Your Contractor's Lowest Bid Was Subsidized by Workers Who Don't Exist on Paper. The AI That Catches Payroll Fraud Can't See Them Either.
A carpenter named Carlos got his start hanging drywall in the non-union residential housing market in Massachusetts. When payday came, there was no check stub, no direct deposit, no W-2 on file. "They would either come to our house or we'd go to their house or meet somewhere," he told researchers at UMass Amherst's Labor Center. "It would be the most underground thing ever. Always cash." When asked if he ever saw the labor broker who hired him on the actual jobsite, Carlos said: "Never, they're never there."
Carlos is one of an estimated 1.1 to 2.1 million construction workers in the United States who are illegally misclassified as independent contractors or paid entirely off the books, according to a national analysis published by The Century Foundation. These workers have no workers' compensation coverage if they fall off a roof. No unemployment insurance if they get laid off between projects. No employer contributions to Social Security. No overtime protections. No access to workplace safety training, employee assistance programs, or the mental health resources that might matter in an industry where the male suicide rate runs 65.6 per 100,000, highest of any major occupational group in America, according to the CDC's 2021 National Vital Statistics analysis.
Meanwhile, a parallel industry has emerged selling AI-powered payroll compliance tools that can detect exactly this kind of fraud. GPS-verified timecards. BLE beacon attendance systems. Computer vision that watches job sites and flags behavioral hazards. Predictive analytics that identify which workers' comp claims will escalate before the third medical visit. These tools work. They reduce fraud. They save lives. And they are almost entirely invisible to the residential construction sector where the fraud is worst.
Where the Fraud Concentrates
Every academic study that has examined the geography of construction misclassification reaches the same conclusion: residential work is the epicenter. A UMass Amherst analysis of Massachusetts Department of Unemployment Assistance audits found that one in six construction employers was misclassifying workers between 2017 and 2019, and that the problem was "especially concentrated in residential construction" and in a specific cluster of trades: carpenters, laborers, painters, roofers, siding contractors, framers, drywall hangers, and finish carpenters. Indirect estimation methods suggested that 9.5 to 15.8 percent of the entire Massachusetts construction workforce was engaged in a fraudulent employment relationship. Annual cost to the state: more than $100 million in lost unemployment insurance revenue, workers' compensation premiums, and income tax.
Texas is worse. A Workers Defense Project survey found a 41 percent misclassification rate on construction jobsites across the state. Austin alone clocked 38 percent. A 2021 Catholic Labor Network study surveyed workers on large commercial and public construction sites in Washington, D.C., and found that 47 percent were paid in cash or via check without payroll deductions. Not on residential sites, where oversight is lighter. On commercial and public projects, where it is supposed to be heaviest.
Nationally, The Century Foundation estimates that unscrupulous construction employers underpay workers and shortchange legally required benefits by more than $12 billion per year. Taxpayers absorb $5 to $10 billion annually in lost revenue to Social Security, Medicare, unemployment insurance, and workers' compensation funds. A UC Berkeley Center for Labor Research study found that 39 percent of construction worker families are enrolled in at least one public safety net program, costing taxpayers $28 billion per year. That $28 billion is not a housing subsidy. It is a subsidy to employers who do not pay their workers enough or classify them correctly enough for those families to survive without public assistance.
What the Technology Can Do
CompScience, a computer vision company that analyzes existing workplace camera feeds using AI, recently partnered with Nationwide Insurance and Swiss Re to offer workers' compensation policies informed by real-time hazard detection. Their platform identifies more than 50 behavioral and environmental risks from video footage: missing hard hats, improper lifting, fall hazards, proximity to heavy equipment, unguarded openings. Nationwide examined two years of actuarial data and confirmed the system reduced workers' comp claims by up to 23 percent. John Lopes, SVP of product expansion at Nationwide, said the platform "provides truly actionable insights into workplace risks."
Workyard sells GPS-verified time tracking designed specifically for construction. Every clock-in and clock-out is stamped with coordinates, eliminating the paper timecards that make wage theft and misclassification easy. Their system automatically applies union pay rules, handles prevailing wage classifications, distinguishes travel time from work time, and generates audit-ready digital logs. Walsh Construction deployed Eyrus BLE tracking on the El Centro Affordable Housing project in Seattle, passively monitoring 250 workers per day through beacon-based attendance, with automatic timesheet generation and zone-specific wage rate tracking.
Gradient AI builds predictive models for workers' comp insurers. Their algorithms analyze historical claims, demographic patterns, and medical trajectories to flag cases likely to escalate into chronic, high-cost claims before they do. Terra integrates Gradient's models into cloud-native claims platforms, where OCR-driven form processing and AI-generated claim summaries cut manual work by up to 40 percent. Liberty Mutual uses AI to generate automatic claim summaries and predictive alerts. Douglas Anderson, the company's workers' comp claims leader, said: "AI creates a superior customer experience. Our commercial customers mainly give us positive feedback."
Commercial customers. That qualifier matters.
Why None of It Reaches Residential
Every tool described above requires the same precondition: a formal employment relationship that generates digital records. GPS timecards need workers enrolled in a system with a smartphone and a login. BLE beacons need an employer who buys them, installs them, and registers workers against them. Computer vision needs cameras that the employer owns and feeds into a monitoring platform. Predictive claims analytics need claims, which require workers' comp policies, which require employers who classify their workers as employees and pay premiums into the system.
Residential misclassification operates outside all of this. A labor broker recruits workers from a parking lot at dawn. He pays them cash at the end of the week. He does not appear on the jobsite. He does not file 1099s. He does not carry workers' comp insurance. There is no camera feed to analyze, no GPS log to verify, no claim file to model, and no employer record for a state auditor to subpoena. As UMass researchers documented, some employers "cannot be located, others will stall and evade DUA representatives for as long as possible and, in some cases, the company simply disappears."
State departments of labor only audit employers registered in the unemployment insurance system. If a contractor never registered, the auditors never see them. Misclassification has gone underground: from the relatively traceable act of filing a 1099 for someone who should be a W-2 employee, to paying workers through shell companies and check-cashing services, to pure cash transactions that leave no paper trail at all. Pennsylvania's labor department called labor brokers "the key contributors to the problem of misclassification." AI compliance tools cannot analyze data that does not exist.
What This Means for Homeowners
If you are building a house or hiring a contractor for a major renovation, the lowest bid on your project may be low for a reason that has nothing to do with efficiency. A contractor who misclassifies a framing crew as independent contractors saves 30 percent or more in payroll taxes, workers' comp premiums, and benefits. That savings goes straight into price undercutting. The homeowner who selects the lowest bid is not choosing the most efficient builder. They are choosing the one most willing to break employment law.
Liability follows. If a misclassified worker is injured on your property and has no workers' comp coverage, the legal exposure does not stop at the subcontractor. Depending on the state, the general contractor, the property owner, or both may face liability for medical costs, lost wages, and damages. In states with joint liability provisions, the homeowner can be named in a lawsuit even if they had no knowledge of the subcontractor's labor practices. In states without such provisions, the injured worker may simply be uninsured and unable to work, absorbing the cost personally or through public assistance programs funded by your taxes.
This is not theoretical. Construction is a dangerous occupation. BLS data shows residential specialty trade contractors (NAICS 238) consistently among the highest for injury rates. A roofer falls. A framer takes a nail gun injury. An electrician is exposed to arc flash. On a commercial site, these events trigger a workers' comp claim, an OSHA report, a safety investigation, and adjustments to future protocols. On a residential site with misclassified workers, the injury happens, the worker disappears, and nobody records anything.
The DOL Rule That Changed Nothing
In January 2024, the Department of Labor published a final rule revising its guidance on employee versus independent contractor classification under the Fair Labor Standards Act. The rule reinstated a six-factor "economic reality" test that examines the nature of the work, the degree of employer control, the worker's opportunity for profit or loss, the investment in equipment, the permanence of the relationship, and the skill required. It replaced the Trump-era 2021 rule that had narrowed the test to two core factors and was widely seen as making misclassification easier.
In practice, the 2024 rule changed the legal framework for analysis but did not meaningfully change enforcement capacity. State labor departments remain under-resourced. Audits cover only registered employers. Labor brokers remain largely invisible to regulatory systems. Cash payments leave no audit trail. Penalties for misclassification, where applied, are modest enough that many contractors treat them as a cost of doing business. Twenty-three percent of the construction industry is composed of undocumented workers, according to the Center for American Progress, and these workers are less likely to report violations due to their immigration status and fear of retaliation.
What Would Actually Work
A few states have taken structural approaches. New York and Massachusetts both operate multi-agency task forces that share data across labor, tax, insurance, and attorney general's offices to identify misclassifying employers. The idea is simple: if a company reports zero employees to the unemployment insurance system but files building permits for projects that obviously require crews of twenty, something is wrong. AI could accelerate this pattern-matching at scale, cross-referencing permit filings, tax records, insurance databases, and OSHA logs to flag employers whose paperwork does not match the size and scope of their projects.
Justicia Lab developed an app called Reclamo that lets workers calculate stolen wages by answering plain-language questions on a phone. It generates demand letters citing state and federal labor law, drafts formal complaint forms, and collects data to identify patterns among employers and geographies. It works because it puts the tool in the hands of the worker rather than the employer. The app was optimized for construction because, as Justicia Lab's team put it, the industry is "egregious with wage theft."
CompScience's partnership with Nationwide shows how insurance incentives could theoretically push the technology downstream. If workers' comp carriers offered premium discounts to residential contractors who adopted AI monitoring, the economics might pencil out. A 23 percent reduction in claims on a residential policy could offset the cost of the monitoring system within a year. But this requires residential contractors to carry workers' comp insurance in the first place, and the contractors most likely to misclassify workers are the ones least likely to have a policy.
Limitations
The Century Foundation's estimate of 1.1 to 2.1 million misclassified workers uses indirect methods that compare household survey data to aggregated payroll records. The lower bound comes from census-to-payroll discrepancies; the upper bound includes adjustments for cash-only and off-the-books work that neither surveys nor payroll records capture well. Both numbers are estimates, not audited counts. The Texas and D.C. survey data come from non-random samples of specific jobsites and cannot be projected nationally with confidence. CompScience's 23 percent claims reduction figure comes from internal data shared through a partnership announcement, not peer-reviewed research. The 30 percent payroll savings from misclassification is a commonly cited industry estimate that varies by state depending on tax rates, insurance requirements, and benefit mandates.
None of which changes the structural problem. AI payroll compliance tools require digital data generated by formal employment relationships. Residential construction misclassification operates specifically to avoid creating that data. Until enforcement reaches the contractors who never enrolled in the system, the technology will keep making compliant employers more compliant while leaving the fraud untouched.