Artificial Intelligence in BIM Clash Management: Assessment of Current Techniques, Automation Levels, and Deployment Readiness

Automated clash detection in Building Information Modelling (BIM) has reached operational maturity; however, subsequent stages of clash management, including filtering, prioritisation, resolution, and prevention, remain largely dependent on manual expert judgement. Artificial intelligence (AI) has been increasingly applied to address these limitations, yet focused synthesis remains limited on how AI techniques support different stages of BIM-based clash management, what levels of automation they achieve, and how ready they are for professional deployment. This review follows PRISMA 2020 and synthesises 21 empirical studies published between 2019 and 2026, providing one of the first focused syntheses of AI applications across the BIM-based clash management lifecycle. Five families of AI techniques are identified, including rule-based systems, supervised machine learning, deep learning, graph neural networks, and reinforcement and generative AI. The results show that research is concentrated on filtering, while resolution and prevention remain underexplored despite their greater coordination significance. A structured automation taxonomy is developed, revealing that systems described as autonomous are evaluated only in simulated environments. The synthesis further indicates that progress is constrained primarily by structural rather than algorithmic limitations, as most studies rely on single-project validation, report performance without class-level metrics, and produce black-box outputs that cannot meet the accountability requirements of professional coordination practice. The primary barrier to deployment is therefore not model capability but the absence of shared benchmark datasets, rigorous cross-project validation, interpretable outputs, and sociotechnical integration with professional practice.