I’ve interviewed a number of startups that develop AI instruments for the AEC sector. They argue that the business creates knowledge for human use and judgment, so you’ll be able to’t depend on it being systematized. Therefore, LLMs are the way in which to go, as they’ll deal with messy knowledge.
Nevertheless, expertise exhibits that the place knowledge is structured and machine-readable, AI performs. The place it’s not, LLMs particularly both fail or produce believable however incorrect outputs.
This isn’t only a query of utilizing AI as a productiveness booster inside a single firm. It’s about provide chain-wide use of knowledge from early ideas to constructing operations, whether or not AI is concerned or not. Is it real looking to imagine that AEC corporations will begin listening to knowledge high quality and put money into enhancing it? In response to Solibri‘s MEP webinar final Thursday (Might 7), it’s doable and surprisingly straightforward when the frameworks and instruments are in place.
The trouble is just not what individuals suppose
On the webinar, Granlund’s Markus Järvenpää defined and demonstrated the usage of Finland’s nationwide knowledge commonplace for MEP objects, system varieties classification, and property units. It was developed through the RAVA2Pro and RAVA3Pro tasks and has been added to buildingSMART Information Dictionary (bSDD) platform (in Finnish).
The Finnish commonplace covers the total scope of MEP: over 800 product kind identifiers throughout HVAC, electrical, and constructing automation, plus system kind classifications and knowledge subject definitions for every. It’s publicly out there and has been applied within the main MEP design purposes utilized in Finland.
The workflow in an utility is a single motion: choose the product kind from a dropdown constructed into the software program (Markus used MagiCAD for Revit), and all required property units export routinely with the IFC. There isn’t any handbook property entry, no separate knowledge administration step, and no further coordination required. In a ten,000-hour undertaking, Markus estimated the overall added effort at roughly one hour!
Granlund has additionally launched a free Solibri extension that checks IFC fashions in opposition to the usual, closing the loop from classification to verification with none further price or tooling funding.
What structured knowledge unlocks
Structured MEP knowledge signifies that each object within the mannequin has agreed semantics, classifications, and property units {that a} machine can learn and compute with out guessing. Markus illustrated this in Solibri Workplace.
When MEP objects are constantly categorised, software program can routinely calculate portions, join them to price databases, simulate alternate options, and replace estimates dynamically. The identical structured basis allows carbon calculations, logistics planning by set up zone, and procurement based mostly on precise mannequin knowledge somewhat than abstract estimates.
Markus made a pointed commentary on the webinar: duplicate objects in unstructured IFC exports are frequent and are hardly ever detected. In the event you begin procurement from a mannequin containing duplicates, chances are you’ll order sixteen electrical distribution boards if you want far fewer.
All of this is applicable whether or not or not AI is within the workflow. However as AI instruments change into extra embedded in building processes, the hole between structured and unstructured knowledge widens additional.
Some argue that LLMs at the moment are sufficiently able to extracting which means from messy knowledge, making construction pointless. There’s partial fact on this; AI is genuinely good at changing unstructured data into structured data. However that doesn’t make the structured output pointless afterward. The extra AI brokers automate procurement, logistics, and lifecycle administration, the extra useful steady, machine-readable knowledge turns into beneath them.
The business barrier is the true downside
Provided that the instruments are prepared, the usual is printed, and the trouble for a designer is minimal, the apparent query is why structured MEP knowledge stays the exception. The reply is business. Purchasers don’t order it, so designers don’t produce it. With no contract requirement, the trouble that allows all the things downstream merely doesn’t occur.
Markus famous on the webinar that he has seen undertaking coordinators suggest adopting the MEP commonplace mid-project, solely to be advised by the MEP designer that it was not within the contract. It is a collective motion downside. Each occasion within the worth chain advantages from structured knowledge, however the fee falls on the designer, and the profit is captured by contractors, homeowners, and facility managers who didn’t pay for it. Till shoppers specify structured MEP knowledge in design contracts, the motivation construction won’t change.
Finland launched BIM-based constructing allowing this 12 months, however the regulatory framework doesn’t but mandate structured MEP knowledge, which might create a transparent authorized anchor for contract necessities. Which will change. Within the meantime, the sensible path ahead is easy: shoppers specify MEP compliance in design briefs, and designers who perceive the inner advantages begin making use of it regardless.
What must occur
Designers can act with out being requested. Markus made this level explicitly: the inner advantages in native MEP software program are rapid. Constant classification makes the mannequin extra usable inside the design crew earlier than any IFC export occurs. The designer who builds the behavior now can be the designer whose fashions will likely be AI-ready when shoppers start to demand it.
The systemic change requires shoppers. A easy contract clause referencing the usual is sufficient to shift the motivation. Purchasers who specify it is going to obtain fashions from which portions, prices, and carbon knowledge will be extracted routinely. Those that don’t will proceed to obtain fashions that require handbook interpretation at each handover. As AI instruments change into extra succesful and extra embedded in building workflows, the hole between these two positions will widen.
Success with AI in building relies upon not totally on the expertise chosen, however on a corporation’s capability to attach AI to its personal processes, knowledge, and strategic pondering. Structured MEP knowledge is exactly that connection level for the constructed asset. It’s not a prerequisite for experimenting with AI; it’s a prerequisite for AI delivering sturdy worth.
