INVENTORY CLASSIFICATION WITH AI: EVALUATING HOW LARGE LANGUAGE MODELS ENHANCE CATEGORIZATION USING UNPSC CODES

Effective item categorization plays a crucial role in transforming unstructured datasets into organized categories, simplifying inventory management for businesses. However, this process is often subjective and lacks consistency across industries and typically requires extensive manual effort for implementation. The United Nations Standard Products and Services Code (UNSPSC) offers a standardized framework for inventory cataloguing. This study examines the use of Large Language Models (LLMs) to automate the classification of inventory data into UNSPSC codes as the chosen taxonomy based on item descriptions. It evaluates the accuracy and efficiency of LLMs when processing datasets that are large and diverse; and when focusing on a specific segment judging the effect of providing context to the LLM. The results demonstrate that LLMs can significantly reduce the manual workload while maintaining high accuracy of upto 90% at UNSPSC segment level when LLM is provided with context. These findings present LLMs as a scalable and efficient solution for businesses seeking to automate inventory management, with the potential for further improvement through advanced model architectures and refined prompt engineering.