AI Can See Supply Chain Disruptions. The Next Challenge Is Letting It Act.

 

Supply chain disruptions cost businesses an estimated $184 billion in 2025, according to the J.S. Held Global Risk Report.

Yet much of the technology investment meant to solve the problem still focuses on one thing: detecting disruption faster.

That is useful, but detection alone does not move cargo, change suppliers, reroute shipments, or protect margins.

Over the past decade, companies have invested heavily in AI-powered visibility platforms, control towers, digital twins, risk scores, forecasting tools, and exception dashboards. These systems have dramatically reduced the time between a disruption happening and a company becoming aware of it.

The bigger problem now is what happens next.

A system may detect a delayed vessel, a supplier outage, an inventory shortage, or a transportation bottleneck within minutes. But action can still depend on someone opening a ticket, joining a meeting, seeking approval, and manually entering the same information into several different systems.

In other words, AI can often see the problem before humans do, but it still has to wait for permission to respond.

Detection is no longer the biggest challenge

Supply chain AI already performs well in areas such as demand forecasting, ETA prediction, inventory optimisation, supplier risk monitoring, and transportation analytics.

Companies can identify a delayed shipment before it misses a delivery window. They can detect supplier risks before customers complain. They can model inventory shortages before shelves go empty.

But the real financial impact often happens after the alert.

Should the company expedite the shipment or wait?

Should it split the order?

Should it switch carriers?

Should it pay a higher spot rate?

Should two partially filled shipments be consolidated?

Should selected high-value products move from ocean freight to air?

These are not always massive strategic decisions. In many cases, they are routine operational decisions governed by rules the company already understands.

Yet they still frequently sit inside a human approval queue.

A 2026 survey by Knosc found that supply chain teams at mid-market manufacturers and distributors spend around 28 percent of their working time responding to disruptions, much of it investigating problems rather than executing solutions.

Meanwhile, AI remains a major priority among logistics leaders. Capgemini reported in 2025 that an AI-driven next-generation supply chain ranked among the top technology trends for 70 percent of large-company executives.

But Gartner found that only 23 percent of supply chain organisations had a formal AI strategy in 2025.

The issue may no longer be the quality of the AI.

It may be how much authority companies are willing to give it.

The problem with the "AI recommendation"

Many supply chain systems follow the same workflow.

AI detects an issue.

It generates a recommendation.

The recommendation becomes an alert.

The alert becomes a ticket.

The ticket waits for a human.

And while everyone waits, the business opportunity disappears.

The alternative carrier may lose available capacity.

The consolidation window may close.

The supplier may allocate its next production slot to another customer.

The system predicted the disruption correctly, but the company still loses money because execution remained slow.

According to research from FourKites and ABI Research in 2025, only 27 percent of organisations allow AI to take autonomous action, while 52 percent restrict AI primarily to decision support.

That distinction matters.

Another dashboard may improve visibility, but visibility alone does not necessarily improve EBITDA.

The next stage of supply chain AI is not simply better prediction.

It is controlled execution.

From AI insights to bounded AI action

The emerging model is what could be called bounded autonomy.

Instead of allowing AI unrestricted control over supply chain decisions, companies define clear rules that allow AI agents to act within specific limits.

For example:

An AI agent could automatically retender a shipping lane if the contracted carrier's ETA exceeds a defined delay threshold and an approved alternative carrier remains within an authorised price range.

It could consolidate outbound shipments when capacity, delivery schedules, and transportation costs make consolidation more efficient.

It could switch selected high-priority products from ocean freight to air when the cost of missing a retail window is greater than the additional freight cost.

It could reallocate safety stock between distribution centres when inventory shortages and transportation constraints appear simultaneously.

These actions do not require AI to become the CEO of the supply chain.

They require companies to define the rules.

The model becomes simple:

If these conditions occur, take this action, within this financial limit, using approved suppliers or carriers, while recording every decision. Escalate only when the situation falls outside the authorised boundaries.

This is not a fully autonomous "lights-out" supply chain.

It is controlled autonomy.

Three things need to change

First, companies need to turn operational knowledge into formal policies.

Rules that currently live inside the heads of experienced planners must become machine-readable decision frameworks.

For example:

If an ocean shipment containing critical inventory is delayed more than 48 hours, and the expected business loss exceeds the cost of air freight, AI may automatically upgrade transportation for approved SKUs within a defined budget.

Second, supply chain systems must allow AI agents to execute transactions.

An AI agent that can recommend a carrier but cannot submit a tender is still just an analytics tool.

Transportation management systems, warehouse management systems, procurement platforms, sourcing tools, and carrier APIs will increasingly need to support authenticated, logged, reversible machine-initiated transactions.

Think of the AI agent as a junior buyer with a strict spending limit.

Third, accountability must evolve.

If an AI agent makes a decision within an approved company policy and the result is unsuccessful, organisations will need to examine the policy, data, rules, and system design rather than simply asking which employee should have double-checked the machine.

Without that cultural change, AI agents will continue to wait for human approval.

And waiting defeats much of the purpose of automation.

The next competitive advantage

For the next few years, many supply chain platforms may look similar.

Almost everyone will claim to have AI.

Almost everyone will have predictive analytics.

Almost everyone will have a control tower.

The real difference may be measured in one metric:

How long does it take to move from detection to action?

One company may receive an AI warning that a major shipment will be delayed.

Another company may receive the same warning and immediately allow an authorised AI agent to retender the lane, consolidate inventory, reroute high-priority goods, and protect the delivery window.

Both companies have AI.

Only one has given AI the authority to act.

Supply chain disruption will not disappear. Geopolitical risk, transportation volatility, component shortages, supplier concentration, extreme weather, and multi-tier visibility problems will remain part of global commerce.

But companies can decide how quickly they respond.

The first generation of supply chain AI gave businesses better visibility.

The next generation will give AI carefully controlled authority.

And the companies that master that transition may discover that the biggest value of AI in logistics is not predicting what will happen next.

It is being allowed to do something about it.