I’ve spent a lot of time at construction conferences over the years. Enough time to notice the patterns. Every major event has a new take on the future of AI in preconstruction and, specifically this year, they all seem to start at AI-powered takeoff.
Faster takeoffs, automated quantities, estimates generated in minutes instead of days; it’s an exciting story to tell, and it’s not exactly wrong.
Quantity takeoff will keep getting faster and more accurate. AI is genuinely good at that kind of pattern recognition.
But I worry that the industry is starting to confuse faster inputs with better decisions.
The irony is that the firms seeing the biggest gains from preconstruction aren’t the ones producing estimates faster. FMI’s State of Global Preconstruction report recently proved that organizations with mature preconstruction practices are 52% more likely to outperform their peers in profitability because they make better decisions around scope, cost, and risk – not because they count quantities faster.
Preconstruction on a complex project can run for months, sometimes years. The time spent waiting on quantities is a very, very small slice of that pie; a day at best. The bulk of the time spent (and wasted) is spent validating scope, reconciling design intent with what the site and budget can actually support, and getting an owner to a number everyone can live with. Quantities are an input to that process, but they are not what governs the pace. (I’d argue that even a tenfold improvement in takeoff speed wouldn’t meaningfully shorten a typical preconstruction schedule.)
The real reason is structural, and it requires that we zoom out and look at how our construction system is currently set up. In the U.S., contractors aren’t paid for quantities produced. They’re paid to deliver a defined scope for an agreed price. Quantities are largely objective: hand the same drawings to five estimators, and you’ll get similar numbers, give or take some interpretation at the margins.
Cost is an entirely different animal. It depends on labor availability, market conditions, means and methods, schedule constraints, and a dozen judgment calls that two equally competent estimators could make differently and both be right. Owners are buying outcomes, and outcomes are priced, not measured.
This distinction matters because it points to where preconstruction actually loses time. It’s rarely a quantity problem. It’s scope clarification, design coordination, constructability review, and cost reconciliation, over and over, as the budget gets revised. Industry research puts the average at six or more budget revisions per project. And each revision reopens the conversation about cost, not quantities.
So when AI vendors pitch faster takeoff as the fix for slow preconstruction, they’re optimizing a step that was never the constraint. It’s a bit like installing a faster elevator in a building where the real delay is the security line at the front door.
AI has made us an efficiency promise. So what would it look like to make preconstruction itself more efficient — not just the tasks within it?
Here’s the harder problem, and the one I think deserves the attention that takeoff speed is currently getting: most firms don’t actually have usable cost data. In a recent industry poll, close to 40% of respondents said their cost data lives scattered across shared drives and folders, and another third said it’s spread across multiple estimating tools that don’t talk to each other. Barely anyone had it in a single structured system. This should not be interpreted as a failure of preconstruction teams. Estimators have effective methods and systems that work for them individually. Rather, it’s a sign that the industry has never agreed on what cost data actually needs to be captured to become useful beyond the project it came from.
Our numbers alone don’t count as cost data. Management thinkers have argued for decades that data without context has little value. Construction cost data is no different. A unit price isn’t knowledge. It’s the final output of dozens of decisions: assumptions about labor, logistics, sequencing, procurement strategy, market conditions, and risk.
Real cost data captures the decision that was made, the context behind it, the cost impact, and what actually happened once the work was built. Capture that consistently across projects, and you start building something an AI system can reason with: a system that can say a particular facade substitution saved a specific amount on three comparable jobs last year, instead of producing the kind of generic, occasionally wrong answer we’ve all learned to expect from a general-purpose model with no access to your actual project history.
AI is only as useful as the data beneath it, and most preconstruction data isn’t well structured enough to feed it. Fix that first, and AI has something real to work with. Skip it, and you’ve bought a faster way to arrive at the same slow, disconnected decision-making process.
None of this is an argument against AI in preconstruction. We’ve needed better systems for a long, long time and the potential AI has to drastically improve speed and efficiency is too significant to overlook. It’s also too promising to point it at a marginally insignificant problem…
My argument is one that any builder would make: the foundation should come first, because AI is only as powerful as the data that runs beneath it.
The firms that will actually pull ahead over the next few years aren’t the ones with the flashy AI takeoff demo. They’re the ones treating cost decisions as data worth tracking in the first place: capturing the trade-offs, the assumptions behind them, and the outcomes, project after project. Those firms will walk into an owner meeting and answer a scope-versus-budget question on the spot, with real precedent behind the answer, while everyone else is still saying they’ll get back to them after the estimate is updated.
Preconstruction needs a memory. Once we build that, this AI conversation gets a lot more interesting.
About the Author
Dustin DeVan is CEO and co-founder of Ediphi, a preconstruction platform for cost modeling, estimating, and contracting, and the founder of BuildingConnected, acquired by Autodesk in 2018.
