$450,000 in, $175,100 out: what the research says a $5 million civil job keeps
A $5 million civil job plans on a $450,000 margin. Published research says it keeps $175,100. Every figure sourced, with the three we cannot measure named.
Take a $5 million civil job. Fifty-two weeks to complete. A forecast contractor margin of almost 9%, which is what eight Australian transport projects actually carried.
That is $450,000. It is the number the job was won on, the number the estimator defended, and the number the bank saw.
Here is what the published research says is left of it.
5.5% of contract value
23% of annual profit, measured
4.9% held, 13.1% never returned
$17,400 at a lower overhead rate
3.5% of contract value
about 1% to 2% of the contract
The diagram runs the high-overhead case, where preliminaries are 20% of the contract sum. Run it at 10% and the wet days cost $17,400 instead of $40,300, head office costs $49,500 instead of $99,000, and the job keeps $247,500. So the honest answer is a range: $175,100 to $247,500, or 3.5% to 4.9% net.
Against a clean run of the same job, where each of those is caught before it turns into a cost, the gap is $153,000 to $176,000 of net profit. On one job.
7.0% of contract value
3.5% of contract value
Same job, same crew, same contract. The difference is whether anything read the record in time.
And that diagram is the optimistic version.
What the diagram leaves out, and why
Three losses leave the flow there, and they belong to two diseases: rework, and missed claims twice over. Cost drift, disputes and lapsed compliance never appear in the diagram at all, along with the largest part of missed claims, for one reason: nobody has published a measurement of what the contractor actually loses.
That is not the same as nothing being at stake. Here is what is at stake on the same job, against the whole margin.
33.4% of budget (HKA)
20% over (Flyvbjerg)
10.6% of contract
10% of contract value
What the job was won on
4.9% of contract value
Four of those five exceed the whole margin on their own. A job does not have to lose all of any of them to lose all of the margin.
Read that chart carefully, because it is the one most likely to be misused. Sums in dispute is not money lost. It is money argued over, on projects that were already in trouble, mostly much larger than this one.
But it is not free either, and this is the part the "at stake" framing hides: you pay to fight regardless of who wins. Win every dispute on the job, recover every dollar claimed, and the legal bill still lands.
And winning outright is not the usual outcome. The clearest Australian number we could find is Queensland's, published annually and checkable line by line: in 2024-25 the QBCC released 214 adjudication decisions on claims totalling A$956,966,293, and awarded A$81,873,482. That is 8.6 cents in the dollar. One A$633,969,448 claim sits inside that total and drags the ratio a long way down, so read it as the shape of the thing rather than your odds.
214 decisions released in 2024-25
8.6 cents in the dollar
How much the fight itself costs is the figure we are least able to stand behind. The most-cited number is Richard Gebken's 2006 PhD study of 44 US projects, which put the mean cost of resolving one dispute at US$330,199 by negotiation, US$1,212,433 by mediation and US$1,167,182 by arbitration. Three caveats, all of which matter: those means come from only 18, 15 and 11 projects respectively; they are US dollars on larger US projects, and we have not converted or scaled them; and we have not read the thesis itself, only CPR's account of it, which makes it a secondary figure by the rule we set ourselves. Treat it as an order of magnitude, not a line item.
What is not in doubt is the direction. Arcadis puts the average time to resolve a North American dispute at 12.5 months - a year of somebody's management time, which appears on no invoice at all.
That distinction is the whole reason we wrote the evidence page before we wrote this post.
Why does rework rank above cost drift?
The obvious ranking is by size, and by size cost drift wins: $1,000,000 against rework's $500,000. We rank rework first anyway, and the reason is the tie-break we used throughout.
Where two diseases have similarly strong evidence, the one that measures money the contractor itself loses beats the one that measures money merely at stake.
Cost overruns are largely overruns against the client's budget, and much of that money is paid by the client. Nobody has reliably measured what share a contractor carries rather than recovers. Rework has been measured directly, from one contractor's own books.
The gap that should bother you
Ask people what rework cost and you get 6% to 10% of contract value. Count what got written down as rework and you get 0.38%. Both are real measurements of different things.
10% of contract value; 6.4% direct plus 5.9% indirect
0.38% of contract value, 0.76% including post-completion fixes
Dashed line: the job's entire planned margin ($450,000)
The same 2026 study that recorded 0.38% also found actual rework costs are underreported by around 300% against site records. Most rework is never written down as rework.
A contractor whose system says rework cost 0.4% this year does not have a good year. It has a recording problem. That gap is not an accounting curiosity: it is the reason a disease can eat a fifth of the annual profit while every report on the desk looks fine.
Do civil contracts pay for wet weather?
Weather is not its own disease. A wet day the contract's extension-of-time clause never got claimed for is a missed claim, the same as a variation nobody raised. What makes it worth its own section is how precisely the contracts are written to leave it with you.
Meanwhile the site sheds, the supervision, the traffic management and the insurances keep running. Preliminaries are commonly put at 10% to 20% of the contract sum, and thirty senior Australian estimators put about 45% of them as time-related. On this job that is $870 to $2,010 a working day that runs whether or not anything gets built. Twenty days of it is the $17,400 to $40,300 in the diagram.
The crew is on top of that. Under the Building and Construction General On-site Award, ordinary hours lost to inclement weather are paid, up to 32 hours in any four-week period. At the CW3 award minimum, each lost day is at least $223.82 per worker before the civil industry allowance and on-costs.
Does 3.5% to 4.9% look right?
It does, which is the part that should worry the industry rather than reassure it.
Operating profit before tax, Australian heavy and civil engineering construction, 2022-23
6.9% of sales in 2023-24, and 2.3% in 2011-12. Large Victorian commercial builders averaged net margins that fell from 4.0% to 1.4% between 2006 and 2015. A model that lands a $5M job at 3.5% to 4.9% net is not a pessimistic model.
ABS, Australian IndustryTwo other numbers sit behind that. The Reserve Bank found over a quarter of Australia's 200 largest builders posted an operating loss in the year to March 2022, and that construction makes up close to 30% of company insolvencies. ASIC put construction at 27% of companies entering external administration in 2023-24.
On a $450,000 margin, a 9% cost overrun that cannot be recovered takes the whole thing. That is less than half the average road overrun.
What we can and cannot say
Every figure in this post is sourced, and the sources are not equally strong. We would rather publish the weakness than paper over it.
| Disease | Evidence class | What the evidence measures | Charted above? |
|---|---|---|---|
| Rework | Contractor's loss | 23% of annual profit, from one contractor's ledgers. | Yes, $103,500 |
| Missed claims: retention | Contractor's loss | 4.9% held, never returned on 13.1% of contracts for smaller contractors. | Yes, $32,100 |
| Missed claims: weather | Event only | Wet days and daily overheads are published; what a contractor fails to claim is not. | As a worked example |
| Cost drift | Event only | Overruns are measured; the contractor's share of them is not. | No |
| Missed claims: variations | At stake only | 10.6% of contract value in change orders; the unrecovered share is unmeasured. | No |
| Disputes | At stake only | 33.4% of budget, skewed to disputed megaprojects. The cost of fighting is measured; what you keep after is not. | No |
| Lapsed compliance | Per incident | $1,200 to $97,408 to the employer. Not a share of contract value. | No |
Lapsed compliance ranks last on measured cost, not on how much it matters. The thing contractors actually fear, being dropped from a licence, a panel or a prequalification list, has no dollar figure published anywhere we could find. It is also the one loss that ends the other five by ending the work.
The $153,000 to $176,000 gap is the most this job could keep if each disease were caught in time: the rework found before the pour, the retention chased the day it fell due, the wet days planned for. Catching them early is what Demiton is for. Nobody has measured how much of each loss the platform prevents, and today it puts the facts one question away rather than raising the flag itself. The cures roll out one disease at a time, and only one of the five has an automated verdict built so far. What each plan protects is priced on the work under cover, not per seat. We will report each one when it has fired on a real job, not before.
Why we published the arithmetic
Most software in this market quotes a percentage saved and declines to show the working. We ran six literature searches, one per disease, under three rules: only report a figure actually read in the source, say exactly what it is a percentage of, and say so plainly when good evidence does not exist rather than stretching a weak number. Every figure was checked a second time against the original before it went up. Where we could only see a figure through someone else's citation, it is marked secondary.
The order those searches produced is the order the diseases appear in when you set Demiton up, and the order we build detection for them. It is a default, not a claim about your business. A contractor with a run of disputed variations, or one heading into a wet season, may reasonably rank them differently, and setup lets you.
The five diseases run on registers built from the systems a contractor already has: what Assignar alone fills, and where it stops, is the companion to this piece. The two weakest inputs to the diagram are the daily site overhead and the head office rate. If you know your own, use them. If you find a figure here that does not match its source, or a better study we missed, tell us and we will change it.
One note before you click through. The docs use the formal register names for the same five things: missed claims are forfeited entitlements, cost drift is the baseline overrun, and lapsed compliance is non-compliance. Same diseases, same numbers, more precise words.
Read the full evidence page, with every source and every caveat
Frequently asked questions
- How much does rework cost an Australian civil contractor?
- A survey of 115 Australian civil infrastructure projects put total rework at 10% of contract value. Six years of one Australian contractor's ledgers found it bore rework costs directly on 93% of its projects and lost a mean 23% of its annual profit to rework. On a $5 million job with a 9% margin, that 23% is about $103,500. Ledger studies that count only what was recorded find far less, 0.38% to 0.76% of contract value, but the same research found rework is underreported by around 300% against site records.
- Is the $175,100 figure a measured Demiton result?
- No. It is what published research says lands on the contractor, worked through one hypothetical $5 million job. Nobody has measured how much of each loss Demiton prevents. Today the platform puts the facts one question away rather than raising the flag itself, and only one of the five diseases has an automated verdict built.
- Why does rework rank above cost drift when the overrun is a bigger number?
- Because the tie-break is what the contractor actually loses, not what is at stake. Roads come in 20% over budget on average, $1,000,000 on a $5 million job, but nobody has reliably measured what share of an overrun the contractor carries rather than recovering through variations. Rework has a direct measurement of contractor profit lost.
- If you win a construction dispute, do you get your money back?
- Not all of it. The ICC calculated from 221 awards that party costs - your own lawyers, experts and witnesses - are 83% of what an arbitration costs on average, and that is spent whether you win or lose; the tribunal is the small part. Nor is a full win typical: the QBCC released 214 adjudication decisions in 2024-25 on claims totalling A$956,966,293 and awarded A$81,873,482, about 8.6 cents in the dollar, though a single A$633,969,448 claim skews that ratio. The most-cited figure for what a dispute costs to resolve, US$330,199 by negotiation, comes from a 2006 US study we have only read second-hand, so we do not lean on it.
- Do civil contracts pay for wet weather?
- Often not. The Transport for NSW major contracts specification prices in 20 wet-weather days out of 226 working days a year and pays Wet Weather Delay Costs only when the time for completion exceeds 52 weeks, above a threshold, and at 50% of a rate it sets. Queensland's Transport Infrastructure Contract grants an extension of time but no delay costs by default. AS4000 and AS4902 give no weather relief at all.
Diagnose your tender losses.
Upload your historical contracts. Demiton Insights classifies them and shows you where your rates drifted - which jobs ran over, which suppliers held firm, where the margin went. AU$10,000/year or AU$999/month, no per-seat.