ChatGPT in Inventory Management — Opportunities and Limitations for Manufacturing Companies

2023 brought an explosion of interest in generative AI in business. In supply chain and manufacturing, the question shifted from "AI or not?" to "how?" and "where?". The study published in LogForum (DOI: 10.17270/J.LOG.2023.917) is one of the first empirical attempts to answer this specifically for inventory management.

The authors tested ChatGPT (GPT-4) on several concrete tasks: historical data analysis, formulating inventory policies, supplier communication, and documenting procedures. The results are mixed — and that is precisely what makes them useful.

Where GPT genuinely helps

The study identifies three areas where language models generate measurable value:

1. Qualitative analysis and documentation. GPT is exceptionally effective at processing unstructured information — meeting notes, supplier correspondence, shortage reports — and turning them into readable summaries or procedures. Tasks that take a planner three hours, GPT handles in ten minutes, at acceptable quality.
2. Generating questions and scenarios. The model can quickly generate a list of risk scenarios ("what if supplier X delays delivery by four weeks?") or review questions. It does not replace analysis — but it shortens the time needed to initiate it.
3. Cross-domain translation. In companies with multicultural teams or global supply networks, GPT helps translate specifications, contract terms, and operational communications — faster and cheaper than traditional translation services.

Where GPT fails — and why it matters

The study identifies three categories of failure with particular operational significance:

1. Quantitative calculations on complex data

GPT-4 makes calculation errors, particularly on multi-step computations (safety stock accounting for supplier and demand variability, EOQ with non-uniform holding costs). Worse — the errors are often stated confidently and convincingly. Without planner verification, results can feed into decisions.

2. Historical data and current stock levels

The model has no access to your ERP. Any analysis relies entirely on what you provide in the prompt. Every analysis requires manual data export and preparation — which partially negates the time savings.

3. Operational context and company specifics

GPT does not know your suppliers, relationship history, seasonal patterns, or internal constraints. It generates responses "for an average company" — which in inventory management can be worse than the intuition of an experienced planner.

How to use GPT in inventory planning — safely

Practical conclusions from the study and our experience:

• Use GPT for text, not numbers. Reports, procedures, communications, scenarios — yes. Calculating ROP, EOQ, safety stock — verify independently.
• Treat output as a draft, not a result. Every GPT response requires verification by someone who knows the context. It is an acceleration tool, not a decision tool.
• Data integration is critical. Systems that connect GPT to ERP data (via API or export) deliver several times better results than text-only prompting. This is the direction of development — but in 2024 it still requires technical work.

One question worth asking yourself

Before introducing GPT into your inventory planning process, answer honestly: does someone in your company have the time and competence to verify what the model generates?

If yes — the gain is real. If not — the model will produce responses that look good and may lead to poor decisions. In inventory management, mistakes are expensive — excess stock freezes cash, shortages stop production.