Construction has a productivity problem that predates the current wave of AI. For two decades, the industry's labor productivity grew at roughly 1% a year while the broader economy moved at nearly three times that — a gap McKinsey has documented for years. So when people ask where artificial intelligence will first make a real dent in how we build, my answer is simple: it starts before the first shovel hits the ground. It starts with the estimate.
At DALCORE, preconstruction is where we've watched AI go from interesting to genuinely useful. It's worth being precise about why — and about what it still can't do.
Takeoff is the clearest early win
The most mature use of AI in estimating is the quantity takeoff: computer vision reads the drawings, measures areas, and counts fixtures that an estimator used to trace by hand. A peer-reviewed study from the University of Kansas put one of these tools against traditional on-screen takeoff and measured roughly a 70% reduction in takeoff time, with quantities landing within about 5% of the manual baseline after an estimator reviewed the output.
That last part matters. The same researchers documented exactly where the AI stumbled — low-quality or scanned drawings, irregular geometry, MEP and specialty items, and anything tucked into a legend or a note. Their conclusion, and ours, is that AI takeoff is a powerful starting point, not a finished estimate. You'll see vendors advertise 98% accuracy and "5x faster." The independent evidence is still genuinely impressive — but it reads as "most of the way there, fast, with a person closing the gap," not "push a button and bid."
Why this matters more right now
The pressure behind all of this is people. In a 2025 AGC of America survey, 92% of contractors reported difficulty filling positions, and worker shortages were delaying projects at nearly half of firms. When experienced estimators are scarce and bid windows are short, getting a clean takeoff in a fraction of the time means we can pursue more work and respond faster — without cutting corners on diligence. It's no surprise that AI is now the top category where contractors say they're increasing technology spending.
The honest limits
Cost is harder than count. Measuring how much tile a bathroom needs is a very different problem from predicting what that bathroom will actually cost — in this market, with these subcontractors, on this schedule. The first problem is largely solved. The second is not. Pricing risk — reading a difficult site, an aggressive timeline, a client's real priorities — remains a human judgment, and I'd argue it always will.
"Garbage in, garbage out" also binds software more tightly than it binds people. A seasoned estimator can mentally correct for a missing dimension or a sloppy plan set; a model can't. That's why the firms getting real value from AI aren't the ones with the flashiest tools — they're the ones that kept clean, organized records of what their past projects actually cost.
And there's a liability question the industry hasn't fully answered. When an AI tool produces a six-figure error, the vendor's terms often cap their responsibility at a few thousand dollars. That gap lands on someone. Having spent years on the legal and compliance side of AI companies before founding DALCORE, I can tell you it's a question to settle in writing, up front — not after a problem.
Where this leaves us
AI hasn't replaced our estimators. It's changed what they spend their time on: less tracing lines on a PDF, more validating, pricing, and pressure-testing whether the numbers make sense for your project. The role is shifting from calculation to judgment — and judgment is exactly what you want standing between a drawing and a number you're going to build to.
That's the standard we hold ourselves to: use AI everywhere it genuinely sharpens the work, and stay honest about the places it doesn't. The estimate is simply where that payoff shows up first.
— Anthony D'Alessandro Jr., J.D.