You’ll Spend $24,000 Training an Apprentice. There’s a 47% Chance He Quits.

Danny Reeves lasted nine months. He was 22, came out of a community college construction management program in Raleigh, and his foreman said he had good hands. By month three he could hang drywall faster than the other first-years. By month six he was reading prints without help. By month nine he took a warehouse logistics job that paid $3 more an hour, had air conditioning, and didn’t require him to be on a roof in August.

His employer had spent roughly $23,500 net getting him to that point — after crediting the productive work Danny did contribute. Wages above his output value. Journeyman time diverted to supervision. Rework on his early mistakes. Recruiting costs. Safety certification. The warehouse job didn’t reimburse any of it.

Danny isn’t unusual. He’s the median.

The Dropout Math Nobody Calculates

The BACH report from August 2024 put the UK construction apprenticeship dropout rate at 47%. Only 8,620 apprentices reached their End Point Assessment in 2022–23, against an industry need of 96,000 new recruits per year. The U.S. numbers are murkier — the DOL’s completion rate dashboard is still in beta — but ABC, NCCER, and every workforce study I’ve read in the last three years tells the same story. Roughly half the people who start a construction apprenticeship don’t finish.

For a residential builder running a 10-person crew, this is a financial wound that reopens every year.

Take a 10-person residential framing crew in a mid-cost market (think Raleigh, Tampa, or Phoenix). You bring on two apprentices per year to replace attrition and grow. Average first-year apprentice wage: $18.50/hour (BLS, construction laborers and helpers, May 2024). A journeyman carpenter supervising them earns $32–$38/hour and loses roughly 20% of productive time to mentoring. Here’s the cost breakdown for one apprentice who drops out at month eight:

With a 47% dropout rate and two apprentices per year, a crew this size can expect to lose one trainee annually. That’s $23,500 in sunk costs that produces no long-term workforce capacity. Over five years: $117,500. Enough to buy a pickup truck and a concrete saw, or — and this is where the VR pitch enters — 78 Meta Quest headsets at $300 each.

What VR Training Actually Claims

The VR training industry has grown loud. Interplay Learning, the Austin-based platform that’s trained nearly 600,000 skilled trades workers, released its 2025 State of Skilled Trades Training Report in December. The headline numbers: 7 in 10 companies say they ramp up technicians faster using simulation-based training, saving an average of 4.3 months compared to traditional methods. 52% of workers who used digital learning said they were more likely to stay at their current job.

PwC ran what’s still the most rigorous enterprise VR study in 2020: VR trained workers 4× faster than classroom instruction, with 275% more confidence applying learned skills. Cost parity with classroom training kicked in at 375 learners.

At Central Washington University, researchers testing Humulo’s VR safety modules found trainees scored up to 30% higher on difficult safety concepts and showed 250% improvement in procedure completion accuracy. A Safety Science meta-analysis covering 52 studies concluded VR safety training increases safety awareness by 30%.

Impressive. But notice what’s missing from every one of those numbers.

None of This Measures What Actually Matters

Not a single published study I found measured the effect of VR training on construction apprenticeship completion rates. Not one.

The PwC study? It trained new managers on inclusive leadership. Soft skills. Not how to toe-nail a stud into a bottom plate. Interplay’s 4.3-month acceleration? Self-reported by their paying customers, who have an obvious incentive to justify the subscription. The CWU study? Twenty people in a fire safety module. The retention finding from Interplay — 52% “more likely to stay” — is an attitude survey, not a measured outcome. Saying you’re more likely to stay is not the same as staying.

Construction apprentices don’t drop out because the training is boring. They drop out because the pay is low relative to alternatives ($18.50/hour vs. $22+ in warehouse logistics), the work is physically punishing, the hours are unpredictable, and the path to journeyman status takes three to five years. A VR headset doesn’t change any of that.

Where It Might Actually Help

VR doesn’t deserve the hit piece. The technology addresses a genuine gap — just not the one the marketing decks emphasize.

The BACH report identified the top reasons apprentices drop out: feeling unsupported by their employer (23%), frustration with the gap between classroom theory and job-site reality (19%), and safety anxiety on complex tasks they hadn’t practiced (15%). Those second and third reasons are directly addressable by simulation.

A first-year apprentice who has practiced residential electrical rough-in in VR twenty times before touching live wire is less likely to freeze on the job site. The BLS CFOI data for 2024 recorded 1,032 construction fatalities — one every 8.5 hours. Tyson Foods cut workplace injuries 20% year-over-year after deploying VR safety training. If simulation reduces the early-months fear factor that drives some apprentices to quit, the per-dropout math shifts.

Here’s the calculation that VR vendors should be making but aren’t. If VR training reduces the 47% dropout rate by even 10 percentage points — to 37% — for a crew hiring two apprentices per year, that’s one fewer dropout every five years instead of one per year. At $23,500 per dropout saved and a VR system cost of roughly $4,800 per year (two headsets at $300 each plus $175/user/month for platform licensing), the ROI turns positive in year two.

Marginal. A rounding error on a residential builder’s P&L. The real payoff would come from the 4.3-month acceleration in time-to-competency, which means apprentices reach productive output faster — but again, that number is self-reported and unverified by independent research.

The Workforce Cliff Nobody’s Ready For

The context makes this more urgent than the numbers alone suggest. ABC estimates the construction industry needs 349,000 net new workers in 2026, rising to 456,000 in 2027. Deloitte’s 2026 outlook puts it higher: 499,000 new workers needed, with 41% of the current workforce retiring by 2031. Only 7% of potential job seekers even consider construction as a career.

If you can’t recruit enough, you have to retain more of the ones you get. The DOL’s RAPIDS data shows roughly 60,000–75,000 new construction apprentices entering registered programs each year (construction accounts for approximately 30% of all registered apprenticeships nationally). At a 47% dropout rate, somewhere between 28,000 and 35,000 of those will never finish. Reduce that dropout rate by 10 percentage points and you retain an additional 6,000–7,500 skilled workers per year — modest against ABC’s 349,000 shortage, but meaningful given that these are workers who already chose construction and needed only a reason to stay.

The DOL’s Mathematica study on registered apprenticeship found that completers earn $240,037 more over a 36-year career than non-participants, with net social benefits exceeding $49,000 per apprentice. Every dropout is a quarter-million dollars in lifetime earnings that evaporates.

The Hard Counterargument

VR can’t teach the feel of wet mortar. It can’t simulate the weight of a 4×12 LVL beam on your shoulder, or how your boots slip on dew-covered OSB sheathing at 7 AM. It can’t replicate the social dynamics of a crew — the hazing, the mentoring, the lunch-break conversations where a 30-year carpenter explains why you don’t nail within two inches of a knot.

And the dropout problem isn’t primarily a training-quality problem. Construction wages grew 4.2% year-over-year through August 2025 and still can’t compete with warehouse, logistics, and gig economy alternatives that offer more predictable hours and climate-controlled environments. Until the industry solves its compensation and working-conditions gap, better training is a bandage on a structural wound.

Small residential builders — the ones who need this most — face another barrier. Interplay Learning charges $30–$250/user/month. Custom VR content development runs $50,000–$195,000 per module. NCCER has standardized curricula, but VR modules mapped to those specific credentials are still limited. A framing contractor in Boise doesn’t need a module on confined-space rescue. He needs one on residential roof truss installation that matches Idaho’s IRC amendments. That module probably doesn’t exist yet.

What This Doesn’t Prove

The $23,500 per-dropout cost is my estimate based on BLS wage data, typical residential crew structures, and employer cost surveys. It assumes a mid-cost market, an 8-month tenure before dropout, and a 45% apprentice-to-journeyman productivity ratio that I derived from NCCER competency benchmarks. Your market, your crew, and your apprentice will differ. A first-year electrician apprentice in San Francisco costs more to lose than a laborer apprentice in Little Rock.

The 47% dropout figure is from UK data. The US DOL completion rate dashboard is in beta and doesn’t yet break down by construction trade. US completion rates may be higher or lower.

Interplay Learning’s outcomes data is self-reported by their customers. No independent randomized controlled trial has measured VR’s effect on construction apprenticeship completion rates specifically. The PwC study — the gold standard for VR training research — measured soft skills, not physical trades competencies. Extrapolating from one to the other requires assumptions I can’t verify.

The 10-percentage-point dropout reduction I used in the ROI calculation is hypothetical. It could be higher. It could be zero. Nobody has measured it, and I want to be clear about that.

What I do know: 349,000 workers short. 47% who start don’t finish. $23,500 per dropout. A $300 headset and a $175/month subscription aren’t going to fix the pay gap or the August heat. But if they keep one more apprentice through that critical first year, the math starts working. Barely. We need better data to know if it actually does.