BIM–AI integration in architecture: a workflow-based review

The integration of Building Information Modeling (BIM) and Artificial Intelligence (AI) has become an important research topic in architecture, engineering, construction, and facility management. However, existing reviews mostly classify related studies according to technical domains or AI techniques, rather than examining how BIM is used within AI-based workflows. This study presents a systematic review of BIM–AI research published between 2015 and 2025. Following a PRISMA-based process, 113 studies were selected from Scopus and Web of Science and analyzed according to five dimensions: (i) the BIM–AI problem addressed, (ii) the specific AI algorithm used, (iii) the broader AI method applied, (iv) BIM’s role within the AI workflow, and (v) the validation approach used to evaluate the workflow. The results show that BIM–AI research is mainly concentrated on Scan-to-BIM automation, condition assessment, and energy performance prediction. Deep learning is the most widely used AI method, while machine learning remains common in prediction and management-related studies. The findings also show that BIM is often used as a data source or a visualization environment, whereas semantic BIM integration is less frequent. The reviewed BIM–AI workflows are mostly evaluated through case studies and experimental demonstrations, while benchmark-based evaluation and cross-validation are less frequently used in the reviewed studies. Overall, the review shows that BIM–AI research is rapidly expanding but remains methodologically fragmented. The study contributes by identifying dominant research directions, underexplored areas, and opportunities for further developing BIM–AI workflows, semantic integration, and validation strategies.