What the Suffolk and MIT AI Study Misses About the Carry Line

A project misses its completion date. The construction loan stays outstanding, general conditions continue, stabilization moves right, and the extension conversation begins.

The Suffolk and MIT white paper models how artificial intelligence (AI) could change project delivery across commercial real estate and construction. For one reference project, the directional model estimates a 17-20% cost opportunity and a 22-25% reduction in overall project duration. It also illustrates an increase of 5-6% in unlevered internal rate of return (IRR).

The study prices AI at the cost line and the schedule line. On a financed project the bill arrives on the carry line.

Nobody asked the lenders’ data. So here it is. It says finishing on time on a financed project is close to a coin flip, and the other side of the coin is ninety days late.

The Study Models What One Project Could Have Done

The model uses one 180,000-square-foot multifamily development in San Francisco, completed in 2024 after 4.5 years from design through closeout. Construction Dive’s September 16, 2026 coverage describes the project as roughly $180 million and the analysis as retroactive.

The paper studies the following six connected levers:

• Design automation
• Offsite manufacturing
• Permitting
• Scheduling
• Skilled labor and subcontracting
• Supply chain and procurement

Across those levers, Suffolk’s integrated model estimates a directional 17 to 20 percent cost opportunity and a 22 to 25 percent reduction in overall project duration. Its modeled economics illustrate an increase of 5 to 6 percentage points in unlevered IRR. The paper notes that the phase lines do not sum to the total.

The work connects research to practical development decisions. James Scott, co-lead of the MIT Center for Real Estate, said in the September 16, 2026 Business Wire release:

“This study is important because it starts to connect academic research with practical industry evidence. 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. That kind of evidence base is essential if the industry wants to move from promising examples to durable, repeatable progress.”

I agree. Outcomes across several hundred projects are the right next step. On the finance side, part of that data already exists, on more than ten thousand projects, and it shows something a model of one project cannot: the loss is not spread across projects, it sits in a tail.

The Ledger Shows a 90-Day Tail

Built’s March 2026 analysis covers 10,352 completed commercial real estate (CRE) construction projects financed on the platform. The projects were activated from 2021 through 2024 and span CRE subtypes with a lender draw process. Durations above 1,800 days were excluded.

A project counts as on time when it finishes within its original construction schedule plus a 14-day closeout allowance, or the current schedule where the original is unavailable. Completion is the construction-complete status change or the payoff date as a proxy.

By that definition, Built’s analysis found that 42.6% finished on time. It is a baseline, not a result. It says nothing about what AI does and nothing about what Built does.

By commitment size, the rate was 44.0% below $1 million, 42.0% from $1 million to $5 million, 41.8% from $5 million to $20 million, and 43.9% at $20 million and above.

The distribution shows where the risk sits:

• 6.5% ran 1 to 30 days over.
• 9.7% ran 31 to 90 days over.
• 41.2% ran more than 90 days over.

You either finish, or you go ninety days over. There is not much in between.

The full-sample median was 36 days over. Among late projects, median lateness ranged from 152 days below $1 million to 244 days at $20 million and above. That is the 90-day tail.

Original schedule dates cover about 65% of the sample. The fallback uses the current end date, which may have been extended and may inflate the on-time rate.

The study models one design-build multifamily project. Built’s ledger covers a separate population of lender-financed CRE projects. They add context without becoming matched populations.

The Carry Line Turns Time Into Cost

Design, scheduling, procurement, and permitting decide what a project costs and how long it takes. The finance layer decides what a slip costs and who sees it first.

The construction loan stays outstanding beyond the planned permanent takeout. General conditions continue, stabilization and revenue move right, and extension fees may apply.

Suffolk’s Table 1 assigns a $7 million modeled finance baseline and $1.2 million in modeled finance savings to its reference project. A separate total schedule line estimates 11 months saved from a 51-month baseline, or 22%. In the model, the finance saving comes from the shorter schedule.

Built’s estimate at current rates is about $580 per day per $1 million of commitment. Swap in your own assumptions. The estimate uses the following:

• An 8.5% construction rate against an assumed permanent rate of about 6.05%.
• General conditions at 10% of commitment annually.
• Delayed stabilization at 75% loan-to-value and a 5.5% capitalization rate.
• Extension fees at 0.375% per 30 days.

Under Built’s current-rate estimate, a $10 million loan costs about $174,000 at 30 days and $522,000 at 90 days. At 180 days, the estimate reaches about $1.04 million. For a $25 million loan, the estimates are about $435,000 at 30 days and $1.305 million at 90 days.

Built’s analysis estimates a median carry cost of about $305,000 among late projects. Medians lead because the $1.9 million mean is driven by a tail of severely late projects. The delayed-stabilization component may overstate carry for non-income-producing CRE.

That is what 5 to 6 points of IRR is made of on the ground. The study’s IRR figure combines cost, duration, risk, and revenue timing, and the model’s finance saving is a schedule saving. This is what a schedule saving is worth per day on a financed project.

Built provides portfolio-level schedule and draw visibility for commercial real estate and construction finance teams, connecting project information with the financial workflows developers and lenders already use.

A Tail Is Something You Can Watch For

Developers should take the study’s execution levers seriously and put the carry line into the pro forma. Use the median late duration for the relevant commitment size instead of zero. Treat the modeled 22 to 25 percent reduction in overall project duration as the prize for reducing schedule risk, rather than an underwriting assumption.

Lenders should ask which active loans are moving toward the 90-day tail and when that path should become visible in the file. Put that question in the approval memo and monitoring plan. Inspection cadence belongs in that discipline, as this analysis of construction loan inspection frequency and credit risk explains.

Built is a real estate and construction finance platform spanning residential, commercial, homebuilder, and private credit, with more than 300 lenders active on it.

The model tells you what a project could have done. A ledger tells you what ten thousand did. The industry needs both, and the second one has been sitting in lenders’ systems the whole time.

See the Finance Side of AI in Commercial Real Estate and Construction

See how Built connects project and draw information across commercial real estate and construction finance portfolios. Book a demo today.

AI in Commercial Real Estate and Construction Finance FAQs

What percentage of commercial construction projects finish on time?

Built’s March 2026 analysis found that 42.6% of 10,352 completed lender-financed CRE projects finished within the original construction schedule plus 14 days, or the current schedule where the original was unavailable. The figure applies only to Built’s sample. Original schedule dates cover about 65%, and the fallback may inflate the rate.

How much does a construction delay cost per day?

Built estimates about $580 per day per $1 million of commitment using its stated current-rate assumptions. Developers and lenders should substitute their own rates, fees, project economics, and general-conditions assumptions.

Can AI cut costs and schedules in commercial real estate and construction?

Suffolk’s model for one reference project suggests a directional 17 to 20 percent cost opportunity and a 22 to 25 percent reduction in overall project duration across six connected levers. These are modeled estimates for that project. The paper calls for broader outcomes data and controlled comparisons.