An independent magazine about AI at work in construction. Every story starts on a real job site, with the people who used the technology and the results they will answer for.
Suffolk and MIT Model 18% AI Savings on One Job and Call It Directional | ConstructionMagazine.ai
Suffolk and MIT Model 18% AI Savings on One Job and Call It Directional
A 16 September 2026 Suffolk and MIT white paper models 17-20% cost and 22-25% schedule savings from six AI levers on one multifamily project. Its own footnote calls the gains directional, not construction-validated.
On 16 September 2026, Suffolk published a white paper with the MIT Center for Real Estate and the MIT Media Lab City Science group titled Construction in the Age of AI: An Industry White Paper and Research Roadmap. Its headline finding is that six AI-enabled levers, applied together, "could generate approximately 17–20 percent total cost savings and 22–25 percent total schedule savings on the sample multifamily residential project analyzed." The same paper says, in a footnote under its lever table, that those gains should be treated as directional rather than construction-validated benchmarks. This article reports the two statements together. It does not report a measured jobsite result, because the paper does not contain one.
What is confirmed as of 18 September: the paper exists, Suffolk and MIT published it, and the percentages are the output of Suffolk's internal financial model run against one reference case, a multifamily development of about 180,000 square feet completed in 2024 after 4.5 years from initial design through commissioning and closeout. The footnote describes it as "a $180M multifamily San Francisco project." The paper does not identify the building, the owner or the design team beyond that. Our figures come from the paper itself, from Hugo Pegley's 18 September analysis at Construction Industry AI, which reproduces the tables and the footnote, and from Construction Dive's 16 September report. We have not audited the model.
What the model put on one project
The paper's table for the reference case sets a $179 million cost baseline against $32 million of modelled savings, which rounds to 18 percent. On schedule it sets a 51-month baseline against 11 months saved, or 22 percent. Those two numbers sit at the low end of the 17-20 and 22-25 ranges in the headline finding. Neither is a cost the contractor recorded or a month it recovered. Both are what the model says would have happened had the six levers been running on a job that finished in 2024 without them.
Sample multifamily project, baseline vs modelled savings, Suffolk/MIT white paper table, 16 Sep 2026 (dollars in $M, schedule in months)
18 September 2026. Hugo Pegley analysis. Page read in session. Reproduces the six-lever table with maxima, the $179M / $32M and 51-month / 11-month project table, the verbatim footnote, the permitting-drops-to-last Borda detail, the Volumetric Building Companies 47 percent figure, and the Fish and Scott quotes.
16 September 2026. Release page. Confirms publication date, partners, six levers, and the up to 20 percent cost / 25 percent schedule framing. Located in session via search; not opened.
16 September 2026. Matthew Thibault. Confirms the $180M San Francisco 2024 case, six levers, design automation as clearest upstream enabler, and next steps on studies and data standards. Located in session via search; not opened.
Free for construction companies.
One email a week on AI in use on real projects, from the contractors and people who ran it.
Here is the footnote in full, as the paper prints it: "Potential Efficiency Gains are estimated from a comprehensive review of existing literature of means and methods applied, in Architecture, Engineering, and Construction (AEC) and adjacent industries predominantly from early-stage pilots (so they should be treated as directional rather than construction-validated benchmarks), as well as means from survey respondents anchored to a $180M multifamily San Francisco project." Read plainly, the per-lever percentages were assembled from published pilots, some of them outside construction, and from what survey respondents expected, then applied to one project's budget and calendar. The paper describes its own contribution as among the first attempts to piece together disparate evidence into a directional view. That is a claim about the evidence base, and it is the reason the 18 percent should not be quoted without the sentence beneath it.
Six levers measured against different baselines
The paper attaches a maximum efficiency gain to each of its six levers. Design automation carries the largest figures, up to 39 percent off design cycle time and up to 21 percent off design and engineering cost. Offsite manufacturing is put at up to 32 percent off project timelines and 14 percent off total cost. Permitting is up to 24 percent off project time and 8 percent off cost. Scheduling is up to 18 percent off the project timeline and 11 percent off total cost. Skilled labor and subcontracting is up to 17 percent off project time and 13 percent off total cost. Supply chain and procurement is up to 15 percent off total project time and 10 percent off cost.
The design automation numbers are not comparable with the other five. They are stated against the design phase, its cycle time and its design and engineering cost, while the other levers are stated against total project time and total project cost. A 39 percent cut to a design phase is a different quantity from a 32 percent cut to a whole schedule, and the paper does not rank the levers by these columns. The figure below keeps the design rows separate for that reason.
Maximum modelled gain per lever, percent. Design rows use design-phase baselines; all others use total project. Suffolk/MIT via CI.AI, 18 Sep 2026
The paper also asked its expert participants to rank the levers, and it published the bias in its own instrument. Design automation ranks first on raw Borda scores with 206 points, "though 89 of those are self-votes from the design cohort." When each expert's votes for their own primary lever are stripped out, schedule optimization ranks first across every expert group with 163 cross-group points, and permitting drops to the least valuable lever of the six. The paper's project table notes that the total does not equal the sum of the lines because project phases overlap.
Where the paper counts instead of models
The permitting section leans least on the model because most of what it does is count. The United States has more than 20,000 independent permitting jurisdictions, each with its own zoning code, review process and approval timeline. Boston divides its 48 square miles into 429 zoning districts under a code of nearly 3,800 pages that has not been fundamentally revised since 1964, and 42 percent of the city's parcels are non-conforming under current lot-size rules. Against that, the paper's permitting lever is a modelled 24 percent off project time. It does not report a permit that cleared faster because a tool read the code.
The labor section cites an Associated Builders and Contractors analysis putting the skilled-worker gap at 456,000 additional workers needed in 2027, and models the labor and subcontracting lever at up to 17 percent off project time and 13 percent off cost. On offsite manufacturing, the paper reports that Volumetric Building Companies claims up to 47 percent schedule savings across geographies while cost savings stay market-dependent. That 47 percent is the manufacturer's own figure as the paper relays it, not a result the paper measured.
The return case and the condition attached to it
The paper converts the savings into developer terms. Coordinated application of the six levers, it says, could raise illustrative unlevered IRR on the sample project by approximately 5-6 percentage points, "for example from 15–20 percent to 20–25 percent." The paper calls the figure illustrative and says it assumes the levers can talk to each other, which in practice means a shared data layer that standardises outputs at every handoff. The levers themselves do not provide that layer.
Construction is at an inflection point. AI presents a real opportunity to transform the way we build. The most meaningful progress will come when the entire ecosystem aligns around better data, smarter workflows and shared accountability. - John Fish, chairman and CEO, Suffolk, white paper release, 16 September 2026
The next step is to keep building the data foundation needed to understand where AI has the strongest impact, where the limits still are and how those findings can be translated into better decision-making across real projects. - James Scott, co-lead, MIT Center for Real Estate, white paper release, 16 September 2026
Scott's sentence and the footnote say the same thing from two directions. The data that would turn a directional estimate into a construction-validated one does not yet exist in a form the authors could use. Construction Dive's report notes the authors pointed to further studies and better data standards as the next steps.
What to ask before you repeat the 18 percent
Ask whether the number is modelled or measured. Everything in this paper is modelled, and a vendor or owner who cites it as proof of a result has skipped the footnote. Ask which baseline the percentage is against, because a design-phase cut and a total-project cut share a percent sign and nothing else. Ask how many of the six levers were actually running on the job in question, since the paper's combined case assumes all six, connected.
Ask what pilots stand behind the lever figure, and whether they were construction pilots or borrowed from adjacent industries. Ask whose survey means went into the model and what cohort they came from, given that the paper's own ranking moved once self-votes were removed. Then ask for the one thing this paper says it does not have: a completed project where the savings were recorded after the fact rather than estimated before it.