Construction is one of the largest sectors in the world economy, with spending equivalent to 13% of global GDP, and one of the least digitised: its productivity has grown just 1% a year over two decades, against 2.8% for the world economy. Closing that gap would be worth $1.6 trillion a year (McKinsey Global Institute). In Belgium, the sector has 16,576 open vacancies and the highest vacancy rate in the country (Embuild). Squeezed between labour shortages, thin margins and safety requirements, AI is arriving on site: planning, BIM, computer vision, safety. Here is what already works, and how to go about it.
Construction's paradox comes down to two numbers. On one side, massive economic weight: around $10 trillion in annual spending, 13% of global GDP and 7% of the world's working population. On the other, stagnant productivity: 1% annual growth over the past two decades, against 2.8% for the world economy and 3.6% for manufacturing. The McKinsey Global Institute estimates that closing this gap would add $1.6 trillion in value a year, the equivalent of half the world's annual infrastructure needs.
Belgium illustrates the tension: 16,576 open vacancies in construction and installation, a 7.07% vacancy rate, the highest of any sector, 73% of companies recruiting and 87% describing that search as difficult or very difficult (Embuild, August 2025). When labour is structurally scarce, every hour of an engineer, site manager or estimator recovered through automation counts double.
Unlike a factory, a job site runs on teams reassembled for every project, dozens of subcontractors and scattered data: PDF drawings, emails, photos, spreadsheets, paper reports. Research by the Get It Right Initiative puts the direct cost of avoidable error at around 5% of project value, and up to 21% once indirect and hidden costs are included. That is exactly the deposit AI goes after, errors, rework, waiting, provided the data is structured first.
Four families of use cases are mature or being rolled out today, from the design office to handover.
AI produces nothing reliable on top of messy data. In construction, the foundation is called BIM: a shared digital model that becomes the project's single source of truth. AI models plug into it to detect clashes between trades, check design rules, compare as-built reality with the design using 3D scans and photos, and prepare the asset's maintenance.
Extended beyond handover, the model feeds the building's or infrastructure's digital twin, which follows the asset in operation. The loop closes with real estate and building management: site data becomes the owner's digital asset.
In 2023, 3,298 people lost their lives in an accident at work in the EU, of which 24.0% in construction, the highest share of any sector (Eurostat). AI does not eliminate the risk, but it multiplies the eyes: detecting helmets and PPE, alerting when someone enters a danger zone, analysing reported near-misses to spot recurring patterns, helping prepare prevention plans.
One caution: filming and analysing workers is not a trivial use case. Algorithmic monitoring of workers touches the high-risk systems of the EU AI Act (employment and workforce management, Annex III) and the GDPR; in Belgium it requires information, proportionality and social dialogue. Design augmented safety with worker representatives, not against them: anonymisation where possible, a purpose strictly limited to safety, no automated sanctions.
The sector is coming out of the euphoria phase. In Autodesk's State of Design & Make 2025 study (over 3,500 leaders surveyed in 28 countries), 68% of construction leaders believe AI will enhance their industry, down from 80% a year earlier, and almost half (48%) expect it to destabilise the sector. Trust is being rebuilt on proven use cases rather than promises.
Above all, the study shows a widening gap: 82% of "digital leader" organisations feel positive about their financial performance, against 52% of beginners. The talent shortage remains the main brake: 55% of leaders cite the lack of skills as a barrier to growth, up from 43% in 2024. Finally, 94% of organisations say they are acting on sustainability and 63% already rely on AI to do so.
The risk is not trying AI, it is trying it without data or a framework. Four steps structure a deployment that holds up.
Inventory what exists: schedules, estimates, models, photos, reports, equipment data. Pick one or two quick-return, low-risk use cases, measurable on a pilot site.
A common data environment (CDE), naming conventions, an up-to-date BIM model, geolocated and dated photos. Without this foundation, even the most advanced model produces unverifiable answers.
One pilot site, simple indicators (hours saved, rework avoided, deadlines met), an honest before/after comparison, then extension to other projects, with training for field teams.
Classify each use case under the EU AI Act and the GDPR, especially if people are filmed or assessed. Document, inform, consult. Anticipated compliance costs less than forced remediation.
Start with documents and planning, not with cameras. Take-offs, specification analysis, progress reports, bid comparison: gains within weeks, with no personal data and no high regulatory risk. Monitoring use cases, on the other hand, are prepared with lawyers and worker representatives.
Belgium's construction and installation sector is already reinventing itself: Embuild stresses that AI, drones, virtual reality and 3D printing are making site work more efficient, lighter and safer, and that modular construction industrialises part of the process in the factory. For an SME, the point is not to robotise everything: it is to recover the hours lost to administration, estimates and coordination, and to make margins reliable project after project.
In practice: a general contractor can automate specification analysis and bid comparison; an installer can generate its progress statements and as-built files; a structural works company can track progress from photos and anticipate equipment failures. Each use case is best classified by regulatory risk, as we detail in our sector analysis of the EU AI Act.
No. The sector's problem is the opposite: labour is missing, with 16,576 open vacancies in Belgium and the highest vacancy rate of any sector (Embuild). AI first tackles office hours: planning, estimates, documents, coordination. On site, it assists (vision, safety, equipment) more than it replaces.
With a low-risk document or planning use case: specification analysis, assisted take-offs, progress reports, bid comparison. The return is measurable within weeks, no sensitive personal data is involved, and it is the opportunity to structure your data (CDE, BIM) before moving to computer vision or safety.
It is strictly regulated. Algorithmic analysis of workers falls under the high-risk systems of the EU AI Act (employment, Annex III) and under the GDPR: limited purpose, proportionality, informing the people concerned, impact assessment and social dialogue are required. A system aimed at collective safety (danger zones, PPE), anonymised where possible and without automated sanctions, is defensible; individual performance surveillance is not.
Automatic clash detection between trades, design rule checking, comparison between as-built reality (3D scans, photos) and the model, help generating schedules from the model and preparation of the operational digital twin. BIM provides the structured data models need; AI makes that data active.
Molderez Consult helps construction and installation companies in Belgium map their use cases, structure their data (BIM, CDE), pick the right tools and frame EU AI Act and GDPR compliance, from estimate to as-built file.
Assess my project