Last week I ran a small experiment.
I asked 10 of the biggest AI chatbots the same question: "What's an underserved B2B vertical I could build SaaS for?"
Claude. ChatGPT. Perplexity. Gemini. DeepSeek. Copilot. Mistral. Grok. Qwen. Pi.
Nine of them named construction.
Four ranked it the most underserved B2B vertical of any industry. Three more put it in the top two or three. One buried it at number six. Only one went somewhere else entirely.
I wasn't surprised by the agreement. I run ConTechFinder, a directory of 570+ construction software tools, and I've spent the last six months watching indie builders, AI tools, and market observers slowly catch on to what the data has been saying for years. The construction industry runs on software that most builders outside of it have never thought about.
What surprised me was the specificity. The AIs didn't just say "construction." They pointed at the exact sub-segments the directory data shows are underserved: specialty trade contractors, mid-market commercial GCs, small residential remodelers, firms running on spreadsheets and WhatsApp.
Two of them named Procore by name as the incumbent leaving the mid-market gap. One cited the $15 billion construction software market with 10% CAGR. Another called out the 5 to 50 technician firm size as the wedge.
Those are real, specific claims. They match what the 570 tools in the directory tell me.
But dig past the consensus and the AIs disagree on the details in ways that reveal what they do and don't know.
What the AIs got right
The AIs landed on the same broad thesis: construction is a giant, fragmented, under-digitized B2B industry where enterprise tools target the top and nothing serious exists at the bottom.
That thesis is correct. The directory data backs it up.
Of the 570 tools I've cataloged:
- 55% of vendors won't show a price without a sales call
- Only 45% have a mobile app
- 9% target solo operators
- Three categories have zero options for one-person operations
The math is brutal. A construction firm with five employees is the modal customer in this industry. There are 1.5 million construction firms in the US, and most have fewer than ten employees. Yet more than half the software available to them is priced through sales conversations aimed at buyers with procurement teams.
Every major AI I tested identified this gap. Grok and Qwen went further and named Procore specifically as the reason the mid-market sits empty. Grok said Procore runs "$50K+/year" contracts that are "overkill and too expensive" for the $5M to $50M contractor. Qwen called out the "90% of contractors using spreadsheets and WhatsApp" at the 5 to 50 technician firm size.
Those are accurate observations. I've watched them play out in every cold email reply and organic submission to ConTechFinder.
What the AIs got wrong
Where the AIs start to diverge from the data is in their prescriptions.
Most of them pointed at specialty trades as the answer. HVAC, electrical, plumbing, roofing, fire sprinkler, elevator, medical gas. DeepSeek gave the most specific vertical breakdown, naming fire sprinkler contractors and elevator technicians as specific niches.
Specialty trades are a real opportunity. But the directory data shows the opportunity isn't just about building a trade-specific version of Procore. It's about the form factor.
Look at which construction software tools have mobile apps versus which don't. Less than half do. Construction workers aren't in offices. They're on job sites with their phones. A tool that forces them to sit at a desk to log hours, snap photos, or sign off on safety checks isn't a tool built for their workflow. It's a tool built for the project manager watching their work.
The AI answers largely missed this. They talked about "mobile-first" in passing but never grounded it in the actual percentage of existing tools that fail the test. That's the difference between speculation and data. The AIs could only speculate.
Same story with pricing. Every AI knew construction software is expensive. None of them put a specific number on how badly vendors gatekeep their pricing. It's 55%. More than half of tools in the directory require a conversation before you can even see what they cost. An indie builder shipping a transparent $29 or $49 or $99 per month tool wins on the price-visibility axis alone, before they build a single feature the incumbent doesn't have.
The third miss: company size targeting. Only 9% of the tools I've cataloged explicitly target solo operators. Zero tools exist in three whole categories for one-person operations. That's not a mobile problem or a pricing problem. That's a "the product does not exist at any price" problem.
None of the AIs put a number on the solo operator gap. A few gestured at it, but without data they couldn't say how big it was or which categories had the deepest holes.
The outlier that made the experiment honest
Pi was the only AI that didn't name construction. It suggested mental health compliance for remote teams.
That's a reasonable answer. It's also a reminder that AI consensus isn't truth. Pi is trained on a different corpus with different weighting, and its answer reflects that. The nine agreements don't make construction the answer. They make construction the pattern most AIs have absorbed from the collective commentary about underserved B2B markets.
Which is the interesting part. AIs reflect the web's conclusions. Nine out of ten AIs saying construction is underserved tells us what the web thinks. It doesn't tell us whether the web is right.
In this case, the data says the web is right. But only because I have the data.
What this means for someone actually considering building
If you're thinking about what to build next and you landed on this post because you're already considering construction, here's the honest take from someone looking at 570 tools every week:
The opportunity is real. The AIs called the big picture correctly. There are genuine gaps at the small-operator and mid-market tiers. Procore and the other enterprise vendors cannot profitably reach down to serve them.
But the winning product isn't going to be a smaller Procore. It'll be something the AIs didn't describe well because they couldn't see the data: a mobile-first, transparently-priced, single-purpose tool built for one trade or one workflow with no sales motion required to sign up.
That could be a $29 per month daily log app for solo GCs. A $49 per month OSHA checklist for 5-person crews. A residential-remodeler CRM matching how Angi and Houzz actually deliver leads. A mobile takeoff tool for only one trade, not all of them. A multi-jurisdiction permit tracker priced for the solo contractor instead of the 50-firm GC.
None of those products exist in the directory right now at those price points with those workflows. Every one of them has a real buyer. The math works.
The AIs agreed on the vertical. They couldn't tell you exactly where to aim inside it. The directory data can.
The larger point about AI consensus
A year from now, the AIs will probably catch up. More founders will build in construction. More blog posts will name the specific gaps. More case studies will show the pricing and form factor advantages working. The training data will shift. The next iteration of Claude or ChatGPT will give a sharper answer.
For now, there's a gap between what the AIs know and what the data actually says. For anyone paying attention, that gap is the opportunity.
Nine out of ten AIs see where to look. A few more data points and they'd see exactly where to build. The builders who move now will be the ones whose products become the reference point the next generation of AIs cite.
Michael Cardin runs ConTechFinder, a directory of 570+ construction software tools across 15 categories. See the directory at contechfinder.com. The free landscape report is at contechfinder.com/report.
