The International Association for Automation and Robotics in Construction
Publications / 2025 Proceedings of the 42nd ISARC, Montreal, Canada
Large Language Models in Construction Bidding: Opportunities and Risks of Algorithmic Collusion
Algorithmic pricing systems have revolutionized industries, and their application in construction bidding is no exception. Integrating large language models (LLMs) for bid price setting introduces new opportunities and challenges in this domain. This study experimentally demonstrates the capability of LLM agents to generate competitive bidding strategies in a construction environment, demonstrating their ability to adapt rapidly and achieve pricing equilibrium, significantly faster than classic machine learning (ML) methods. Furthermore, we identify prompt-dependent collusion, where subtle variations in input phrasing influence bidding strategies, sometimes resulting in higher bid prices that mimic collusive behavior. Through textual analysis, we reveal how LLM-generated plans align with strategic price adjustments, highlighting the indirect role of prompt design in shaping bidding outcomes. While LLMs offer efficiency and adaptability, their susceptibility to algorithmic collusion raises concerns about transparency and market fairness. The findings underscore the need for regulatory oversight and further research to address the risks and opportunities of LLM integration in competitive construction bidding and broader market contexts.
