You’re Paying $25,000 Per Home in Rework. A $500 Camera System Catches Most of It Before Drywall.

A framing crew on a 3,200-square-foot custom home in Raleigh installs a load-bearing header at 4x8 instead of the engineered 4x12 spec. Nobody catches it during the weekly site walk. Insulation goes in. Drywall goes up. Paint, trim, final punch list, certificate of occupancy. Seven months after move-in, the homeowner notices a crack propagating across the living room ceiling. Structural engineer confirms deflection beyond allowable limits. Remediation estimate: $43,000.

That header was visible for eleven days between framing completion and insulation. A 360-degree camera clipped to a hardhat would have captured it in a four-minute walkthrough.

Five Percent of Everything

Construction rework is the industry's most consistent line item that nobody budgets for. Liu et al. (2018) analyzed residential projects and found rework averaged 4.95% of total project cost. Mahamid (2024) put the range at 5-10%. The Construction Industry Institute's landmark study found it could reach 20% on troubled projects.

On a $495,000 new home (the U.S. Census Q4 2025 average), 5% is $24,750. At 10%, you are looking at $49,500 in work that gets done, torn out, and done again.

Where does it come from? PlanRadar's 2025 survey of 811 construction professionals across 13 countries broke it down: miscommunication causes 26% of all rework. Bad or inaccurate data causes another 14-22%. Design errors account for the rest, with the split between design-related (6.9% of cost) and construction-related (3.3%) documented by Cnudde as far back as 1991.

A number worth sitting with: companies running consistent QA/QC programs kept rework under 2% of project cost. Without consistent QA/QC, it climbed to 6-10% or higher. That delta on a half-million-dollar home is $20,000 to $40,000.

What the Cameras Actually Do

Three companies are building AI-powered visual inspection tools for construction. They work differently, and the differences matter.

OpenSpace ($168M+ raised, Series E) takes the simplest approach. Clip a 360-degree camera to your hardhat. Walk the site. OpenSpace automatically stitches your footage into a navigable capture, pins it to the floor plan, and creates a timestamped visual record of every wall, junction, and penetration. Their BIM+ overlay compares as-built conditions against the BIM model, flagging deviations automatically. Field Notes lets supers flag issues directly in the capture and assign them to responsible subs. General contractors like Nibbi and RG Construction use it on active projects.

Buildots ($45M raised at a $300M valuation, May 2025) goes further. Their hardhat-mounted cameras feed into an AI engine that automatically compares actual construction against the BIM model, identifying missing fire stops, incorrect pipe routing, insulation gaps, and mechanical conflicts. It integrates with Autodesk BIM 360. The system does not just document what was built. It tells you what was built wrong.

Doxel ($40M+ raised, led by Andreessen Horowitz) uses deep learning on HD photos and LIDAR scans. Autonomous rovers scan interiors after the workday, cross-referencing installed quantities against the budget and schedule. The company claims an 88% reduction in cost overruns on a medical center project. That number has not been independently verified, so treat it as aspirational until someone else confirms it.

Water Gets In, Money Goes Out

Water intrusion is the single most expensive defect category in new residential construction. Envelope failures at windows, flashing, and penetration points cause the majority, and they share one characteristic: invisible after drywall.

Pre-drywall inspection is the critical window. Once walls close, you are betting that your weekly site walk caught every flashing detail across 40+ windows and dozens of penetrations on a 3,000-square-foot home. In California, HOAs have a 10-year window to sue over construction defects. Insurance brokers estimate 80-85% of condo projects in the state eventually face defect litigation. The Terner Center at UC Berkeley found that defect insurance alone adds $8,000-$18,000 per condo unit. A comprehensive visual record captured during construction is the best legal defense you can buy, and it costs a fraction of one deposition.

A Calculation Nobody Published

Here is the math that should end the conversation for any residential builder doing ten or more homes per year.

OpenSpace starts at approximately $5,000 per year for small teams. A builder completing ten homes annually pays $500 per home for AI-powered visual capture and defect detection.

If that system reduces rework from 5% to 3.5% of project cost, a 30% improvement, each $495,000 home saves $7,425 in avoided rework. That is a 14.8× return on the $500 investment.

Even a conservative 20% reduction in rework saves $4,950 per home. At $500 per home for the tooling, that is still 9.9× ROI.

Inputs: U.S. Census Q4 2025 average new home price of $495,000. Baseline rework rate of 5% from Liu et al. (2018) residential-specific study. OpenSpace annual cost of $5,000 spread across 10 homes. Reduction percentages are estimated; no peer-reviewed study has measured the specific impact of AI visual inspection on residential rework rates.

Why Almost Nobody Uses This Stuff

Most AI QA/QC tools were designed for commercial and multifamily construction, where BIM models are standard and project values justify the overhead. Residential is different. Custom home builders rarely produce full BIM models. A set of 2D architectural drawings and a structural package is the norm. Without a BIM model, Buildots and OpenSpace BIM+ lose their best feature: automated deviation detection.

Photo documentation without BIM comparison is not really "AI." It is good practice with a nice interface. Valuable, sure. But the promise of an AI that catches your framing crew hanging the wrong header before insulation covers it requires a digital model to compare against. Most residential builders do not have one.

Production builders are moving. D.R. Horton, Lennar, and Toll Brothers have the scale and standardized plans to make AI QA/QC work. A custom builder doing eight homes a year in eight different configurations faces a harder sell. OpenSpace at $5,000 per year is reasonable above $5M in annual revenue. For a small remodeler running three $200K projects, $1,667 per job in QA/QC software is tougher math.

What You Should Actually Do

If you are a GC running $2-5M+ in residential projects: Start with 360-degree photo capture. OpenSpace or a comparable platform, approximately $5,000 per year. Capture every phase. Capture especially pre-drywall. Even without BIM comparison, the visual record alone creates a defensible timeline of every wall cavity, every window rough opening, every penetration seal. When a homeowner calls in month fourteen about water staining, you will have timestamped proof of what the envelope looked like before it was covered. If your rework rate is already under 2%, the tooling may not pay for itself in avoided rework, but it will pay for itself the first time you avoid a warranty dispute.

If you are building a custom home as an owner: Ask your builder whether they do pre-drywall photo documentation. If the answer is no, you are trusting that a once-a-week walkthrough caught every mistake in a four-month build. Ask for it. Offer to pay for it. At $500 per home, it is the cheapest insurance you will buy on a half-million-dollar project.

If you are a production builder doing 50+ homes per year: Evaluate Buildots or Doxel alongside OpenSpace. At your scale, per-home cost drops below $100, BIM models exist because your plans are standardized, and automated deviation detection catches systemic errors: a framing sub consistently installing headers one size down, an HVAC crew routing ducts through structural members, a plumber missing fire stops on every third penetration. Those patterns cost millions across a 200-home development. An AI that flags them in week two instead of month eight is worth multiples of its cost.

What This Analysis Cannot Prove

Nobody has run a proper study on whether AI visual inspection actually moves the needle on residential rework rates. My ROI numbers use estimated reductions (20-30%) extrapolated from PlanRadar's finding that consistent QA/QC drops rework from 6-10% to under 2%. Whether strapping a 360-degree camera to a hardhat delivers that level of improvement is genuinely unproven in the residential context.

Company claims need scrutiny. Doxel's 88% reduction comes from a single case study the company published about itself. Buildots and OpenSpace have testimonials, not controlled studies. That is standard for construction tech startups, but it means the ROI math above is built on estimates, not measurements.

Rework data is thinner for houses than for hospitals and office towers. Liu et al. (2018) at 4.95% is the strongest residential number, but from a limited sample. And my calculations use the national average of $495,000. In a market where median new builds run $350K, the dollar savings shrink. In San Francisco, they explode. Your mileage will vary by region, labor costs, and how complicated your local code is.