AI is genuinely useful and improving fast. Every RFP response now includes an AI slide touting predictive risk dashboards, generative design renderings, schedule optimization, and estimating tools that promise faster, more accurate numbers. For a portfolio manager or site selection consultant vetting a general contractor, that slide is becoming as standard as the safety record and bonding capacity — and honestly, just as easy to misread.

Here’s the thing that actually separates contractors: not which AI platform they've licensed or built, but how many years of real project cost data they're running it against.

Large language models are converging fast. Any construction firm of reasonable size can now point ChatGPT, Gemini, or Claude at a set of drawings and generate a plausible conceptual estimate.

What isn't commodity is the ability to check that estimate against decades of what a building type actually cost, market by market, when it was real. AI is capable of generating a conceptual estimate early in a project. It's not as good at determining what the real number as the design progresses and more details become. We've watched a handful of outside vendors try to build estimating tools for GCs from scratch and they hit the same wall every time: no data, no traction. A model can produce a number. It can't tell a decision-maker which procurement strategy shaved six weeks off a comparable project's schedule last year, or which subcontractor packages on a given building type reliably generate change orders. That knowledge doesn't live in any training set. It lives in a contractor's own estimating databases and closeout reports. That’s exactly what determines whether a schedule commitment made in month one still holds in month eighteen.

For a corporate real estate team weighing time-to-market and cost certainty on a capital project, that distinction is the whole ballgame. An AI-generated estimate from a contractor with thin project history is a guess with better formatting. The same tool, layered over sixty years of actual outcomes, is a forecast.

Decision-makers should expect contractors to be protective of that historical cost data — and shouldn't read the guardedness as evasion. It's the same reason Coke and Pepsi don't publish their recipes: the ingredients aren't secret because they're complicated, they're secret because they're the business. A contractor with a real proprietary dataset has every reason to keep it out of a shared platform, even a compelling one. The useful due-diligence question isn't "will you show me the data" — it's "how deep is it, and how long have you been building it." There's a real difference between summary-level numbers and the depth that actually holds up: a detailed, line-item takeoff versus a rough guess at what a whole trade package should cost.

And this isn’t unique to general contractors, by the way. Utilities carry decades of outage and equipment performance history that predicts grid reliability better than any published capacity study. Site selectors themselves carry a version of this in thousands of confidential siting decisions that never became public record, including why a seemingly ideal market fell through. None of that shows up in a language model's training data, and none of it should be assumed just because a vendor says "AI-powered."

There's a labor dimension worth factoring in too. The contractors best positioned to make good on AI-assisted schedule and cost commitments are the ones using the technology to augment a stable craft workforce, not thin one out. We want people to see this as a career. An industry watching its skilled labor pool age out faster than it's replaced needs contractors who are using productivity gains to invest in training and optimize its workforce, not headcount reduction. In my experience, AI helps you do more with what you have. That says as much about a contractor's staying power as its tech stack, and it’s worth asking about directly.

None of this argues against AI-enabled contractors. I think it’s a pretty exciting time, actually. It argues for scrutinizing the claim the way any other capability gets scrutinized: not by the vendor logo on the estimating tool, but by what's underneath it. The really cool thing is, the contractors who have been building that foundation for decades are the ones best positioned to use this technology well. We’re closer to that than we’ve ever been.

And even with all of this, it's still a relationships business. You could automate the outreach — have AI draft every follow-up email, schedule every check-in. But that's not the point of any of it. The technology should free people up to do more of the things that don't scale: the handwritten note, the phone call instead of the email, showing up in person when it matters. That's not a nice-to-have. That's the business.