I had a call last Thursday with a GC who runs about 40 jobs a year out of a small office near the Heights. Halfway through, he stopped me and said: “Every vendor that pitches us this year says they have AI. I’ve sat through twelve demos and not one of them showed me anything different than the permit search I already pay $200 a month for. Is this real?”
Fair question. The honest answer is that most of what gets sold as “construction AI” in 2026 is window-dressing on a database. The real stuff is rare, but it does exist—and it’s starting to change which contractors win bids and which ones spend their week chasing dead leads.
Here’s how I’d separate the two.
What “AI” usually means right now
If you scroll through the construction tech category on Capterra or G2 today, you’ll find roughly forty products with “AI” in the headline. In practice they fall into three buckets:
- 01 ·A chatbot bolted to a permit database. You ask it “Show me roofing permits in 77019 over $50K” and it returns a SQL query result. That’s not AI. That’s a search box with a casual outfit.
- 02 ·A summarizer. The product takes a long permit description and produces a one-sentence version. Useful, sometimes. Worth its own product line, no.
- 03 ·An autocomplete on email outreach. Generates the cold-outreach copy you would’ve written anyway, in a voice that sounds slightly more like a LinkedIn comment than your own.
None of these things are bad. They’re just incremental. They don’t change the shape of your week.
What real project intelligence looks like
The thing that actually changes how a GC operates is something narrower than “generative AI” and much more useful: a system that watches every signal across your market continuously and tells you which ones matter to your business specifically.
That’s a different problem than search. Search assumes you know what you’re looking for. Project intelligence assumes you don’t—and that the work worth bidding on is buried in a thousand records you’d never have time to read.
Real intelligence does four things at once:
- Joins data the city publishes in fragments. A permit lives in one system, the plan-review status in another, the plat in a third, the owner in a fourth. Stitching those together is genuinely hard. Most tools don’t.
- Scores every project against your business. Not against “the market” in the abstract, against you—your trade, your zip codes, your valuation bands, your recent wins. The score should be defensibly explainable, not a black box.
- Surfaces signal as it lands. If a plat application drops on a Monday, you should know about it Monday morning, not when the permits come out eight months later.
- Connects you to the decision-maker. An address is not a contact. The contact is whoever is on the application, the LLC behind the parcel, the developer’s VP of construction. AI that surfaces a name and a way to reach them is doing real work.
The signal-to-noise problem is the whole problem
Houston publishes about 25,000 to 30,000 permits a year just inside the city limits. Add Harris County, plus the surrounding metros (Fort Bend, Montgomery, Galveston), plus the plats heading to Planning Commission every two weeks, plus the plan-review queue—and a serious GC has tens of thousands of potentially relevant signals to evaluate.
No one reads them all. The GCs who win more bids aren’t reading more. They’re reading the right ones. The thirty or forty per week that actually fit their trade, their territory, their crew capacity.
That filtering job is where AI earns its keep. Not generating prose. Not summarizing PDFs. Reading 25,000 records and handing you the 30 worth a phone call.
What it can't do (yet)
I want to be honest about where the technology is. AI is not going to:
- Cold-call a homeowner and convince them to sign with you. That’s still you.
- Replace the relationship between a GC and a developer they’ve worked with for fifteen years. It can put new developers in front of you; it can’t fast-forward trust.
- Read between the lines on a project that’s permitted but politically dead. Local context—which projects will actually break ground—is a human judgment call.
- Predict what your competition will bid. There are companies pretending to do this. They’re bluffing.
What it can do is buy you back the four to six hours a week you’re currently spending on the wrong half of your funnel.
How to think about ROI
If you’re running a $4M–$30M revenue GC in a major metro, the math on AI lead intelligence is unusually simple. One won bid that you wouldn’t have heard about pays for the platform for the year. Two won bids and you’re funding next year’s software stack from the margin.
That’s why the smart contractors I talk to don’t evaluate this category like they’d evaluate accounting software. They evaluate it like they’d evaluate a senior estimator. The right tool isn’t the cheapest one—it’s the one that pays for itself out of one project.
What to ask in your next demo
If you’re sitting through demos this quarter, save yourself an hour by leading with these three questions:
- 01 ·Show me a project I’m already tracking. If they can’t, the index is shallow.
- 02 ·How does your scoring work? If it’s a black box (“our proprietary algorithm”), it isn’t ready for production. If they can describe the inputs in plain English, that’s a green flag.
- 03 ·How fast does new signal land? The honest answer is hours, not minutes. “Real-time” is rarely real-time. But “once a week” is too slow for plats and plan reviews—that’s where you lose the lead-time advantage.
If you’d like to see what a properly-stitched, scored, Houston-and-beyond market looks like—with the data join, the contact resolution, the daily inbox brief, all of it—that’s why we built Platineer. No long demo cycles. No prose generators dressed up as products. Just the leads that actually match your business, every morning.