The construction sector is digitising fast. Site management apps, digital quoting tools, speech recognition for daily reports — the offer is wide and the promises are real. But most contractors face a choice that rarely comes up in the sales pitch: does your business data go to an external provider’s servers, or does it stay with you?
AI has clear applications in construction. Writing quotes takes time you’d rather spend on site. Site reports are repetitive work. Tender documents follow the same structure every time. The efficiency gains are real. The risk is not in AI itself — it is in where that AI runs and what data passes through it.
Your data is more complex than it looks
As a contractor, you process three types of data simultaneously: your clients’, your subcontractors’, and your employees’. That combination is not unusual — it is the standard reality of the construction sector — but it makes data management considerably more complex than in most other SME sectors.
A private client’s name and address are personal data under the GDPR. A site attendance sheet with your workers’ names and arrival times is personal data. A LIMOSA certificate for a foreign subcontractor is personal data. Those three streams run through your business every day — and every app you use to manage them is a processor that requires a data processing agreement under the GDPR.
Fewer than twenty per cent of construction companies have a formal data processing agreement in place for the site apps they use daily. That is not negligence — it is an information gap. Apps are sold on functionality, not on their data processing practices.
In Belgium, LIMOSA legislation requires contractors to pre-register foreign workers and subcontractors before they start work. The responsibility does not rest solely with the foreign party: as main contractor, you share liability if a foreign subcontractor is not registered when the social inspectorate visits. In the Netherlands, the Chain Liability Act (Wkba) creates documentation obligations across the subcontractor chain for wage tax and social security contributions. Both systems generate administrative flows that need managing — and that is exactly where AI can make a difference, provided the data sits on the right system.
What can go wrong with a site app
Imagine: you migrate to a modern site management app with integrated AI for planning optimisation and reporting. Workers clock in and out via the app, site managers dictate their daily reports through speech recognition, and you enter client data for the quoting module. Convenient — until you read the fine print.
Your workers’ attendance and location data flows continuously to the provider’s external servers. Your site managers’ voice input — project details, supplier conversations, problem descriptions — is processed by cloud-based speech recognition outside your organisation. Client personal data and project descriptions you enter into the quoting module are absorbed into the provider’s AI training process. And if a worker representative requests access to the location data processed via the app, you have no direct access to that data at the cloud provider.
In a tender procedure, the client asks for the data processing agreement covering tools that manage project data. That agreement does not exist.
“Writing quotes takes time you’d rather spend on site. Site reports are repetitive. Tender documents follow the same structure every time. AI handles all of that — quickly and reliably. The one thing we avoid is sending that data to an external server. Client addresses, project details, employee attendance — that’s your information. A local system keeps it with you.”
The AI Act adds an additional dimension for contractors considering site software with camera analysis. Systems that analyse camera footage to detect unauthorised persons, monitor evacuation routes, or log site access using image recognition potentially fall into the high-risk category of AI Act Annex III. A security camera that records is not AI. A system that analyses that footage — that is a different matter.
Local AI: same efficiency, your data
Local AI runs on your own hardware or server. Nothing goes to external systems. The efficiency gains are identical to cloud solutions — but ownership of the data stays with you, demonstrably and directly verifiable.
For contractors, three applications deliver immediate value:
Quotes: AI generates a quote based on your project description, your historical pricing data, and the client’s specifications. A quote that used to take two hours is ready in twenty minutes. If you write twenty quotes per month, that is forty hours per month freed up for work on site. Your margin rates, your subcontractor discounts, and your strategic project choices never leave your system.
Site reports: a site manager dictates notes at the end of the day. Local AI structures these into a complete site report — date, work performed, parties present, issues observed, next steps. Site reports serve as legal evidence in disputes about work performed. They belong on a system you control directly and whose contents you can demonstrate immediately if a dispute arises.
Chain documentation: AI helps maintain your subcontractor chain records — who is on site, for whom, with what LIMOSA status or Wkba documentation. This is a compliance task that prevents concrete fines and is immediately demonstrable when the inspectorate visits. Confederatie Bouw (BE) and Bouwend Nederland (NL) publish sector guidance on these obligations.
Three steps to get started
Step 1 — Map your data flows. Which apps and tools currently process client, employee, or subcontractor data? Which of them have a processing agreement in place? The answers to those two questions give you a clear picture of your actual data risk.
Step 2 — Choose AI for one task. Do not start with a full platform. Choose the task that costs you the most time — quotes, site reports, or tender documents — and implement a local AI solution for that one task. Measure the time saving after four weeks. The return on investment in construction is typically visible quickly.
Step 3 — Make the processing agreement a standard requirement. Every tool that processes personal data requires a data processing agreement. Request it before you put a tool into use. If a provider cannot supply a processing agreement, that is a clear signal about how they handle your data.
AI in construction is not futurism — it is an efficiency tool with a concrete and measurable return. The only thing that separates the efficiency tool from a risk is the question of where your data ends up.