorous work of data integration, process redesign, and governance definition that has to happen before autonomy is safe to grant.

‍

That reframing matters for anyone planning around these numbers. The distance between low and high adoption isn't going to close because the technology gets better in the interim, though it will. It closes because enterprises that start the foundational work now will be well ahead of that curve in a few years, and the ones waiting for the technology to be "ready" will find that readiness was never really the constraint.

‍

‍

The Diagnostic

Before asking whether your supply chain is ready for agentic AI, it's worth asking whether your supply chain data is trustworthy enough for a human to act on autonomously, let alone a machine. If a planner would still double-check a system-generated recommendation before acting on it today, an agent replacing that planner isn't removing a bottleneck. It's removing the check.

‍

‍

AI That Pays is published by Avaali — an enterprise AI and intelligent operations firm working with large enterprises across Asia, the Middle East, and Europe on Source-to-Pay transformation, Intelligent Finance Operations, and AI-enabled procurement.