Can LLMs Plan Your Supply Chain?

For the past couple of years, we have all been inundated with Artificial Intelligence (AI) and how it is going to change the world as we know it. There’s no doubt it will have an impact, but the real question is, how? When it comes to supply chain planning, there is a lot of buzz about how we can perform not just supply chain planning but also factory planning and scheduling using AI/ML (Artificial Intelligence and Machine Learning). 

Well, I have news for you, AI/ML has already been used in these areas for over three decades, leveraging search techniques such as gradient descent, divide and conquer, and constraint propagation. 

So why does this matter now? Because the way we interact with these complex systems is fundamentally changing. With LLMs, there’s potential to reduce decision-making time, improve responsiveness, and democratize access to insights, meaning faster response to disruptions and fewer costly mistakes. 

But now the claim is that we can use Large Language Models (LLMs) to do all kinds of planning. Supply chains, like the stock market, are complex and unpredictable. We can assess risks and hedge against them. However, when an event occurs, the question is how quickly we can be informed about it, how fast we understand its impact, and what actions can be taken. LLMs can assist in interpreting the nature of the event to keep us informed. 

That means your business can stay ahead of disruptions, maintain service levels, and avoid costly downtime, not by replacing humans, but by amplifying human capability. 

To grasp the impact of the event, we need a thorough and up-to-date (real-time) understanding of the operations, the state of the supply chain, and all its dependencies. The first two stages help to reduce decision latency. The next challenge is determining the right response and how optimal that response should be. The latter is also referred to as a Prescription. 

Prescribing a solution requires analyzing the entire supply chain and all the possible choices, followed by recommending actions. This process must account for evolving business goals, as well as limitations and constraints in the supply chain, such as capacities, lead times, material availability, or even tariffs and carbon emissions. 

In short, understanding where LLMs help, and where they don’t, can save your company from overhyped tech investments and guide you toward smarter adoption strategies that actually move the needle. 

Of the three stages mentioned above, where would you say LLMs can help? Perhaps in the first stage. They can interpret a message, such as an oil spill at a supplier site or an incoming tornado, to warn of the event. To understand the impact of the event, one must have a model (digital twin) of the environment. To know what to do, one needs to optimize or use heuristics that can quickly prescribe appropriate actions. LLMs can also help by translating system recommendations into natural language for users or by explaining outcomes in more accessible terms. 

This helps cross-functional teams understand and act faster, whether it’s procurement, logistics, or executive leadership, bridging the gap between technical data and strategic decisions. 

Before we conclude this topic, let me give you a poor example of the use of Machine Learning (ML) in planning & scheduling. Many believe that one can predict manufacturing lead times using ML. Yes, this is possible. However, it is a silly way of doing it because they can be calculated easily without relying on a technique that merely reinforces historical norms. How could you then improve your lead times if you keep learning from past practices? 

Generative AI (GenAI), using “tools,” can indeed perform optimization; however, any kind of intelligence needs to have an understanding of the world around it, what the constraints are, physical and emotional, and then apply techniques to arrive at an “optimal” answer. 

Understanding this distinction helps organizations avoid wasting time on the wrong use cases and focus on where AI truly adds value, augmenting human decision-making, not blindly replacing it. 

LLMs are great at understanding and communicating events in the supply chain (e.g., parsing alerts), but not at optimizing or planning what to do next, which still requires proper models and algorithms. So, don’t expect them to magically replace decades of supply chain logic. Use them as a helpful interface, not the brain of your operations. 

In other words, care about this because knowing the real strengths and limits of LLMs could be the difference between leading your industry or falling behind.