It is 4:40 on a Friday in the job trailer, and the project manager has four months of email open on a laptop with a cracked corner. Somewhere in those threads is the moment the owner said yes to moving the kitchen island six inches, which meant rerouting the gas line, which meant the cabinet shop re-cut two boxes. Nobody wrote a change order. Tile crew was coming Tuesday and somebody had to decide. So the island moved. Now the owner is holding final payment over $11,000 in disputed extras, and the PM is reconstructing a verbal approval from a text message that says "sounds good."
I have watched this exact scene play out, with different islands, for twenty years.
New research from Dodge Construction Network, commissioned by the change-order software firm Clearstory and published this June, finally puts industry-wide numbers on what that Friday afternoon costs. Ninety-eight percent of general contractors have lost fee to change order negotiations, not some of them but nearly all of them. And of that group, nearly half say the erosion topped 10 percent of their fee on at least some projects (Clearstory/Dodge, 2026).
Read that again. Half the industry has, on at least one job, watched a tenth of its profit dissolve in arguments about work everybody agrees was actually performed.
The Closeout Tax Nobody Budgets
Fee erosion is only the visible wound. Ninety-three percent of GCs in the Dodge survey have had project closeout delayed by change order disputes, and more than two-thirds of those delays stretched past a month. A month of open job cost, a superintendent still on the payroll, retention still sitting in the owner's account, and the PM spending his days as an archaeologist, digging through correspondence instead of running the next job. Nobody puts "argue about the island" on the baseline schedule, which is why it always lands in the same place: contingency, then fee, then the relationship.
On the trade side the numbers are uglier, because trades finance the gap. From work performed to authorized change order, the full cycle averages nearly seven weeks. Ninety-seven percent of trade contractors start the work before it is approved, because the schedule does not pause for signatures, and 83 percent say the resulting lag damages their cash flow. Seventy-seven percent end up writing off unapproved or downward-negotiated change orders as bad debt (Dodge/Clearstory via Business Wire). Understand what that means in plain terms. Your electrician is giving your project an interest-free loan for seven weeks, then forgiving part of it.
And the industry knows its process is broken. Only about one in three GCs, and one in three specialty trades, say their current change order process works very well, which means the other two-thirds have looked directly at the thing eating their margins and decided, collectively, to keep doing it. Meanwhile the other two-thirds run the Friday-afternoon archaeology expedition as standard operating procedure, which tells you most of what you need to know about who buys process software and who buys bigger filing cabinets.
The Paradox in the Old Data
Now here is the part that took me a while to believe, because it contradicts everything the Dodge numbers seem to say. There is an older dataset, from CoConstruct's 2020 analysis of residential projects, and it says change orders are profitable. Home builders running a single change order averaged $11,180 on it. More striking: for home builders, every additional change order on a project correlated with about 0.4 percent of extra profit margin (CoConstruct). Change orders that get processed make money. Change orders that get argued about lose money. Both datasets are true at the same time, and the entire difference between them is paperwork.
That is the original finding of this article, and I want to state it plainly because nobody in the vendor literature will. The margin does not live in the change. It lives in the 48 hours after the change is requested, a window in which the price either gets written down or evaporates into a text thread. Price it then, document it then, get the signature then, and the change order is a 0.4 percent margin gift. Let it slide, and it becomes a 10 percent fee amputation six months later at closeout. Ninety-one percent of GCs in the Dodge survey admit they sometimes do not pay full change order amounts, and the reasons they give are disputed pricing, incomplete documentation, and scope disagreements. Every one of those is a paperwork failure, not a construction failure.
Show the Math
Methodology, since I insist on it. Take a $1.2 million custom home, a 12 percent GC fee, which is $144,000. Apply the Dodge finding that fee erosion exceeds 10 percent on troubled projects: $14,400 gone. Now take six change orders priced at the CoConstruct three-change-order average of $7,479 apiece. That is $44,874 in change order value, about 3.7 percent of the contract, which is conservative next to the 5 to 20 percent of project value that change orders represent for most firms in the 2026 ServiceTitan/Thrive survey of over 1,000 contractors (Contractor Magazine). At a routine 15 percent markup, the documented margin on those six changes is roughly $5,850, which does not fully cover a $14,400 erosion event, and that is exactly the point. Documented changes were never the problem; the undocumented ones ate the fee, the island and the gas line and the re-cut cabinets, each one a verbal agreement dissolving into a text thread. Work performed without a price becomes a negotiation, and negotiations at closeout get settled with the owner's money.
Academic literature backs the mechanism. A 2025 peer-reviewed study of residential cost overruns ranked periodic design changes the third-strongest driver of overruns, tied with inadequate design and inspection, behind only schedule lag itself (PMC, 2025). PlanRadar's 2026 rework analysis, citing a 2026 ASCE study, found mid-project changes adding 10 to 25 percent in unplanned cost on typical mid-size projects, with two in three professionals saying changes drive overruns on many or most jobs (PlanRadar). Nobody is imagining this. The industry keeps measuring it, publishing it, nodding at it. Then everyone goes back to the text thread. Every time.
What the Software Actually Does
This is where AI finally walks onto the job, and I will describe what it does without the vendor adjectives. Clearstory, which commissioned the Dodge study and therefore deserves a disclosed conflict of interest, structures the change order communication log: requests, approvals, and pricing live in one thread instead of scattered across text, email, and memory. Procore and CMiC sell approval workflows with audit trails, so the signature happens before the tile crew arrives instead of after the lawyers do. Agave's Tom Reno, in Autodesk's 2026 trends roundup, argues the real 2026 story is AI moving into "the least forgiving parts of the business: finance, compliance, and controls," naming change orders, pay apps, and lien waivers explicitly, with AI proposing the coding and humans keeping the judgment (Autodesk). FlowForma offers no-code workflow automation that includes change order routing for firms without an IT department.
Strip away the branding and every one of these tools does the same three things: it timestamps the request; it attaches the price within hours, sometimes drafted by AI from the scope description and historical pricing; and it will not let the work proceed in the system without the approval, which forces the conversation that used to happen in the trailer to happen when it is cheap instead of at closeout when it is ruinous. One avoided erosion event a year pays for any of them. That is the entire ROI case, and it fits in one sentence, which is how you know it might be true.
The Case Against, at Full Strength
Now the objections, stated properly, because I have buried too many software rollouts to skip this. First, the Dodge numbers were commissioned by a company that sells change order software. Dodge did the survey work, and Dodge is reputable, but the questions serve the pitch, and no independent replication exists, so the responsible move is to treat the headline percentages as the top of a range and wait for the independent study that will either confirm them or embarrass everyone quoting them. Treat the 98 percent as directionally true and precisely uncertain.
Second, the data skews commercial, while the CoConstruct numbers it is being combined with are residential and six years old, so the whole paradox rests on joining a 2026 commercial survey to a 2020 residential dataset across an inflation wave, and any honest reader should hold both numbers loosely. Material and labor inflation since 2020 means the dollar averages are stale. Trust the ratios more than the levels. I have labeled them as such in the math above.
Third, and heaviest: software cannot make an unwilling owner pay, it cannot make a disorganized GC document, and it cannot turn a handshake-deal culture into a countersignature culture, which means the tool only works for firms already willing to have the uncomfortable pricing conversation early. Firms reporting "measurably better outcomes" in the Dodge study may simply be better-run firms that also happen to buy software. Selection bias is the ghost haunting every vendor case study. A structured process on paper, enforced by a foreman with a clipboard and a 48-hour pricing rule, beats an AI workflow nobody opens. The technology removes the excuse, not the discipline.
Fourth, the graveyard. Katerra burned through $2 billion, and Veev took $647 million with it. Construction tech history is a long list of tools that added clicks instead of removing them, and every one of them had a beautiful ROI slide. If the tool does not get faster than the text thread, the text thread wins, because the text thread is already in the PM's pocket. Pockets beat platforms.
What This Analysis Could Not Prove
Honest boundaries. Dodge and Clearstory surveyed commercial GCs and specialty trades, so applying the findings to residential custom homes is inference, not measurement, and any residential GC reading this should discount accordingly. CoConstruct's dollar figures are 2020-vintage and predate the inflation wave. Treat $11,180 and $7,479 as historical markers, not current prices. That $14,400 worked example assumes a 12 percent fixed fee and the greater-than-10-percent erosion scenario; cost-plus contracts behave very differently, and CoConstruct itself found fixed-price change orders running $3,114 higher than open-book ones on million-dollar-plus projects. I found no independent, third-party measurement of how much any AI tool shortens the seven-week trade approval cycle; vendor claims on cycle-time reduction are marketing until audited. And the 0.4 percent margin correlation is exactly that, a correlation, not proof that adding change orders causes profit, and any GC who reads this as permission to manufacture changes deserves the closeout fight he is about to have.
What to Do Monday Morning
For builders and remodelers running one to five million a year in residential work, four moves, in order. First, adopt a 48-hour pricing rule: every change gets a written price within two days of the request, before the work starts, no exceptions, because the data says the signature timing is the whole game. Second, standardize the change order request form with photo documentation attached at creation; the phone in every foreman's pocket is already the capture device. Third, price the seven-week approval gap explicitly, either with faster approvals or with terms that compensate the carrying cost, since your trades are currently lending you money interest-free and writing off the difference. Fourth, if you buy software, buy it for the timestamp and the audit trail, not the AI adjectives; the value is forcing the conversation early, and any tool your PM will not open on a Friday afternoon is a donation to a vendor.
For homeowners, the advice is shorter and cheaper. Never say "just do it and we'll figure out the price later." That sentence is the most expensive in residential construction. Demand every change in writing with line-item pricing before work begins. Expect 10 to 20 percent markup on change orders as normal; it covers the disruption pricing no estimate can foresee. Treat 50 percent as a red flag and a second-bid situation. Walk away if they will not itemize. And understand that your contractor's 48-hour pricing rule protects you too, because a builder who prices honestly and early is a builder who will still be solvent at closeout, which is when you need him most.
Twenty years of Friday afternoons have taught me that the industry does not have a change order problem, only a documentation habit problem, and habits are exactly what software is good at enforcing when humans are too tired, too rushed, or too deep in the trailer on a Friday to enforce them on themselves. It has a memory problem. Work gets done, the price gets forgotten, and the forgetting gets billed to whoever has the least leverage at closeout, which is usually the person who did the work. AI that timestamps, prices, and routes approvals is not glamorous. It will never keynote a conference. It just quietly converts the most expensive sentence in construction into the cheapest one: "Sign here first." That sentence is worth $14,400 a job. I have the archaeology expeditions to prove it.