أثر استخدام الذكاء الاصطناعي في علوم سلاسل الإمداد

Introduction: a question older than the current wave

Interest in artificial intelligence among supply chain scholars did not begin with generative models. It began when demand forecasting, inventory planning and shipment routing became very large mathematical problems that planner intuition and spreadsheets could no longer solve. But the current enthusiasm mixes two things: what has been established by peer-reviewed research, and what is said in marketing presentations. This article tries to separate them.

The article answers three questions. What do the reviewed studies say about the effect of artificial intelligence on supply chain performance? In which functions is that effect clear? And what has not yet been established? The article does not offer ready-made return figures, because the literature we reviewed does not offer them either.

What the peer-reviewed literature says

The broadest review I found on demand planning is by Walter, Ahsan and Rahman, published in 2025 in The International Journal of Logistics Management. The review covered 141 peer-reviewed studies published between 2012 and 2023. About 81% of them appeared after 2019, and the number of studies peaked at 37 in 2023. In other words, the field is growing quickly but is still young.

These studies divide into mathematical models (55%), empirical studies (37%) and conceptual work (8%). Predictive AI dominates 79% of the publications. The authors concluded that the use of artificial intelligence in demand planning is still, in their words, at an early stage. Benefits recorded in the literature include lower costs through inventory control, faster response to external changes, and support for operational resilience.

A review by Logožar, published in the journal ENTRENOVA, agrees with this; it relied on the Scopus and Web of Science databases. The review identifies four main application areas: demand forecasting, inventory management, logistics optimization, and risk mitigation. It concludes that these applications help make better decisions, reduce forecasting errors and control inventory levels.

Where artificial intelligence changes practice

The two reviews agree on four places where the effect appears, and I present them in the order in which decisions connect within the chain.

Demand. This is the largest body of literature. Demand forecasting is a predictive problem at its core, which explains why 79% of the publications in the Walter et al. review belong to predictive AI. The methods used go beyond traditional regression to models that learn patterns from sales data and external indicators. The review summarizes the expected effect as a greater ability to capture fluctuations, not as guaranteed accuracy, because most empirical studies are built on a single dataset.

Inventory. Inventory is tied to demand as an effect is tied to its cause. The better the forecast, the lower the need for safety stock. The Walter et al. review cites the study by Fu and Chien (Fu & Chien, 2019), which indicates that excess safety stock can reach 60% of total inventory investment in intermittent-demand manufacturing. This figure describes the size of the problem that better forecasting tries to reduce; it does not describe the savings that artificial intelligence achieves. The difference between the two is what marketing presentations overlook.

Logistics. The Logožar review lists logistics optimization as a separate application area, covering route planning, load distribution and transport scheduling. These are optimization problems that have been known in operations research for decades; what modern methods add is the ability to recompute when inputs change, not the invention of the problem itself. The Walter et al. review notes that the literature is thin on distribution, spare parts and returns, areas that sit at the heart of distribution companies’ work.

Risk. Risk mitigation covers the early detection of disruptions: a supplier delay, volatility in demand, or a breakdown on a route. The Walter et al. review links this capability to operational resilience, meaning the chain’s ability to absorb a shock and return to its previous level.

This article’s reading of the above is that the clearest value lies in the first link, demand, because the other links depend on its outputs. This is the author’s analytical reading, not a conclusion stated by either review.

The new direction: autonomous agents, physical AI and decision governance

Professional sources differ from the peer-reviewed literature in that they look forward more than they measure the past. In this context, Gartner identified three themes in its 2026 supply chain technology trends: Agentic AI, Physical AI and Decision Governance. These are forecasts from an advisory firm, not findings of peer-reviewed studies, and the published report contains no figure measuring their effect.

The first theme refers to software agents and multiagent systems that take over steps of the decision rather than merely handing a forecast to the human planner. The second refers to linking artificial intelligence with the Internet of Things (IoT) and robotics, that is, moving the decision from the screen to the warehouse and the truck. The third addresses the question that follows from the other two: who is accountable when an autonomous system decides, and under what controls? Gartner’s analysis states that trust and governance are a pillar parallel to autonomy, not an appendix to it.

Interpreting these trends against the earlier literature opens a striking observation. The peer-reviewed reviews study predictive systems whose outputs a human reviews, while market talk is about systems that execute. So far, among the systematic reviews I have seen, none measures the effect of this shift. This is an observation by the author, not a conclusion taken from a source.

Gaps and limits of the scientific evidence

The Walter et al. review sets out its own limits, and these limits make a useful list of questions before any investment decision:

• Generalizability. Empirical studies are mostly built on a particular dataset, so a result should not be transferred to another company without testing.
• Weak theoretical grounding. Few studies rest on a theory, so many results remain descriptive.
• No economic analysis. Few studies compare cost with return, so the literature offers no reliable figure for return on investment. The review itself does not measure effect size.
• Organizational factors. Trust, readiness and skills receive marginal attention, even though they often decide whether an implementation succeeds in companies.
• Topical gaps. Spare parts, returns and distribution are less present than demand forecasting.
• Method limits. The search was limited to material published in English in the Scopus and Google Scholar databases, so work published in other languages, including Arabic, may be missing.

In addition, Gartner’s trends do not enter the evidence count at all, because they are an estimate of what will happen, not a measurement of what has happened.

Conclusion and what it means for managers and researchers

Taken together, the material leaves three supported conclusions. First, the presence of artificial intelligence in supply chain research is documented and accelerating (141 studies, 81% of them after 2019), but the field has not matured. Second, the clearest place where it has an effect is demand forecasting and inventory control, while logistics and risk are less well documented. Third, the financial return is not measured in the reviews I consulted.

For managers. The decision starts with a question about what the current problem costs: what share of your total inventory is safety stock, and what is the actual forecast error for each item? That is the baseline against which any model is measured. A limited trial on a single product category, compared against the existing method, is advisable, because the results of published studies do not necessarily carry over from one project to another. As for autonomous agents, do not take a step forward before deciding who approves their decisions and who bears responsibility for them.

For researchers. The priority is what the review itself points to: studies grounded in a theoretical framework, economic comparisons of cost and return, research on organizational factors, and applications in markets that have not been studied enough, including those of Egypt and the Gulf. This is a conclusion by the author, not a published statistic.

Note: the Fu & Chien (2019) study is cited as reported in the Walter et al. review; its original text was not consulted.