Last Thursday, researchers from Tampere and Aalto universities reported on preliminary results of two AI- and digitalization-related projects. One project demonstrated how AI can automate the Bill of Materials process; the other tried to quantify the impact of data-driven construction and AI. The conclusion was that AI-powered automation can dramatically increase productivity, but unevenly. However, the substantial benefits may come from better outcomes rather than project efficiency.
Various standards, company-specific practices, project-specific schemas, and traditions are designed for humans to comprehend. 2D drawings and text documents as PDF files, Excel sheets, and design and planning software have the information, but not the data in ideal form for automation.
AI can navigate the data maze
The Talotekniikka 2030 consortium’s project, led by the Tampere research team, was tasked with demonstrating whether the “messy” data could be used to automate the BOM process with agentic AI. The team shared a proof of concept showing it is possible.
The researchers started with an MEP design model (IFC), used an agentic process to analyze it, enriched it with external data, and created a procurement-ready Bill of Materials with actual products. They could do this on a locally run LLM, without paying for tokens.
BIM models are only one source of information. In mechanical and electrical design, information is spread across various documents, schematics, and diagrams. A comprehensive system must account for this information as well, unless it’s included in the model.
Even with non-ideal data, the efficiency boost agents provide is remarkable. Based on preliminary results, agentic AI delivers accurate enough results in hours rather than the days it would take a human estimator. It can also detect details humans would miss, which would cost a lot later in the process.
Benefits, but not evenly distributed
The other, Building 2030 consortium project, set out to map the future of digitalized construction and AI as an enabling technology. Naturally, the immediate benefits of AI-powered automation are in knowledge-intensive processes: design, estimation, procurement, and management. Logistics and on-site assembly will see more modest savings, at least in the short run.
The Aalto researchers presented figures on potential cost savings, both from verified previous projects and a bullish estimate for the future. In some existing project tasks, the measurable efficiency gains are remarkable. However, because 80% of the work happens on-site where AI’s impact is relatively small, the efficiency gains across the overall project are harder to quantify.
Interestingly, the discussion that followed led to a key finding. Rather than making designers work less, AI can broaden their scope and enable them to deliver new services that were previously too expensive or time-consuming. Another benefit beyond efficiency is the quality of the output. When AI helps clients, design teams, and contractors collaborate and optimize from the get-go, the entire project and product will improve.
Strategies for the future
The proof of concept we saw, and what many startups are currently developing, is about fixing a broken process after the fact. When participating organizations work in silos, you need intelligent agents to parse the information into a usable form.
For now and for many years to come, this approach makes sense. Many construction startup founders I’ve spoken with wouldn’t bet on an imminent shift to standardized, machine-readable data that enables fully automated processes without human intervention. They accept this reality and try to add industry intelligence and context to agentic processes that use LLMs.
Some startups have begun solving the problem upstream. Why not use AI to automate design tasks so the output is well-organized, machine-readable, and connected to external data? Combine that with standardization and digitalized supply chain processes, and we’ll have a 21st-century manufacturing information process.
Moving beyond experimentation
AI has moved beyond being a smart predictive typist to teams of agents that develop plans to fulfill their tasks and use creative tactics to access the data they need. Do we use this powerful technology to patch holes in the current process or to explore novel paths?
I would like to see more of the latter. For decades, we have applied new technology on top of old practices, limiting what we could have achieved. Outdated general contract conditions don’t help either.
A practical step forward is to form risk- and reward-sharing project alliances committed to removing one roadblock to a digitalized, AI-powered process at a time. Unlike alliances for large, complex projects, these can include forward-looking SMEs. The alliance should serve a client willing to continue beyond the first few projects, because this approach needs repetition to deliver the ultimate benefits.
P.S. I’ll keep you posted when the research projects I mentioned release their final reports