AI in the Chain

Artificial intelligence is moving quickly into supply chain conversations. Every week, new discussions emerge about AI agents, copilots, autonomous planning, digital twins, predictive analytics, and intelligent control towers. The promise is attractive: faster decisions, fewer manual tasks, better forecasts, improved service levels, lower inventory, and more resilient operations.

But there is a risk hidden behind the excitement.

Many companies are trying to apply AI to planning processes that are already fragile. They have disconnected systems, inconsistent master data, unclear decision rights, conflicting KPIs, and planning routines that still depend heavily on spreadsheets, manual follow-ups, and individual experience. In these environments, AI may accelerate activity, but not necessarily improve decision quality.

This is the uncomfortable truth: AI will not fix broken planning.

If a company has poor data, unclear accountability, and outdated planning logic, an AI agent will not magically create a high-performing supply chain. It may only automate confusion faster.

The next phase of supply chain transformation should not begin with the question, “Which AI tool should we buy?” It should begin with a more fundamental question: “Which decisions are we trying to improve?”

The Problem Is Not Lack of Technology

Supply chain leaders are not short of technology. Most companies already have ERP systems, planning platforms, warehouse systems, transport systems, supplier portals, reporting tools, and business intelligence dashboards. Many have also invested in advanced analytics, forecasting models, robotic process automation, and now generative AI.

Yet the planning experience remains difficult.

Demand planners still spend time cleaning data instead of interpreting demand. Supply planners still chase missing confirmations instead of evaluating scenarios. Procurement teams still react to supplier issues instead of anticipating risk. Customer delivery teams still manually reconcile promises, constraints, and exceptions. Senior leaders still ask why the numbers in different reports do not match.

This is not because technology is absent. It is because planning is often built on weak foundations.

The traditional planning model was designed for a more stable world. It assumes that historical demand is a reasonable guide to the future, that lead times are predictable, that suppliers perform within known patterns, and that planning cycles can be managed through periodic reviews. But the supply chain environment has changed. Demand is more volatile. Geopolitical risk is higher. Customers expect faster responses. Product portfolios are more complex. Supplier networks are more exposed. Sustainability and compliance requirements are growing. Disruptions are no longer rare exceptions; they are part of normal operations.

In this environment, simply adding AI to old planning structures is not transformation. It is decoration.

Why AI on Top of Old Planning Can Create New Problems

AI is powerful, but it is not neutral. It learns from available data, follows defined objectives, and produces outputs based on the structure it is given. If the planning process is poorly designed, AI can reinforce the wrong logic.

For example, if forecast accuracy is measured only at an aggregated level, AI may optimize the total number while missing critical item-level service problems. If inventory targets are disconnected from supplier risk, AI may recommend lower stock without understanding fragility in the supply base. If customer prioritization rules are unclear, AI may suggest allocations that are mathematically efficient but commercially wrong. If master data is unreliable, AI may produce confident recommendations based on incorrect assumptions.

This is especially important with agentic AI.

Unlike traditional analytics, AI agents can break goals into tasks, interact with systems, prepare actions, and sometimes execute steps. In supply chain, an agent could review forecast changes, identify shortages, contact suppliers, recommend reallocations, draft customer updates, or trigger workflow actions. This creates enormous potential. It also creates new operational risk.

An AI agent working inside a weak planning process may not just report bad information. It may act on it.

That is why governance, decision rights, and data reliability matter more than ever.

Lora Cecere’s Warning: Do Not Automate the Wrong Architecture

Lora Cecere has been one of the strongest voices challenging supply chain leaders to rethink planning before chasing the next technology wave. Her recent message is clear: companies should not simply place AI, agents, or advanced analytics on top of outdated planning architectures.

This warning matters because many supply chain transformations fail by treating technology as the starting point. The organization buys a new platform, adds more dashboards, deploys automation, and expects behavior to change. But if the underlying decision model remains the same, the results are limited.

Planning transformation requires more than better tools. It requires a better architecture for decision-making.

That means asking difficult questions:

What is the purpose of the plan?

Which decisions should be centralized and which should remain local?

Which data is trusted?

Which assumptions are still valid?

Which KPIs create the right behavior?

Which exceptions require human judgment?

Which decisions can be automated safely?

Which ones should never be automated?

Without this clarity, AI becomes another layer of complexity.

From Data Visibility to Decision Quality

Many companies still confuse visibility with decision quality.

Visibility is knowing what happened. Decision quality is knowing what to do next.

A dashboard may show that demand increased. A report may show that a supplier is late. A system may show that inventory is below target. But the real planning question is more complex: Should we expedite? Should we reallocate? Should we adjust the forecast? Should we inform the customer? Should we accept the risk? Should we increase safety stock? Should we challenge the demand signal?

This is where AI can help, but only if the planning logic is clear.

The value of AI is not simply to produce more alerts or faster reports. The value is to help planners compare options, understand trade-offs, and make better decisions under uncertainty.

A mature AI-enabled planning process should move from “What is the number?” to “What decision does this number support?”

That shift is fundamental.

The Need for a Semantic Layer

One of the most important concepts in modern AI-enabled supply chain planning is the semantic layer. In simple terms, a semantic layer helps systems understand business meaning. It connects data to context.

For example, “available inventory” may mean different things depending on whether stock is unrestricted, quality blocked, allocated, in transit, reserved for a customer, export restricted, or physically available in the right location. A human planner often understands these differences intuitively. An AI system needs the meaning to be structured.

The same applies to lead time, forecast, order status, supplier confirmation, customer priority, and service level. If these terms are not consistently defined, AI may interpret them incorrectly.

A semantic layer helps translate raw data into business language. But it cannot compensate for poor data quality at the source. If the data is wrong, incomplete, outdated, or inconsistently maintained, the semantic layer only organizes the weakness more elegantly.

This is why AI readiness is not only an IT topic. It is a supply chain operating discipline.

Data Fabric, Data Mesh, and the Reality of Source Data

Data fabric and data mesh are often discussed as modern architecture concepts. They can help connect data across systems and domains, making it easier for AI to access information from demand, supply, procurement, logistics, finance, and customer operations.

But architecture alone does not solve the planning problem.

If source data is unreliable, better connectivity may simply move bad data faster. If ownership is unclear, data issues remain unresolved. If teams do not agree on definitions, different parts of the organization will continue to interpret the same signal differently.

The real challenge is not only technical integration. It is business ownership.

Who owns customer master data?

Who owns supplier lead time accuracy?

Who validates product lifecycle status?

Who maintains planning parameters?

Who confirms whether a data field is operationally meaningful?

AI requires answers to these questions. Otherwise, it becomes dependent on data that no one fully trusts.

The Planning Process Must Become Outside-In

Traditional planning is often inside-out. It starts from internal history, internal targets, internal constraints, and internal planning cycles. But today’s supply chains are shaped by external signals: market volatility, customer behavior, supplier disruptions, geopolitical risk, weather events, regulation, port congestion, labor constraints, energy costs, and technology shifts.

An AI-smart planning process should be more outside-in.

This does not mean abandoning internal data. It means enriching internal planning with external context. Demand history is important, but it may not capture sudden market changes. Supplier performance history is important, but it may not capture new financial stress or regional disruption. Inventory policy is important, but it may not reflect changing risk exposure.

AI can help companies monitor weak signals and connect them to operational decisions. But again, this only works if the company knows which decisions should change when a signal appears.

A signal without a decision rule becomes noise.

The Human Role Changes — It Does Not Disappear

A common fear is that AI will replace planners. A more realistic view is that AI will change what planners do.

Today, too many planners spend their time collecting information, reconciling numbers, chasing updates, and preparing reports. These activities are necessary, but they are not where human judgment creates the most value.

In an AI-enabled planning model, routine data gathering and first-level analysis should increasingly be handled by machines. Human planners should spend more time on exception management, scenario evaluation, cross-functional trade-offs, supplier and customer judgment, and decision governance.

The planner becomes less of a spreadsheet firefighter and more of a decision orchestrator.

This shift requires new skills. Planners will need to understand data quality, AI recommendations, scenario logic, exception prioritization, and risk trade-offs. They do not need to become data scientists, but they do need to become more fluent in how AI supports decisions.

The future planner will ask better questions:

Why is the AI recommending this action?

Which assumption drives the recommendation?

What trade-off is being optimized?

What risk is not visible in the data?

What would change if demand, supply, or lead time assumptions were different?

This is not a lower-value role. It is a higher-value role.

A Practical Framework: Automate, Augment, or Keep Human-Led

Not every planning decision should be treated the same way. A useful approach is to classify decisions into three categories.

1. Automate

These are low-risk, high-frequency decisions with clear rules. Examples may include report generation, exception classification, routine status updates, basic data validation, or standard workflow reminders.

AI and automation can handle these tasks efficiently because the cost of error is relatively low and the decision logic is stable.

2. Augment

These are decisions where AI should recommend, but humans should approve. Examples include forecast overrides, replenishment adjustments, supplier escalation prioritization, inventory rebalancing, transport mode recommendations, or customer order risk assessments.

In these cases, AI can analyze options and prepare recommendations, but human judgment remains important.

3. Keep Human-Led

These are high-risk, strategic, or commercially sensitive decisions. Examples include customer allocation during shortages, major supplier changes, significant inventory write-offs, compliance exceptions, or decisions that affect contractual commitments.

AI can support these decisions with analysis and scenarios, but accountability should remain with people.

This framework helps prevent two common mistakes: automating too much too soon, or blocking AI from areas where it could safely create value.

The KPI Problem: What Are We Optimizing?

AI systems need objectives. But supply chain objectives are often conflicting.

A company may want high service levels, low inventory, stable production, low logistics cost, strong supplier relationships, and high forecast accuracy. These goals do not always move in the same direction. Improving one metric can damage another.

If AI is asked to optimize planning without clear KPI hierarchy, it may produce recommendations that look good in one area but create problems elsewhere.

For example:

Reducing inventory may increase customer risk.

Improving forecast accuracy at an aggregate level may hide item-level shortages.

Minimizing transport cost may reduce responsiveness.

Maximizing supplier efficiency may reduce resilience.

Improving planning stability may reduce agility.

This is why AI-enabled planning must be connected to business strategy. The system must understand which trade-offs are acceptable and which are not.

The best supply chains will not use AI to optimize isolated KPIs. They will use AI to manage trade-offs transparently.

A Control Tower Is Not Enough

Many companies have invested in supply chain control towers. These can improve visibility, but visibility alone does not create better planning. A control tower that only shows alerts may become another source of noise.

The next generation of control towers should be decision-centered.

Instead of simply showing that a supplier is late, the system should explain which customers are affected, which alternatives exist, what each option costs, what risks remain, and who needs to decide. Instead of showing that forecast changed, it should identify whether the change is meaningful, whether supply can respond, and whether the business should accept or challenge the signal.

AI can make control towers more useful by moving from monitoring to recommendation. But for that to work, the company must define the decisions that the control tower is meant to support.

A control tower without decision logic is only a better screen.

The Real Starting Point: Decision Mapping

Before deploying AI agents, companies should map their planning decisions.

This does not require a massive transformation program. It starts with a practical exercise:

List the recurring supply chain decisions made every week.

Identify who makes each decision today.

Define what data is used.

Assess whether the data is trusted.

Clarify which KPI the decision is meant to improve.

Classify the risk level.

Decide whether the decision should be automated, augmented, or human-led.

Identify what would need to change for AI to support it.

This exercise often reveals the real barriers to AI value. The issue may not be the algorithm. It may be unclear ownership, poor master data, conflicting KPIs, missing business rules, or lack of process discipline.

Once these issues are visible, AI deployment becomes more grounded.

Unique Insight: AI Transformation Starts Before the AI Tool

The most important work in AI-enabled supply chain planning happens before the AI tool is deployed.

It happens when leaders define what good planning means.

It happens when teams agree on data definitions.

It happens when decision rights become clear.

It happens when KPIs are aligned with business strategy.

It happens when planners move from manual reconciliation to exception-based decision-making.

It happens when companies stop treating AI as a shortcut and start treating it as an amplifier.

AI amplifies what already exists. If the planning process is disciplined, AI can scale discipline. If the planning process is fragmented, AI can scale fragmentation.

This is why the maturity of the operating model matters as much as the maturity of the technology.

Practical Prompt for Supply Chain Leaders

Use this prompt in a leadership workshop:

“Identify one planning process where the organization wants to deploy AI. Before discussing tools, map the top 10 decisions inside that process. For each decision, define the owner, data inputs, business rules, KPIs, risk level, and approval requirements. Then classify each decision as automate, augment, or human-led. Finally, identify the data and governance gaps that must be fixed before AI can create value.”

This prompt changes the AI conversation. It moves the discussion from technology excitement to operational readiness.

Final Thought

AI will play a major role in the future of supply chain planning. Agentic AI, copilots, predictive models, and intelligent control towers will become increasingly common. But technology alone will not create better planning.

The real opportunity is not to add AI to broken processes. The real opportunity is to redesign planning so AI can support better decisions.

Companies that skip this step may automate noise, accelerate bad assumptions, and create new risks. Companies that do the hard work of improving data, decision rights, governance, and planning logic will use AI as a true competitive advantage.

The future of supply chain AI is not about adding intelligence to broken planning.

It is about rebuilding planning so intelligence can create value.