AI Won't Fix a Broken Supply Chain Process
Updated: Sep 9
AI can help a supply chain move faster. That is true! But there are limitations!
If the process is fragmented, the data is late, and teams do not trust the signals in front of them, AI may only accelerate the confusion. It can make bad handoffs faster. It can surface more exceptions than people can use. It can automate work that should have been redesigned first.
I recently read IBM’s The New Supply Chain Equation, and one point stayed with me: the supply chain operating model that worked in the past will not carry us into the future.
I agree with that. The old model is under pressure from demand volatility, supplier risk, transportation constraints, labor gaps, and rising customer expectations. Supply chains need better tools. They need better sensing. They need better decision support.
But the fundamentals still matter.
Before deploying AI, leaders need to understand how work actually flows. Where does information stop? Where are people re-entering data? Which decisions require judgment? Where are teams waiting on one another? Where is the organization carrying extra inventory because no one trusts the process?
Those questions may sound basic. They are also where many transformations are won or lost.
Start with the flow of work
Many AI conversations begin with the wrong question.
The question is usually some version of, “Where can we add AI?”
That question sounds practical, but it often leads to point solutions. A planning assistant here. A chatbot there. A forecasting tool somewhere else. Each one may help, but none of them will fix a process that is broken across functions.
A better first question is:
How should this work flow from beginning to end?
That question forces a different conversation. It moves the focus away from technology and toward the operating model. It asks how demand becomes purchasing intent, how purchasing intent becomes supplier action, how supplier action becomes logistics planning, and how logistics planning becomes customer service.
In many supply chains, that flow is not clear. Customer demand lives in one system. Supplier production updates arrive by email. Freight planning happens in a spreadsheet. Inventory policies sit in a planning tool that only a few people trust. Finance sees the cost impact after decisions have already been made.
People then build workarounds. They copy and paste data. They hold extra safety stock. They call suppliers directly. They build private trackers. They create side processes that help them get through the week but make the system harder to manage.
That is the real problem AI has to face.
AI can summarize, predict, classify, and recommend. It can read unstructured messages. It can find unusual patterns. It can help people compare options. But if the flow of work is unclear, AI will struggle to know what matters. Worse, it may support the informal process instead of improving the real one.
The foundation has to come first:
Clear process ownership
Agreed decision rights
Shared definitions of demand, supply, capacity, and risk
Clean handoffs between functions
Connected data at the points where decisions happen
A practical way to separate routine work from judgment-heavy work
That foundation is not glamorous. It does not grab headlines the way AI does. But without it, AI and supply chain investments often produce activity without transformation.
AI should not automate confusion
A broken process usually reveals itself through friction.
Someone enters the same data twice. A planner waits for a supplier update that should already be visible. A buyer expedites an order because they did not see the demand change soon enough. A warehouse team receives product earlier than expected, but transportation does not know the freight plan changed. Finance questions inventory levels after the working capital is already tied up.
These are not just technology issues. They are workflow issues.
If AI is layered on top too early, it may help people move through those points of friction faster. That can be useful in the short term, but it is not the same as changing how the system works.
For example, an AI agent may draft an email to a supplier faster than a buyer can. But if supplier commitments are still handled through scattered email threads, the organization has not solved the visibility problem. It has simply made the email process faster.
A planning model may identify demand anomalies. But if no one has defined who reviews them, what threshold triggers action, and how decisions flow into supply and inventory plans, the model may create noise.
A procurement tool may classify transactions. But if approval paths are unclear, catalog data is weak, and buyers still need to chase context manually, the process will still feel heavy.
The useful test is simple. Before adding AI, ask what will change about the work itself.
Will the handoff disappear?
Will the decision happen earlier?
Will the person doing the work have better context?
Will the team stop maintaining a shadow spreadsheet?
Will the system reduce the need for extra inventory, extra touches, or extra approvals?
If the answer is no, AI may still have value, but the work needs more design.
A better process made the technology useful
I have seen the difference process discipline makes.
At one distribution company, customer demand, supplier production, and freight planning operated in separate silos. Each group worked hard, but the work did not move cleanly across the system. Demand signals did not translate into supplier communication fast enough. Supplier production updates did not connect cleanly to freight planning. Transportation decisions often happened with incomplete context.
The team did not start by asking where to add AI.
They started by redesigning the workflow.
Kanban principles helped make purchasing intent clearer. The process defined how demand signals should trigger replenishment. It clarified how that intent would be communicated to suppliers. It also connected suppliers and logistics partners to real-time demand signals, so teams could respond to the same information instead of reconciling different versions after the fact.
Once the workflow was clear, technology had a much better job to do. It did not need to compensate for ambiguity. It could support a process that people understood and trusted.
The result was a 22% decline in lead times.
That improvement did not come from a magic algorithm. It came from making the work visible, reducing delay between functions, and giving partners a clearer signal earlier.
This is one of the most overlooked lessons in supply chain transformation. The best technology usually performs well when the process around it is disciplined. The same technology performs poorly when the process is fragmented.
AI is no different.
Planning agents work best when planners remain in charge
Another planning transformation followed the same pattern.
The team first strengthened the demand and supply planning fundamentals. That meant clarifying planning horizons, exception logic, data inputs, inventory rules, review rhythms, and decision rights. Only after that foundation was in place did the organization deploy Claude-based agents.
The agents helped planners in several practical ways:
Identify unusual demand patterns
Prioritize inventory exceptions
Evaluate scenarios
Prepare decision-focused briefings
Bring relevant context together faster
They did not make decisions for the planners. That distinction matters.
Planners still brought the judgment. They understood customer history, supplier behavior, service requirements, commercial priorities, and operational constraints. The agents helped them spend less time gathering and formatting information, and more time investigating what mattered.
That is where AI can be most useful.
It can reduce the burden of repetitive work. It can scan larger volumes of information than a person can handle manually. It can highlight changes that deserve attention. It can compare options and prepare a concise briefing.
But it should not remove accountability from the people who understand the business. In supply chain planning, many decisions require context that is not fully captured in the data. A customer may accept a substitution. A supplier may be reliable despite a delayed update. A promotion may be more important than the baseline forecast suggests. A late order may matter more because of the account relationship behind it.
AI can support those decisions, but it should not pretend they are purely mathematical.
In this transformation, the broader change released $12.7 million in working capital while maintaining a 98% fill rate and reducing excess inventory by 59%.
Those results came from combining process discipline, connected data, and human judgment with AI support. The agents created capacity. The planners used that capacity to make better decisions.
That is the kind of outcome leaders should look for when evaluating ai in supply chain examples. The question is not whether AI appeared in the process. The question is whether it helped people make better decisions inside a process that had already been designed well.
Procurement gains come from redesign before routing
The same thinking applies to procure-to-pay.
Procurement teams often face high transaction volume, unclear routing, inconsistent purchasing behavior, and manual effort across approvals, sourcing, and invoice resolution. AI can help classify requests or route transactions, but it cannot fix an approval model that no one follows. It cannot repair weak category strategies. It cannot make competitive sourcing effective if the process lacks discipline.
In one procure-to-pay transformation, the team began by mapping the process.
That map showed where requests entered the system, how approvals moved, where buyers had to intervene, which categories were candidates for competitive events, and which transactions needed a better routing path.
The team then strengthened the Coupa workflow, expanded the use of reverse auctions, and introduced an AI agent to help route incoming purchasing transactions.
The order mattered.
The workflow came first. The sourcing approach became clearer. The AI agent then had a useful role. It helped manage inbound purchasing activity and direct work to the right path faster.
The results were meaningful. Competitive sourcing reduced costs by 28%, while the redesigned workflow increased transaction capacity by 40%.
Again, AI helped. But the gains did not come from AI alone. They came from redesigning how the work should move, then applying technology where it could reduce manual effort and delay.
This is especially important in procurement because too many processes grow around exceptions. Someone needs a rush buy. Someone bypasses a catalog. Someone asks for approval after the fact. Someone uses a preferred supplier without enough competitive pressure. Over time, the exception path becomes the real process.
AI can help find and route those exceptions. But leaders still need to decide which exceptions should remain, which should be prevented, and which should be turned into standard paths.
The right role for AI in supply chain work
AI belongs in supply chain transformation. The opportunity is real. But its role needs to be clear.
AI is strongest when it helps with work such as:
Reading and classifying large volumes of unstructured information
Detecting unusual demand, supply, or inventory patterns
Comparing scenarios quickly
Summarizing risk and options for decision-makers
Routing transactions based on defined rules and context
Preparing briefings that help humans act faster
Monitoring signals that would be difficult to track manually
AI is weaker when leaders ask it to compensate for poor process design.
It cannot create trust in a planning number if the inputs are unclear. It cannot make supplier collaboration work if suppliers are not connected to the right signals. It cannot reduce excess inventory if no one has agreed how service levels, risk, and working capital should be balanced. It cannot improve decision quality if decision rights are vague.
That means the transformation sequence matters.
Start with the work. Map the flow from demand to supply, from request to purchase order, from forecast to inventory action, from exception to decision. Study the handoffs. Find the waiting. Look for rework. Identify manual data entry. Ask where people use judgment and where they are simply moving information from one place to another.
Then improve the process. Remove steps that do not add value. Clarify ownership. Define triggers. Build standard paths for routine work and clear escalation paths for exceptions. Connect the data needed at each decision point.
Only then should the AI question come back into focus.
Now ask where AI can help people move through the better process with more speed and better context.
That sequence protects the organization from automating the wrong thing.
Build the foundation before adding the accelerant
The lesson is simple.
Do not begin by asking, “Where can we add AI?”
Begin by asking, “How should this work flow from beginning to end?”
That question changes the work. It brings the process into view before the technology gets selected. It forces teams to confront the real sources of delay, rework, excess inventory, and poor decision-making.
Once that foundation is in place, AI can create real value. It can remove repetitive work. It can identify risk earlier. It can evaluate options faster. It can help planners, buyers, and operators focus on the decisions that matter most.
But AI is the accelerant. Process discipline, connected data, and human judgment are still the foundation.
A faster broken process is still broken. A well-designed process, supported by AI, can become a supply chain that learns, adapts, and performs under pressure.
Comments