Your Sub Billed 15% Ahead of the Work. That Was His Entire Profit, Paid Early.

By Frank DeLuca, September 15, 2026, Project Management & Operations

It is the 25th of the month, which means the pay applications have arrived, which means you will spend tonight at the kitchen table with a red pencil, twelve schedule-of-values sheets, and a memory of what the job site looked like last Tuesday, because the contract fixed the 25th as draw day years before you ever walked this lot and nobody has moved draw day for anyone's convenience since. This is the pencil draw, and every general contractor knows it the way sailors know weather. Your sub says the electrical rough-in is 70 percent complete. You walk the house in your head, counting boxes and home runs, and you decide whether you believe him. Then you sign a document that moves tens of thousands of dollars on the strength of that belief.

Twenty years of projects going sideways has taught me that the pencil draw is the most consequential control on a residential job and the least scientific. A 2024 Payapps survey of more than 1,000 quantity surveyors and contract administrators found that 64 percent of construction projects experience overbilling. Nearly half of respondents said they face it occasionally. Seventeen percent said frequently, so read that sentence again slowly, because it means that on nearly two out of three jobs somebody is billing for work that is not done yet.

Most of it is not theft, a distinction that matters and one we will return to once the arithmetic has had its say. But first that arithmetic, because it is worse than you think.

Eight points is the whole margin

Construction Dive's payment reporting puts typical subcontractor margins between 5 and 10 percent. Take the middle of that band, and a $300,000 trade package at an 8 percent margin carries $24,000 of profit. Divide $24,000 by $300,000 and you get 8 percent. Which means a sub billing just 8 percentage points ahead of verified completion has already collected his entire profit in cash, weeks or months before the work is finished.

Every point beyond that, you are financing his next job. His job, not yours.

At the 5 percent margin end of the band, the number is just 5 points, which means a framer billing 55 percent on work you would walk at 50 has banked everything the job was supposed to pay him, leaving the remaining 45 percent of the contract as labor and materials he must fund out of money you already gave him. Construction Dive's analysis spells out the incentive this creates: a sub billing 20 percent ahead has made his profit, which leaves very little reason to come back for the punch list. Ask any PM who has begged an electrician to return for six device plates, and he will tell you he has seen this movie before, usually around the third unanswered text. Every PM has.

The inputs are simple enough to check: contract value times points-ahead equals cash advanced; compare against margin times contract value. Underneath, the calculation assumes margin spreads evenly across the schedule of values, though in practice schedules of values are themselves front-loaded, heavy on mobilization and rough-in and light on trim and closeout, which does not weaken the rule but makes the real number worse, since profit concentrates exactly where the billing runs ahead.

What the pencil draw actually costs you

Now price the control itself. Twelve sub pay apps a month is normal on a serious custom home. Forty minutes each is honest: the field verification, the math check across the G702 summary and the G703 continuation sheet, the lien waiver chase, the email asking why line 14 moved twelve points with no change order attached. That is 8 hours a month, compounding across a 10-month build to 80 hours, which at a loaded PM cost of $95 an hour means you are spending $7,600 per project to review billing by memory and red pencil.

That figure is modeled, not measured, and your mileage will vary: some months the draw takes twenty minutes because everything reconciles, while other months it eats three days because nothing does. But hold the $7,600 next to the exposure table above and the economics of the status quo get uncomfortable. You are spending the price of a used pickup to operate a control that misses on 64 percent of projects.

Misses sometimes go spectacular, as Massachusetts proved when S&R Construction and its sub settled for $420,000 after Massachusetts prosecutors said they knowingly front-loaded pay applications on an MBTA station job, billing for work not yet performed in violation of the state False Claims Act. S&R is barred from public work in Massachusetts for five years. That was fraud, proven, with a debarment attached. Most residential overbilling never gets near a courtroom. It dissolves into the fog between "percent complete" as the sub sees it and "percent complete" as the red pencil sees it, and the fog is where the money goes.

An independent witness arrives

Two technologies are converging on this fog from opposite sides, and neither one is really about catching thieves.

From the field side, camera-verified progress: Buildots' Dot assistant answers plain-language questions about exact progress percentages per trade, per area, drawn from 360-degree site captures matched against the BIM model and the schedule. A superintendent can ask which units have drywall closed but tile not started, and get an answer grounded in what the lenses saw, not what the sub reported. Our own prior reporting has put this class of tool at a few thousand dollars a month on residential-scale work. Against the exposure table, the break-even is roughly one caught front-load per year. Most PMs I know catch that many by Thanksgiving.

From the document side, pay-app validation. The AIA G702 and G703 forms are arithmetic wrapped in notarization, which means the arithmetic is checkable. GCPay's GC review checklist already pushes automated validation of G702/G703 alignment as standard practice. More telling is Buildbase-payapp, an open-source project that runs AIA calculations and anomaly detection over billing amounts with a Claude API backend, and when the fraud-detection layer for the industry's standard billing form is a GitHub repo, the capability has commoditized. What remains undecided is not whether the math can be checked automatically, but why you are still checking it by hand.

Neither tool replaces the walk, since cameras cannot see inside walls and no document AI has ever caught a bad waterproofing detail before the tile went over it. What they replace is the memory half of the pencil draw, the part where you reconstruct last Tuesday from a brain that has looked at four hundred Tuesdays, because an independent percent-complete per trade sitting next to the billed percent on the same screen turns the 8% rule from a parlor calculation into a monthly routine: run the subtraction on every app, flag anything past the margin line, and the whole system fits on an index card.

The case against this article, stated properly

Now the objection, at full strength, because it is largely correct: front-loading is usually not fraud but survival, and the distinction decides whether the right response is a camera or a faster check.

Siteline's billing data shows 5 percent of subcontractors are paid on time, with the rest waiting an average of 96 days for money on work already completed. Levelset's survey found 37 percent of GCs always collect an advance deposit, versus 13 percent of subs. His crew's payroll lands every Friday while the payment chain, GC to owner to lender and back down, takes three months, so of course he bills ahead; the surprise would be if he did not. That is the payment chain working exactly as designed, which is to say badly for everyone except the lender. The Even the CFMA's own analysis concedes that billing for stored materials or anticipated change-order approvals is often a rational response to being starved of cash, not a scheme. Pointing cameras at a man to verify he is only 45 percent done, while you take 96 days to pay him for it, treats a cash-flow disease as a honesty problem. Fraud is what the MBTA case was, genuinely and provably. Your framer is not the MBTA case; he is trying to make Friday payroll.

Tighter verification without faster payment just moves the pain downhill, and downhill is where the smallest companies live. Rabbet's 2024 payment report, via CFMA, puts the industry's annual losses to slow payment at $280 billion. Any GC who installs verification AI while keeping 90-day pay practices has built a very expensive mirror and learned nothing.

What this analysis does not prove

The 64 percent figure comes from a vendor survey with self-reported data, and "overbilling" in that survey includes honest mistakes: typos, mismeasured quantities, change orders billed before formal approval. Whatever the true fraud rate is, it sits below that 64 percent, and nobody knows by how much. No public dataset I could find isolates residential overbilling rates, so the commercial and public-project figures here are proxies, used openly as such. Vendor claims about per-trade progress accuracy have thin independent verification at residential scale; treat camera-verified percentages as a second opinion, not a survey. That $7,600 review-cost figure is modeled from stated assumptions, and shops with a dedicated project accountant will land somewhere else entirely. And the 8% rule assumes you know the sub's margin, which you do not; use your best estimate and watch the trend across months, because a consistent 6-point gap is telling you something even if the exact margin is a guess.

What to do on Monday

Break the schedule of values into milestone-verifiable line items before the job starts. No single line that reads "rough-in, 0-100%." If a line cannot be verified by standing in a room and looking, it is not a line, it is a wish.

Require the pencil walk before the pencil draw. No pay app gets approved without same-week field verification, and the verifier initials the app. This costs nothing and catches more than any software.

Run the 8% rule on every application: billed percent minus verified percent, times contract value, compared against your best estimate of the trade's margin. Anything past the margin line gets a phone call, not a signature.

Cap stored-materials billing to what is on site, insured, or in bonded off-site storage with proof. Materials in somebody's warehouse are somebody's problem, and it should not be yours.

Pay faster, because running verification AI on subs you pay in 90 days gets the order of operations backwards. A check that arrives when the contract says it will remains the cheapest anti-overbilling technology ever invented, and everything else is instrumentation around a trust problem you created.

Nobody is retiring the red pencil. For the first time in twenty years, it does not have to work from memory alone.