AI-Driven Peer Company Identification: A Semantic Text-Similarity Approach Beyond Traditional Industry Classification Systems

Background: Traditional classification systems, such as NACE and NAICS, primarily classify businesses by industry, limiting their ability to identify related companies. Objectives: This research aims to improve the identification of related companies by analysing their descriptions, utilising a more semantic approach. Methods/Approach: A pre-trained BERT model was employed to assess semantic text similarity for suggesting peer companies. The goal was to create a system that assists experts in comparing companies based on their descriptions, rather than developing a perfect classification tool. Results: Trained on publicly available data, the model achieved 73.6% accuracy in identifying related companies, with accuracy exceeding 90% for selected industry-pair combinations. Conclusions: While the system demonstrates promise, its outputs are intended to guide professionals who must ultimately validate the results. The findings emphasise the strengths and limitations of using AI models for this purpose, providing a foundation for future enhancements and real-world applications. However, the solution remains a conceptual idea, limited to only 13 industry categories, highlighting the need for broader testing and development.