Every project manager has seen the moment. A chart on the wall shows 80 percent complete, morale is high, and someone says "home stretch." And then something happens that no Gantt chart predicted: progress decelerates. Tasks that should close in a day drag on for three, and a punchlist that fit on one page becomes a binder. Progress reports stop naming dates and start saying "a few more weeks."
Now somebody has the numbers.
On June 25, Buildots launched the Intelligence Lab, a research hub offering free AI analysis of hundreds of projects spanning hundreds of millions of square feet worldwide, where head-mounted cameras capture site conditions continuously and machine vision tracks every activity across each project's lifecycle. What emerges is construction's first large-scale empirical answer to a question schedulers have argued about since critical path methods were invented: when does a project stall?
At the 80 percent mark, on every project the cameras tracked, every time.
That spread between "almost done" and "actually done" is structural. It is about the physics of finishing: trades compressed into tightening spaces, cascading dependencies where one late inspection holds three crews idle, punchlist items that take 20 minutes each but two weeks total because nobody can reach the ceiling until the fire-protection sign-off. Ask any foreman. Buildots confirmed with cameras what the industry has known in its bones for half a century.
What Else the Cameras Found
The stall point was not the only finding. In data center construction, planned weekly MEP output exceeds actual delivery by 20 to 50 percent. Half never arrives. And within those same projects, top-performing MEP teams work at three times the pace of average crews on identical scopes, which is a diplomatic way of saying that the gap between your best subcontractor and whoever answered the phone last Tuesday is wider than most schedules can absorb.
Schedule adherence varies wildly by building type. Healthcare projects hit planned milestones 65 percent of the time, which would embarrass most industries but represents the ceiling for construction. Data centers land at 57 percent, commercial and industrial projects settle in the low-to-mid 40s, and education projects finish dead last, under 39 percent. Schools. Every one. Less than two milestone hits out of five, on buildings every taxpayer assumed would be straightforward.
Why This Matters for Residential
None of this data comes from residential construction. That gap matters. Buildots tracks commercial jobs with full-time schedulers and million-dollar monitoring, oversight no residential contractor has or can afford. If a billion-dollar data center still sees 20-to-50-percent MEP gaps with AI cameras on every trade, what is happening where nobody is counting?
Census data gives us a hint: an average single-family home took 9.1 months from permit to completion in 2024, according to the Survey of Construction, and custom homes averaged 15.1 months.
What You Can Do With This
If you are building a custom home or managing a renovation with any complexity, ask your general contractor one question: what is the plan for the last 20 percent? Not the schedule, not the Gantt chart, but who coordinates punchlist sign-offs across trades, when final inspections get scheduled relative to the last finishing crew, and what happens when HVAC commissioning delays the flooring installer.
If the answer is "we push through it," that is not a plan but the absence of one, and it is exactly how a nine-month project becomes twelve.
Here is what works. Pad the final fifth of any schedule by 35 to 40 percent, not 10 percent, not "add a week." If your GC quotes eight months, budget the last two months as three.
Twenty years of watching timelines collapse has taught me one reliable truth about how buildings actually get finished, and cameras on hundreds of job sites just confirmed it with hard numbers I can finally cite in a meeting instead of just a tired gut feeling: the last fifth is where schedules go to die. Plan for it, or plan to live inside it.
What We Don't Know
Buildots has not published detailed methodology, including sample sizes per sector or confidence intervals, and findings may skew toward its client base of large general contractors like Turner and JE Dunn running projects that look nothing like a four-bedroom custom home where the schedule lives on a whiteboard and nobody is watching. Useful? Yes. Gospel? No.
Sources: Buildots Intelligence Lab launch (PR Newswire, June 25, 2026); ENR (June 26, 2026); U.S. Census Bureau Survey of Construction (2024); Physica A, "Activity delay patterns in project networks" (2024). Findings freely available at buildots.com.