AI Agents and OR Algorithms Complement Each Other in Inventory Control
A new study explores how large language models (LLMs) and operations research (OR) algorithms can work together in inventory control. Traditional OR methods rely on rigid assumptions and fail under demand shifts, while LLMs offer flexible reasoning but lack structure. The researchers introduce InventoryBench, a benchmark of over 1,000 inventory instances with synthetic and real-world data, to test decision rules under demand shifts, seasonality, and uncertain lead times. The study finds that combining OR algorithms with LLM-based agents improves performance over either alone, suggesting a complementary human-LLM-OR approach for complex inventory decisions.
Key facts
- Inventory control is a fundamental operations problem.
- OR algorithms rely on rigid modeling assumptions.
- LLMs can reason flexibly and incorporate contextual signals.
- InventoryBench contains over 1,000 inventory instances.
- The benchmark includes synthetic and real-world demand data.
- The study tests decision rules under demand shifts, seasonality, and uncertain lead times.
- Combining OR algorithms with LLMs improves performance.
- The research proposes a human-LLM-OR complementarity framework.
Entities
Institutions
- arXiv