The most dangerous number in residential construction is the one nobody checks.
Every custom home built on a construction loan runs on a draw schedule. The builder submits a pay application claiming a certain percentage of the work is complete, the bank releases funds, the homeowner's loan balance climbs, and the only evidence behind that percentage is the builder's own word, scrawled on an AIA G702 form or something that looks like one, signed by a guy who has every financial incentive to round up.
I've watched this play out for twenty-three years. Same story, different zip code. The framing is "90% complete" when the second floor hasn't been sheeted. The rough-in is "done" when three plumbing stacks are still missing. Nobody catches it until the project stalls and the lender sends someone to figure out where $380,000 went.
How Draw Schedules Actually Work
On a typical $500,000 custom home, the bank structures five to seven draws tied to milestones: site work and foundation, framing, mechanical rough-in, insulation and drywall, trim and finishes, final. At each milestone, the general contractor submits a pay application. The bank dispatches an appraiser or inspector who visits the site for fifteen, maybe thirty minutes, checks broad categories against a checklist, and confirms the work looks roughly consistent with the claimed percentage.
Roughly consistent, not measured, not verified against the plans, not compared task-by-task to the schedule. The inspector is looking at the skeleton and saying yes, this looks like a house that's about 60% built, and the bank wires $300,000.
Between those milestone inspections, the builder's self-reported percent-complete is the only number anyone sees, which means a homeowner with no construction experience, no independent verification mechanism, and nothing but trust and a stack of receipts is signing off on six-figure wire transfers based on a single column in a spreadsheet that the person receiving the money filled out themselves. And here's the part that nobody in the industry likes to talk about: the builder controls both variables in the percent-complete calculation. They report the costs incurred, and they set the total estimated cost. Divide one by the other, and you get a number that is exactly as honest as the person doing the division.
The 80/20 Problem Has a Name
Industry data from the Birmingham Group shows that 80% of cost overruns occur during the final 20% of project completion. Late-stage complexity is the standard explanation: scope changes pile up, inspections fail, finishes get delayed, trades stack on top of each other in compressed timelines, and all of that is real.
But there's a simpler explanation that accounts for at least some of those overruns, one that nobody puts in the conference presentations: the first 80% of the budget got burned when the project was only 65% done, because the percent-complete numbers were inflated from the start, and by the time anyone noticed, the money was gone and the house wasn't.
This isn't hypothetical. It happens. Arizona's Attorney General issued a warning in April 2026 about a growing pattern of construction fraud targeting ADU projects specifically. Contractors collecting large deposits, never pulling permits, never breaking ground. Families losing up to $250,000. It's the same mechanism that operates at smaller, legal-but-corrosive scales on legitimate custom home projects every day: the builder's number is the only number, and nobody verifies it until the money is spent.
A Camera on a Hard Hat Changes the Equation
OpenSpace, a construction technology company that has captured more than 40 billion square feet of project data, makes a 360-degree camera system that mounts on a hard hat. Walk the site in ten minutes, and the Vision Engine maps every photo to the project plans automatically, creating a time-stamped Google Street View of the entire job. Their progress-tracking module compares the visual record to the BIM model and the schedule, task by task, and reports verified percent-complete with 95% accuracy.
Ninety-five percent verified accuracy. That's not a self-reported number on a form. It's a human-in-the-loop AI system cross-referencing what it sees in the photographs against what should be there according to the plans, with a trained analyst confirming every output before it reaches the project team.
Buildots offers a similar system and claims 90% or better task recognition on complex commercial sites, with automated daily reports that compare actual conditions to the planned schedule at the individual-task level. If the plumber was supposed to rough-in the second-floor bathrooms by Tuesday and the camera shows bare studs, the dashboard says so, and it says so to everyone with access, not just the general contractor.
Cleveland Construction, a general contractor, deployed OpenSpace across their entire portfolio after testing it on a single project. Cleveland Construction's senior VP of operations, Elliot Christiansen, put it plainly: the data was accurate enough that they could have effective two-way conversations with field teams and get answers quickly, without having to "play twenty questions with them like we used to."
The Residential Gap
Here's the problem. Both OpenSpace and Buildots are commercial-scale platforms. Their pricing and onboarding are built for general contractors running $10 million hospital wings and 200-unit apartment complexes, not a homeowner building a $600,000 house in the suburbs. Monthly costs for commercial implementations run $3,000 to $5,000 per project, which is reasonable when the project budget is eight figures. On a half-million-dollar custom home, that's a harder sell.
But the technology doesn't care about project size. A 360-degree camera captures a 3,000-square-foot house in under five minutes. AI comparison works the same whether it's matching against a commercial BIM model or a set of residential plans. The barrier isn't technical. Nobody has packaged and priced it for the market that needs it most: the individual homeowner who has no construction expertise, no independent verification mechanism, and a six-figure loan balance climbing with every draw.
California law already says contractors can't collect payment for work they haven't completed or materials they haven't delivered, and the state caps down payments on home improvement contracts at $1,000 or 10%, whichever is less. These rules exist because the problem is real and old, and the technology to enforce them in real time exists now, but the pricing model for residential doesn't.
What You Can Do Today
Until residential-priced AI verification arrives, homeowners building custom homes can negotiate three protections into their contract before signing:
First, require time-stamped photo documentation at every draw milestone, and not a handful of flattering angles the builder selects but comprehensive 360-degree coverage of every room, every elevation, every rough-in area. Consumer-grade 360 cameras cost $300 to $500, and making the photos part of the pay application package costs nothing beyond the effort of carrying the camera.
Second, tie your draw schedule to specific, measurable tasks instead of subjective percent-complete estimates, because "framing 80% complete" is an opinion, while "all exterior walls framed and sheeted, all interior bearing walls framed, roof trusses set, roof deck installed" is a checklist that either matches reality or doesn't. Percentages invite interpretation; checklists don't.
Third, hire your own inspector, and not the bank's appraiser who spends fifteen minutes and checks a box but an independent building inspector or construction consultant who visits the site before you authorize each draw and confirms that the claimed tasks are actually, physically complete. At $300 to $500 per inspection on a half-million-dollar build, that's insurance at 0.5% of the project cost.
Limitations
AI progress tracking isn't omniscient, and photo-based systems can verify what's visible but can't see behind finished drywall to check whether the plumbing was pressure-tested or the electrical was properly secured. OpenSpace's 95% accuracy figure is impressive, but on a 200-task project, that's still 10 tasks the system might misclassify, and while human review closes most of that gap, the number isn't zero.
More importantly, most residential builders are honest, reporting progress accurately and managing their draws responsibly and delivering the house. The problem isn't widespread fraud but rather that the system provides zero verification friction for the ones who don't, and the consequences fall entirely on the homeowner who trusted a number that nobody checked.
Twenty-three years of watching projects go sideways taught me one thing the hard way: the most expensive mistakes in construction aren't the ones you can see. They're the ones hidden inside a percentage on a pay application that everyone assumed was accurate because no one had a reason to count.
Now something can count. The question is whether anyone will package and price it for the person standing in an unfinished kitchen, writing checks against a number that exists only because someone with a clipboard said so.