UK Construction Rework & Admin Cost Report 2026

There is a figure in almost every talk, tender and trade article about UK construction quality: rework costs between 10% and 25% of project value.

It is repeated so often that nobody checks it. I did. It does not mean what most people using it think it means, and the defensible version is more specific and more useful than the range.

This report traces it back to the primary research, separates what was measured from what was estimated, and puts an operator number on the cost per £1m of project value. Then it looks at the second cost nobody counts: the admin consumed by variations, notices and retentions.

This is not a software pitch. Margins are thin, sites are hostile to anything with a screen and a battery, and the primary source below is explicit that the industry does not measure the thing everyone quotes.

What GIRI actually found — and what it did not

The primary source is the Get It Right Initiative, set up in 2015 out of the Institution of Civil Engineers Best Practice Panel, with research by Expedition Engineering. The quoted number comes from its Strategy for Change: "the measured direct costs of avoidable errors are in the order of 5% of project value… When unmeasured and indirect costs are included the situation gets much worse with estimates of total costs ranging between 10% and 25% of project cost" (GIRI, revised June 2019, p.3).

Neither figure is GIRI's own: the 5% comes from a literature review of international studies, and the 10–25% range summarises the spread of other people's estimates.

GIRI's own number is a single figure. Research Report section 6.6: "If we assume that the direct cost of recorded error is around 5% the results of the Delphi analysis suggest the total cost of error to the UK construction industry is around 21% of the total spend" (GIRI, Revision 3, April 2016, p.56). The composition is exact: recorded direct 5%, indirect 7%, unrecorded process waste 6%, latent defects 3%.

The derivation matters. GIRI used the Delphi Method — structured consensus among eleven senior quality and commercial figures from firms including BAM Nuttall, Costain and Wates — to assess the relative magnitude of the four categories: recorded direct 24%, indirect 34%, unrecorded process waste 29%, latent defects 13%. Anchoring that 24% share to the 5%-of-spend figure scales the whole to 21%. So the headline is one literature-derived anchor multiplied by eleven people's view of proportions: the best published UK estimate available, which I use throughout, but not a measurement.

The finding that should worry you more than the 21%

Buried in the interview results is something more damaging: GIRI could not get the data, because the data does not exist.

"All of the systems that we saw only capture the direct cost of error, and do that only partially. None of the organisations that we interviewed was able to provide data relating to indirect costs or unrecorded process waste" (GIRI, 2016, p.18). And: "No Contractors were able to provide an estimate of the total cost of error… Tier 1 Contractors have very little understanding of the true cost of error which is borne by their supply team."

The best-resourced research the UK industry has run on this question could not measure it, because contractors record defects at handover and almost nothing before it. Most of the direct cost of error is, in GIRI's words, "in effect, experienced as 'process waste' at Tier 2 and below" — resolved before anyone hands over, never written down. I have written about this pattern as ghost workflows: the work that consumes the most time is the work no system was asked to record. Construction's version is the most expensive, because that unrecorded portion sits in a supply chain of 885,485 businesses, 78.1% of them with no employees (DBT, 2 October 2025).

The margin arithmetic

Here is why 5% matters more than 21% for anyone running a contracting business.

The Construction Index's analysis of accounts filed before 30 June 2025 found the average pre-tax margin across the Top 100 UK contractors was 2.4%, up from 1.9% a year earlier, on combined turnover of £76.5bn and aggregate pre-tax profit of just over £1.9bn (The Construction Index, 29 August 2025). Forty-three of the hundred were below 2%, and sixteen were loss-making.

On £1m of turnover at a 2.4% margin a contractor keeps £24,000. GIRI's recorded direct cost of error on the same £1m is £50,000.

The rework you already know about — the part already in your cost reports — is roughly twice your entire pre-tax profit. GIRI said in 2016 that error cost "around 7 times the total annual profit of the UK Construction Industry"; on 2025 accounts it is 8.5 times. That calculation is mine.

No procurement squeeze available to a contractor is worth as much as a few points off the error bill — and that argument works on the 5% alone.

The productivity backdrop, and one claim that is wrong

Construction productivity is falling: output per hour fell 2.2% in 2025 and 1.8% in 2024, with hours worked up 5.5% against gross value added up 1.2% (ONS, 19 May 2026, series DJP6 and DJP8). More labour, not much more finished work — which is what a rework problem looks like in national statistics.

Where the admin hours go: variations

The second cost is administrative and much harder to source, because nobody publishes it. So let me be exact about what is verified and what is modelled.

Verified is the contractual clock. Under the NEC4 Engineering and Construction Contract — the standard form for most UK public infrastructure work — the contractor must notify a compensation event within eight weeks of becoming aware, or lose entitlement to any change in the Prices or Completion Date entirely (clause 61.3). Exhibit 3 sets out the full cycle from NEC's own guidance (NEC, Compensation events; NEC, Accepting quotations).

One compensation event on which everybody performs therefore has a six-week administrative cycle, and up to fourteen weeks if notification runs close to the bar. A contractor with a hundred live events is not managing a hundred pieces of paperwork. It is managing a hundred simultaneous six-week clocks, each with a commercial cliff-edge.

What is not verified is how many internal hours one variation consumes. I looked for a published UK dataset and found none — not from GIRI, CIOB, RICS, NEC, ONS or gov.uk. There is also no statutory framework for variations at all: the Construction Act says nothing about how a change is instructed, valued or priced.

So here is a model with every input stated — my arithmetic, not a published statistic. For one moderate variation: site manager assembling evidence, 1.5 hours; quantity surveyor measuring and pricing it, 3 hours; project manager reviewing and submitting, 0.75 hours; responding to queries, 1.5 hours; commercial team maintaining the register and chasing the reply, 1.25 hours. That is 8 hours per variation, or £336 at an indicative fully-loaded blended £42 per hour — so a contractor processing 120 a year spends roughly 960 hours, about £40,000, purely moving variations through the cycle.

Small against the £210,000-per-£1m error bill, yet it decides whether you recover the money, because the entitlement disappears at eight weeks whether or not the claim was good. It is the margin leak I have written about in the sales-to-operations handoff: value is lost not in the work but in the paperwork that proves the work happened.

Inadequate contract administration is the number one cause of disputes

Here the number is measured, and remarkable. The current UK data is the third and final report in the King's College London and Adjudication Society series, based on ten adjudicator nominating bodies and 166 individuals (Nazzini and Godhe, KCL, November 2024). Referrals hit 2,264 between May 2023 and April 2024 — the highest since statutory adjudication began in 1998, up 9% year on year (p.17).

The causes: "The two leading causes of disputes in construction adjudication in the past year are inadequate contract administration, at 50%; and lack of competence of contract participants, at 42%", followed by exaggerated claims and changes by client at 30% each. Half of all formal construction disputes are attributed by practitioners not to bad building or bad design, but to paperwork handled badly.

The claim categories agree. "Smash-and-grab" technical payment claims — where a party recovers a sum because the correct payment notice was not served in time — were the most common by a wide margin at 63%, followed by "true value" final accounts at 38%, variations at 26%, and defects at just 14% (p.28). Defects, the thing the industry calls its quality problem, sit near the bottom.

That 63% is driven by statute, not contract, which is why it bites whichever form you use. Under Part II of the Housing Grants, Construction and Regeneration Act 1996 as amended, a payment notice specifying the sum due must be given not later than five days after the payment due date (sections 110A, 110B), and the payer "must pay the notified sum" by the final date unless a valid pay less notice is served (section 111). Any party may adjudicate at any time (section 108).

A five-day notice window, applied to every payment cycle on every contract, is therefore the highest-leverage administration in the business. Missing it does not reduce your margin — it transfers the disputed sum to the other side.

Retentions: the rules are about to change

If you read one section for practical reasons, read this one, because the law is moving.

The baseline evidence is the Government's 2017 research: retentions average 4.8% of contract value and are released far later than intended — 12.5 months intended against 18 months actual, with 71% of contractors reporting delays (Pye Tait Consulting for BEIS, October 2017). DBT has since revised the total sharply upwards, to a central estimate of £8.4bn held across the UK in 2025.

After a consultation drawing 867 responses — 238 from construction — the Government decided to act: "Prohibition of retention payments: we propose to ban the practice of deducting and withholding of retention payments under the terms of a construction contract" (DBT, Time to pay up, March 2026). Respondents reported that when defects or insolvency actually occurred, "the amount withheld rarely, if ever, covered the costs."

That ban is being legislated through the Commercial Payments Bill [HL], at Report stage in the Lords as I write (Bill 4128). It is not law yet and has not been to the Commons. Exhibit 4 has the timetable.

Note where DBT says the benefit sits. Its impact assessment records a negative headline net present value, explaining this is "largely because not all benefits are monetised, including the reduced administrative burden and resulting cost and time saving by payees not needing to chase late payment of retentions." The biggest gain from abolishing retentions is administrative — and the Government could not put a number on it.

A reporting duty already bites: since 1 March 2025 qualifying large companies must publish their retention policies, rates and release mechanisms (SI 2025/75).

Being honest about construction and software

I work in AI advisory, so you would expect a tooling recommendation. Here is the honest position instead.

Construction is the hardest environment in the UK economy to put software into, and adoption reflects it: UK construction sits at 11% AI adoption against 19% for manufacturing. That gap is not stupidity. Sites are wet, loud and temporary, connectivity is unreliable, a 2.4% margin does not fund experiments, and 69% of CIOB members named lack of time as the main barrier to development in construction management roles (CIOB, 13 June 2025, 2,556 respondents). GIRI's own July 2025 report warns that "AI tools themselves can introduce errors, especially through hallucinations or incorrect assumptions. This is a particular concern in high-hazard environments such as construction" (GIRI, p.7).

So the useful starting point is almost never on site. It is the commercial office document flow — variation registers, notice deadlines, retention release dates — already digital, already deadline-driven, already the leading cause of your disputes. That is also where AI fails differently by sector: the failure mode is not model accuracy but that the process was never written down, which is why most AI projects fail.

Methodology and sources

What was checked, and when. All figures were read on the primary source in July 2026. GIRI reports were read as PDFs in full, not taken from the website summaries, which paraphrase the research in ways that change its meaning. ONS figures came from published time series and reference tables; statutory provisions from legislation.gov.uk; Bill facts from Parliament's own Bills data. All figures are sterling.

Measured versus derived. GIRI's 5% direct cost of error is derived from a literature review, corroborated by — not measured from — figures reported by Tier 1 contractors. The 21% total is derived by applying Delphi relative magnitudes (24% / 34% / 29% / 13%) to that anchor. Neither is a bottom-up measurement. Mine rather than published: all £-per-£1m figures, the 8.5× profit multiple, the cycle-length totals, and changes versus 2019 computed from published index levels. The 8 hours per variation, £42 blended hourly cost and 120 variations a year are an illustrative template, not a benchmark. GIRI's own sample: 19 companies interviewed, an 11-member Delphi panel, and 143 survey responses of which only 59–73 completed each detailed question.

What I excluded. The unattributed "10–25% of project value" claim, because it is GIRI's characterisation of other studies. GIRI's "£10–25bn per annum", because those totals imply a mid-2010s spend base of roughly £100bn. The claim that error costs Tier 2 contractors 20–25% of turnover, because GIRI states "the respondents were not able to provide evidence to support these statements." And the 2017 retentions total, superseded by DBT's 2026 figure.

What I could not verify. GIRI announced in May 2022 that it had commissioned research to update the annual direct cost of error (GIRI, 18 May 2022). I could find no published updated figure, so the April 2016 Research Report remains the most recent GIRI cost-of-error research, and any newer GIRI figure should be treated with suspicion until the source is produced. Contrary to common assumption, CIOB publishes no defects or cost-of-quality survey — no figure for the share of projects with defects, their cost, or who bears it, and no estimate of time lost to rework or admin. Figures circulating with a CIOB attribution generally originate in academic studies quoted in CIOB blog posts. I found no adjudication report later than November 2024, which KCL calls the third and final in its series.

What to do on Monday

Three things, none of which require buying anything.

Split your rework number in two. Pull last month's cost report and find the rework and defect lines. GIRI's composition says that total is roughly a quarter of your real error cost. Then pick one live project and ask site managers to log, for four weeks, every instance of work redone before handover, with an hour count. GIRI could not obtain that data from nineteen major contractors. Produce it for one project and you know something your competitors do not.

Time your variation cycle against the clock. Take your last twenty variations and record elapsed days from instruction to implementation, and internal hours consumed. Compare the first against the six-week cycle in Exhibit 3, and use the second to replace my 8-hour assumption. Then count how many were notified inside eight weeks — that decides whether the entitlement was ever recoverable.

Audit your notice and retention dates. Smash-and-grab claims are 63% of adjudications, and below roughly £30,000 adjudication is not economic, so small sums stay unrecovered. Build one list of every payment notice deadline and retention release date across your live contracts, with a named owner. And if the Commercial Payments Bill passes, retention clauses signed after the transition period will be void — the contracts you are pricing now will outlive the current rules. To find the rest of the hours, that is what the hidden waste audit produces — or book thirty minutes and we will map your variation and payment flow.

Related Reading

• The UK Manufacturing Productivity Gap: 40 Statistics (2026)
• AI Adoption in UK Manufacturing: The 2026 Benchmark Report
• The Hidden Cost of Manual Processes in UK Manufacturing
• Ghost Workflows: The Hidden Manual Tasks Nobody Recorded
• Why AI Fails Differently in Manufacturing, Logistics and Services
• The Sales-to-Ops Handoff Where Margin Disappears
• Why 70% of AI Projects Fail — And What the 30% Do Differently

Frequently Asked Questions

How much does rework cost in UK construction?

The Get It Right Initiative estimates the total cost of error at 21% of construction spend, of which 5 percentage points is recorded direct cost (GIRI, April 2016). On a £1m project that is £210,000 — £50,000 visible in cost reports, £160,000 in indirect costs, unrecorded process waste and latent defects.

Is the "rework costs 10–25% of project cost" figure accurate?

It is widely misattributed. That range is GIRI's summary of the international literature, not its own finding. GIRI's own estimate is a single figure, 21% of construction spend, with the 5% recorded direct component stated separately.

What is the basis of GIRI's 21% figure?

It is derived, not measured. GIRI took a 5%-of-spend direct cost figure from a literature review, then applied relative magnitudes from a Delphi panel of eleven senior industry figures — recorded direct 24%, indirect 34%, unrecorded process waste 29%, latent defects 13% — to scale it to 21%.

What causes most construction disputes in the UK?

Inadequate contract administration, at 50%, followed by lack of competence of contract participants at 42% (King's College London and the Adjudication Society, November 2024). "Smash-and-grab" payment claims were the most common category at 63%; defects accounted for only 14%. Referrals reached a record 2,264 in the year to April 2024.

Are construction retentions being banned in the UK?

The Government has committed to banning them, through the Commercial Payments Bill [HL]. As at late July 2026 it is at Report stage in the Lords, not yet law. As amended in Committee, retention clauses agreed after a two-year transition would be void. DBT estimates £8.4bn was held in retentions in 2025.

How long does one variation take to administer?

Under the NEC4 Engineering and Construction Contract the cycle is six weeks when everyone performs — one week to confirm validity, three weeks for the quotation, two weeks for the reply — and up to fourteen weeks if notification runs to the eight-week bar. No published UK figure exists for internal admin hours per variation.