Suffolk Construction and MIT published a white paper on Tuesday that does something the construction AI conversation has mostly avoided: it puts a number on the thing. Applied together across six processes on a sample 180,000-square-foot multifamily project, artificial intelligence modelled out to a 17% to 20% cost reduction and a 22% to 25% schedule reduction.
Read the methodology before you quote the number.
Six AI levers, and the case for compounding
“Construction in the Age of AI” identifies design automation, offsite manufacturing, permitting, scheduling, skilled labor and subcontracting, and supply chain and procurement. The paper’s central argument is that these only pay when they run together, with design automation as the upstream enabler feeding permit compliance checking, procurement automation, schedule generation and subcontractor coordination off one data foundation.
Suffolk chief technology officer Jit Kee Chin framed it as arithmetic that doesn’t behave like arithmetic. “This is not like a math exercise for 1+1+1 equals 3,” she said. “It’s 1+1+1 equals, potentially, 6.”
What the study actually measured
Nothing was built. The case study is a retroactive analysis of a $180 million San Francisco multifamily project completed in 2024, and the findings were assembled from a roundtable, a literature review and expert interviews rather than a controlled trial or a survey with a defined sample frame. Chin says so herself, acknowledging the results aren’t causal and pointing to a roadmap that asks for more studies and better construction data standards.
It’s also worth naming the obvious: the modelling comes from a general contractor with a technology arm, co-branded with a university. That doesn’t make it wrong. It does mean the 17-20% is a projection produced by a party with an interest in the answer, and it should be quoted that way.
Why anyone is bothering now
The authors tie the urgency to three things. Models are getting better. Construction data volume has finally reached a level where training on it is worth doing. And economic pressure has closed off the option of absorbing waste. Running alongside that is the load the data center boom is putting on the industry: planned or operating projects nearly tripled between March and August 2026 according to Cleanview, a third-party figure nobody has independently checked.
Chin’s closing framing is the right one for contractors reading this. “There’s enough evidence out there now that says that this vision is possible. The question is: What do we all do about it?”
For most firms the honest answer is not much yet. Design automation is the only lever with off-the-shelf tooling that works on a normal job, and the compounding case collapses without it. Anyone selling the schedule number without the design foundation is selling the conclusion without the premise. The projects where this eventually pays are the complicated ones with thousands of coordination points, the Centre Block rehabilitations of the world, and those are exactly the jobs with the least clean data to train on.