Does AI Eliminate the Need for Supply Chain Expertise?

Artificial intelligence has become a dominant topic of discussion throughout industry.  The technology is on a path of rapid adoption and implementation, with many use cases reported in a broad array of industries. Which raises the following questions: Why do we still need to teach the fundamentals of supply chain management and analytics to students? And why do we need supply chain professionals with a deep understanding of the discipline?  In other words, why can’t any bright professional with limited supply chain knowledge successfully use AI to guide his or her decision-making and management of supply chain functions?

The answer is that while generative and other forms of AI have demonstrated extraordinary capabilities and future promise, the technology represents just another tool that requires human supply chain understanding, experience and expertise. As with any major decision-support system (DSS), effective supply chain management requires individuals who can (1) intuitively understand and assess a supply chain problem or need, (2) determine the appropriate decision variables that must be considered to make a decision, (3) develop the appropriate decision-making input data that is required, and (4) evaluate the outputs of their management decision-support systems.

An important first step in evaluating the output of a DSS is to assess the validity and reasonableness of a recommended action or analytic solution. Such action requires that the evaluator has knowledge and understanding of supply chain management fundamentals, principles and practices.

Is there, for example, a basic flaw in the logic underlying the AI-generated answer? In AI parlance, did the tool have a hallucination, or fail to consider all appropriate variables? Or perhaps the problem the AI is being asked to address wasn’t properly defined, leading to a response that’s off target.

In deploying current AI tools for supply chain management, one must be facile in their use — similar to how a user of spreadsheet software must develop the relevant expertise.   Additionally, the user must possess the functional supply chain knowledge needed to frame a problem or question, and understand the variables that affect a problem and solution.

A Network-Design Case

The following distribution network-design case illustrates the type of supply chain management knowledge, acquired through education and experience, that a manager must have.

Consider the following fictitious scenario. Firm XYZ is a domestic consumer products manufacturer with one manufacturing plant located in approximately the geographic center of the U.S. It’s currently deciding whether its future U.S. distribution network should have two or four distribution centers.

XYZ distributes to a customer base of major, mid-size and small retailers. They stock its products in their own stores and sell them to local customers. XYZ offers delivered pricing to its retail customers, and thus pays the freight costs both from its plant to the DCs and from the DCs to retail customers. Additionally, XYZ utilizes truck transportation for all plant-to-DC and DC-to-retailer shipments.

XYZ has determined that its analysis of the two alternative networks should address several key questions, as well as the total annual costs of each option. Specifically, it wants to know whether the following costs will be higher, lower or about the same on a two-DC network compared to one with four DCs:

• Plant-to-DC freight costs;
• DC-to-retail customer freight costs;
• Total finished goods inventory investment and inventory carrying costs, and
• Total warehousing costs.

 Figure 1 depicts the two alternative networks under consideration.

The AI Scenario

For illustrative purposes, assume that an employee of XYZ has developed the firm’s individual cost inputs (such as truckload cost per mile) and annual demand forecasts for its future distribution network. Further, assume this employee provided their favorite AI model with this input data, and asked the model to project the total annual costs of both the two- and four-DC networks. The AI model analyzed all the data inputs and developed the cost analysis shown in Table 1.

Taken at face value, this analysis would suggest that XYZ could save $2.5 million annually by utilizing a two-DC network rather than a four-DC one. However, a well-trained, experienced supply chain professional would immediately recognize that several of the cost projections in Table 1 appear directionally incorrect.

What would a well-trained SCM professional spot in the Table 1 results that a non-SCM colleague might miss, and how would they identify the directional errors? The professional has studied and understands all of the relationships and trade-offs of the individual cost and service variables that collectively make up a distribution network design problem. They would assess the validity and potential accuracy of the distribution network analysis guided by the following general SCM rules:

Plant-to-DC freight costs. The annual total freight costs of all shipments from the plant to the DCs should either be lower on the two-DC network than on the four-DC one, or they should be equal. (The plant ships the identical total tonnage in either network option. Therefore, on the two-DC network, the average pounds per shipment will be higher or equal to that on a four-DC network. Further, the total shipments on a two-DC network will be either lower or equal to the total shipments on a four-DC one.)

DC-to-retail customer freight costs. The annual total freight costs of all shipments from the DCs to the stores of XYZ’s retail customer locations should be lower on the four-DC network than on the two-DC one. (The shipment weights will be identical on each network. However, the average distance that trucks must travel between XYZ’s DCs and its retail customer locations will be lower on a four-DC network than on a two-DC one, and therefore the freight costs of the four-DC network will be lower.)

Total finished goods inventory investment. XYZ’s total finished goods inventory investment requirement and, therefore, its annual inventory carrying costs, should be lower on a two-DC network than on a four-DC network. (Serving the same demand and market from fewer stocking locations smooths the variability of demand, and therefore reduces total safety stock requirements — as demonstrated by the inventory pooling principle.)

Total warehousing costs. In general, XYZ’s total annual warehousing costs should be lower on a two-DC network than on a four-DC one. (Fewer facilities generally reduce total fixed costs across a network and facilitate potential economies of scale.)

Note that Table 1 indicates the total inventory investment — therefore, annual inventory carrying costs on a two-DC network will exceed those of the four-DC one. Conversely, the AI model projects that DC outbound freight costs from XYZ’s DCs to its retail customers on a four-DC network will surpass those on a two-DC network.  An experienced supply chain professional would immediately recognize that both the inventory and outbound freight cost projections violate standard logistics operating fundamentals — they’re directionally incorrect — and the model’s outputs can’t be trusted. Clearly, the inputs provided to the AI model must be scrutinized. Are there issues with the data provided to the AI model? Were the prompts and directions given to the model misguided?  

The point is that a manager not trained and experienced in supply chain and logistics management wouldn’t have the expertise to recognize the incongruities in Table 1’s results. Should XYZ move forward based on these results, the firm would risk making sub-optimal decisions.

The simple distribution network-design problem cited above illustrates an intuitive conclusion: While AI models provide extraordinary current capabilities and future promise, there remains a need for well-trained, experienced supply chain professionals. The education, training and practice of supply chain management must incorporate AI tools and advances, as with the prior adoption of technological innovations such as optimization models, spreadsheet software, the internet and robotics. That said, the demand for expert supply chain functional oversight and management will remain.

(Editor’s note: The network design problem in this article is excerpted from Practical Supply Chain and Logistics Analytics, first edition, by Tan Miller, Editor Katya Seward, Copyright 2026 by Imprint. Reproduced by permission of Taylor & Francis Group.)

Tan Miller is Professor of Global Supply Chain Management and adviser to the Global Supply Chain Management Program at Rider University’s Norm Brodsky College of Business.