The world’s largest industry has the most to gain from AI.
Construction generates some 13 trillion dollars of gross output a year worldwide and accounts for about 9% of EU GDP. Eurostat’s national accounts show that an hour worked in EU construction produced about a tenth less in 2025 than in 2015 — the baseline against which any adoption figure should be read.
Les Insights sont publiés en anglais.
Construction is one of the world’s largest industries and the one whose measured performance has improved least. McKinsey’s assessment of 9 August 2024 puts the sector at some 13 trillion dollars of gross output in 2023, about 7% of global gross output, and labour productivity growth at 0.4% a year between 2000 and 2022 — against 2% for the total economy and 3% for manufacturing.
On 16 September 2025 the Royal Institution of Chartered Surveyors published a survey of more than 2,200 construction professionals worldwide. Forty-five per cent of their organisations reported no use of artificial intelligence at all. One per cent had scaled it across projects.
The two findings are usually presented as a contrast: a very large industry, a very small adoption rate. They are more useful read together. In construction, the baseline a firm competes against has barely moved in two decades.
In Europe, an hour worked in construction produces less than it did in 2015
The comparison can be made directly from Eurostat’s national accounts, without recourse to any consultancy estimate. Between 2015 and 2025, real value added in EU construction — chain-linked volumes, so price inflation is stripped out — rose by 6.5%. Hours worked in the sector rose by 18.3% over the same period. Output per hour worked therefore fell by about a tenth. For the EU economy as a whole, the same calculation gives a rise of about 7.6%.
Switzerland differs in level but not in direction. Swiss construction value added in 2025 stood 2.0% above its 2015 volume while hours worked were 0.6% below, implying a gain in output per hour of under 3% across ten years. The Swiss economy as a whole gained about 15% on the same basis.
The pattern is not confined to Europe. Austan Goolsbee and Chad Syverson, in NBER working paper 30845 of January 2023, document a decades-long decline in United States construction productivity and test whether measurement error accounts for it. On measures of physical productivity in housing construction, they find productivity falling or, at best, stagnant over multiple decades, and conclude that measurement error is probably not the sole source of the stagnation.
The sector’s weight is what makes those decimals consequential. The European Commission puts construction at about 9% of EU GDP and 18 million direct jobs, with up to 95% of construction, architecture and civil-engineering firms classified as micro-enterprises or SMEs. In Switzerland, Eurostat’s national accounts record 355,500 people employed in construction in 2025, close to 6.6% of domestic employment.
The RICS survey shows an industry that has started almost everywhere and finished almost nowhere
Expectation is not the constraint. In the RICS survey, close to 70% of project managers and quantity surveyors believe AI will help them deliver greater value, and 40% expect the largest impact over the next five years to fall on design optioneering, with scheduling, risk management and cost control also prominent.
Commitment is more divided. A quarter of respondents — 25% — intend to increase AI spending over the next twelve months, 28% have no plans to do so, and 22% are unsure. Maureen Ehrenberg, acting president-elect of RICS, framed the task as ensuring AI is “adopted responsibly, ethically and in ways that deliver real public good”.
The number to carry into a management meeting is the distance between 70% expecting value and 1% having scaled anything. That gap is not evidence of scepticism. It is evidence that the step from a useful tool to a changed process has not been taken.
The barriers practitioners name are organisational, not technical
Asked what holds adoption back, RICS respondents put a lack of skilled personnel first at 46%, system integration second at 37% and poor data quality third at 30%. None of the three describes a shortage of available technology.
The reason sits in how the industry is organised. Each project is delivered by a temporary organisation — client, designers, main contractor, specialist subcontractors, suppliers — assembled for one building and dissolved on handover. Information crosses company boundaries as documents rather than as structured records, and the firm that generated a cost or a delay is rarely the firm that keeps the outcome data. Statistical systems, including the current generation of AI, learn from repetition; construction is organised so that repetition accumulates nowhere. That organisational fact, rather than any reluctance to buy software, is the most plausible reading of the 30% who cite data quality.
It also indicates where value appears first: in the parts of the work a single firm repeats on its own account. Tender pricing, the procurement of recurring packages, scheduling, site reporting, variation orders and the administrative tail that follows them are all processes in which one company holds its own history — and can therefore act on it without waiting for the rest of the value chain.
In Swiss construction, costs have moved and volumes have not
The price side has been anything but static. Eurostat’s construction producer price index for Swiss residential building stood at 96.1 in 2015 and 111.1 in 2025 on a 2021 = 100 base — a rise of roughly 16% over a decade in which the sector’s real output was essentially flat. In the EU, construction costs rose 5.8% in 2021, reached double-digit annual rates in most member states in 2022, and were still climbing by 2.3% in 2024, while production volumes in building construction turned negative in 2023 and more clearly in 2024.
For a Swiss contractor, planner, façade specialist or equipment supplier, the reading is straightforward. Neither volume growth nor price restraint is available as a source of margin. What is available is execution: the hours consumed by tender preparation, by rework, by chasing information between parties, and by the documentation that surrounds every project. Those are the variables a single firm controls, and they are the ones on which the available tools already work.
What a management team can decide before its next tender
Nothing in the evidence suggests that the firms capturing value bought better technology. The decisions below are ordinary management decisions.
- Pick one process the firm repeats every month — tender pricing is the usual candidate — and measure it before changing anything: hours per tender, tenders answered per quarter, win rate, and the gap between estimated and final cost.
- Inventory the firm’s own records before evaluating any product: past tenders won and lost with their prices, supplier quotations, hours booked per work package, defect and rework logs. Nothing that follows works without them.
- Budget for skills ahead of licences. Lack of skilled personnel was the leading barrier in the RICS survey at 46%, ahead of integration at 37%; a subscription without trained users produces neither.
- Write the confidentiality rule first. Client drawings, tender documents and pricing are normally contractually restricted; the question of what may be entered into an external system should be settled in writing before the first upload, not after.
- Agree in advance what would count as success and by when — a defined reduction in hours per tender, or in the interval between site report and invoice — so that the decision to extend or stop is made on evidence.
- Ask two or three principal clients what data format they will require at handover on the next project. Their answer determines how much of this becomes a contractual requirement rather than an internal option.
What to watch, and where it will show
Three markers are worth a place in the calendar. A further edition of the RICS survey would be the first place a movement away from 1% scaled adoption became visible. Eurostat’s national accounts are updated each year, which means the productivity gap between construction and the total economy can be recomputed from public data rather than inferred. And Eurostat’s statistical overviews of construction costs and production volumes are scheduled for their next update in October 2026.
The sector’s record is, in commercial terms, the opportunity. Where a competitor’s productivity has been flat for a decade, an improvement that would be unremarkable in manufacturing becomes a difference in tender price and delivery date. That is an unusual position, and not one that can be assumed to hold indefinitely.