BIMINAM AI AND DATA ENGINEERING WORKSTREAM MEETING, 16-04-2026
The meeting centered on the practical use of AI in BIM (Building Information Modelling) and asset management, highlighting both its potential and its current limitations. Early discussion emphasized that AI systems tend to be overly complementary and confidence-boosting, which can mislead users into overestimating output quality. Participants stressed that AI lacks real-world context and must not be blindly trusted.
A key theme was data quality and human oversight. Multiple contributors reinforced the principle of “rubbish in, rubbish out,” noting that AI outputs are only as good as the input data and prompt structure. Real-world examples (e.g., SFG20 fire door scheduling) showed AI failing to produce reliable results, reinforcing the need for validation, auditing, and domain expertise. Prompt engineering and structured instruction were identified as emerging skills essential for effective AI use.
The discussion then shifted to AI agents and automation. Unlike simple chat-based AI, agents can perform multi-step tasks using predefined skills, tools, and workflows. Demonstrations showed how AI can:
- Extract structured data (e.g., COBie data) from documents
- Build drawing registers from PDFs
- Conduct audits across multiple data sources
- Self-correct and iterate until tasks are complete
These capabilities highlight AI’s strength in process automation and scalability, rather than independent decision-making. However, risks were noted, including unintended behaviors (e.g., AI agents interacting with each other and generating unnecessary costs).
A major portion of the meeting focused on classification systems and data interoperability (e.g., UniClass, NRM, ETIM). Participants agreed that:
- Mapping between classification systems is inherently complex due to different structures and levels of granularity
- A data dictionary and “parent-child” hierarchy approach are critical for consistency
- No single classification system fits all use cases; instead, multiple systems must coexist and be linked
The group emphasized the importance of granular, well-structured asset data that can support various use cases (maintenance, compliance, logistics). AI can assist in managing and mapping this data, but human judgment remains essential, especially where context and interpretation are required.
Another key outcome was the group’s collaborative objective:
- To create a shared environment for experimentation
- To avoid duplication of effort (“reinventing the wheel”)
- To test AI solutions on real projects (e.g., NHS and construction case studies)
- To improve asset information quality and ensure it is accurate, auditable, and usable
Advanced applications were also presented, including AI systems that:
- Perform regulatory compliance checks on design documents
- Act as “engineering spell-checkers” by comparing designs against standards
- Use multiple specialized AI models (LLMs) orchestrated together for higher accuracy
Importantly, the meeting concluded that AI should be viewed as computational support—not a replacement for human thinking. Its role is to enhance productivity, automate repetitive tasks, and surface insights, while professionals remain responsible for interpretation, validation, and decision-making.
Key Takeaways
- AI is powerful but not inherently reliable without human oversight
- Data quality, structure, and context are critical for success
- AI agents enable scalable automation, but require careful design
- Classification and data interoperability remain major industry challenges
- The future lies in human–AI collaboration, not replacement