Rolling Out AI in a Construction Company, A Step-by-Step Guide for Swiss SMEs (incl. 2026 Security Playbook)

The construction industry is at a turning point in 2026

AI tools like ChatGPT, Claude and Gemini are already on the building site, in the estimating department and in the back office, most of them via personal accounts. Quotation cover letters are drafted, tender documents summarised, cost calculations sanity-checked, variation orders written. In roughly 80% of Swiss construction companies this is already the everyday reality, only the management rarely knows on what scale.

The decisive question that almost nobody asks: Do you know what data your employees are entering into public LLMs right now? And more importantly: are you aware that your competition could, in theory, end up with insight into your costing logic?

This guide walks through, step by step, how Swiss construction companies can introduce AI professionally, securely and compliantly, without endangering trade secrets and without falling behind the competition.

Why AI in construction is unavoidable today

In 2026, construction companies are dealing with a combination of pressures, each of which would be heavy on its own, and together becomes crushing:

• Rising administrative burden. SUVA notifications, EKAS rules, BUV documentation, VAT-compliant construction invoicing, 30–40% more paperwork per project than ten years ago.
• More complex tender procedures. Public tenders demand sustainability evidence, CO₂ figures, subcontractor lists with collective-agreement compliance statements. That doesn't get written between two phone calls.
• Margin pressure. Material prices fluctuate, subcontractors are scarce, and the client still expects the 2022 price.
• Skills shortage. Foremen, site managers and estimators with ten years of experience are barely findable. The ones you have age faster, and walk out the door with their know-how.
• Knowledge loss at retirement. Your seasoned site manager knows which subcontractor in region X is reliable and which supplier delivers on time for special formwork. Is any of that written down anywhere?

AI can help concretely against every one of those points: quotation and variation-order drafting, summarising site minutes, automating the site journal, accounts-payable checks, expense and payroll handling, and systematic capture of internal knowledge. But: how you introduce it decides whether it becomes a risk or a competitive advantage.

The big risk: public LLMs inside a construction company

Today, employees in a typical construction SME use:

• ChatGPT (Free or Plus, on a personal account)
• Claude Consumer
• Google Gemini
• Microsoft Copilot with a personal Microsoft account

The problem starts the moment a project manager types something like this into the chat:

• Unit prices from your costing template
• Internal margin and surcharge assumptions
• Subcontractor discounts and supplier conditions
• Variation-order strategies vis-à-vis a specific client
• Margin models and contribution-margin calculations

From that moment on, the very information that distinguishes your company from the competition has left your controlled IT environment. Even if the provider insists data is not actively used for training, consumer tiers often come with different data terms than enterprise plans. Logs are stored, content can be used for model improvement, and employees simply work around internal policies because the tool is convenient.

Your competition doesn't need to hire a hacker. It only needs to understand how models are trained. When ten Swiss construction companies enter similar costing patterns into the same public model, an implicit industry knowledge emerges that nobody controls exclusively anymore.

The question is not academic. It is a direct consequence of how we prevent data leaks in the AI era, or fail to.

Step 1: Make Shadow AI visible

Before you officially introduce AI, you need to know what is already happening inside the company. Concretely: who uses which tools? On what accounts? With what data?

In practice, we recommend three low-friction measures before you buy any new tool:

• Internal survey. Anonymous, one page, five questions. "Do you use AI tools at work? Which ones? For what? How often?" The answers surprise every management team.
• Awareness workshop. 90 minutes with concrete construction-industry examples: what happens when a tender document ends up in ChatGPT, what happens when a staff appraisal is summarised in Claude.
• A clear, short policy on public LLMs. One A4 page maximum. "What goes into which tool, what must never go into a public tool?" This kind of policy gets lived, not filed.

This step is not bureaucracy. It is the precondition for every subsequent step to actually work.

Step 2: AI governance before tool selection

This is where the most expensive mistakes happen. Many construction SMEs do it like this: pick a tool, hand out accounts, manage the chaos. The right order is the other way round.

Define first:

• Roles. Who is allowed to use AI for what? Site managers for quotation text, accounting for invoice review, management for strategic analysis?
• Data classification. Which data is public, internal, confidential, strictly confidential? In construction, typically sensitive: tender data, costing models, HR files, payroll, contracts, subcontractor conditions.
• Audit logging. Which inputs are logged? For how long? Who can read the logs?
• Access control. Who can actually enter which data into an AI tool? Enforced technically, not only by policy.

This governance work is the foundation for everything that follows. Skip it, and you build on sand. For a feel of what this looks like in real life, see our piece on the Swiss AI Governance Gap.

Step 3: A secure platform instead of a tool patchwork

A typical Swiss construction company in 2026 works with this tool landscape:

• Bexio, Abacus or Sage for bookkeeping
• Outlook for email and calendar
• SharePoint or a network drive for documents
• WhatsApp for fast site communication
• Excel for everything the other tools can't do
• ChatGPT for the AI tasks nobody is talking about openly

The result is data silos, loss of control, compliance risks and just plain friction in daily work. A foreperson hopping between six tools is not more productive, she is more tired.

The answer is an integrated Swiss AI platform that brings the key building blocks under one roof: chat assistant, knowledge management, ERP, payroll. That is exactly what hegyon.ai is built for.

Why hegyon fits construction SMEs

🇨🇭 Data stays in Switzerland

Hosted at Hostpoint in Zurich, no US cloud provider, no extraterritorial legal uncertainty, no CLOUD Act through the back door. When SUVA asks where your HR files live, the answer is: in Switzerland.

🔐 EU AI Act guardrails

Content and intent classification with a four-tier logic: Block (clearly problematic inputs are stopped), Warn (employees get a heads-up before passing sensitive data on), Transform (PII is anonymised automatically) and Allow (anything harmless runs through without friction). That is not a marketing line, it is EU AI Act conformity applied in daily use.

🧾 ERP built in, aligned with Swiss law

• Receivables with QR-Bill in the SIX standard
• Payables workflow with Pain.001 export and CAMT.053 reconciliation
• Payroll based on ESTV rules (Lohnausweis form 11)
• Default interest under OR Art. 104 (5%)
• Swiss VAT rates and filing

🏗 Construction-specific advantage: knowledge management

Construction companies lose a huge amount of know-how at retirements, project changes and subcontractor switches. Hegyon includes an integrated wiki (process manual) with version history, a tamper-resistant audit chain, 15 ready-made Swiss-SME process templates and a RAG link to the AI assistant.

The crucial part: the AI assistant uses your internal knowledge as context, not the publicly trained industry knowledge. When the new site manager asks about the right approach for a renovation in a listed building, the system answers with the experience of your house, not with an average drawn from the open internet.

Step 4: Multi-provider AI instead of dependency

With hegyon you can use OpenAI, Anthropic, Google and Mistral in parallel. You as the management decide:

• Which model for which task? (Quotation text → Anthropic, cost-calculation review → OpenAI, image analysis → Google.)
• Which data may go where? (Payroll never to US providers, generic text everywhere.)
• Which prompts get logged? (Anything above the critical thresholds.)

No black box, no vendor lock-in. If one provider changes its data policy tomorrow, you switch without touching your business processes. More on that in our piece on RAG vs. CAG vs. Agentic RAG.

Step 5: Integration into existing construction processes

Three concrete use cases from the construction day-to-day so this doesn't stay abstract:

📑 Quotation and tender drafting

The AI assistant generates, on the basis of your costing and your internal knowledge: cover letters, fully worded scopes of work, summaries of bulky tender documents, variation-order rationales. The key point: the model works with your internal costing knowledge as context, the content stays in-house.

💰 Accounts payable

QR-Bill is captured automatically, an expense account is suggested, the invoice is routed through a 4-eyes approval workflow and finally bundled into a Pain.001 file for the next payment run. What that looks like in detail is in our piece on the Swiss SME accounts-payable process.

👷 Payroll

Swiss payroll module with AHV/IV/EO/ALV, BVG integration, correct Lohnausweis in ESTV form 11, a correction-pay wizard and audit-proof archiving. A construction company with 30 employees, a mix of hourly and monthly wages, collective-agreement-relevant, the module is built precisely for that case.

📚 Process manual and knowledge management

15 ready-made SME process templates to fork, adapt and version. Construction-specific processes such as site setup, SUVA notifications, handover, final invoice and warranty handling can be documented on this basis, and become a knowledge base for the AI assistant.

Step 6: Onboard employees the right way

Historically, AI projects don't fail because of technology, they fail because of culture. Four success factors we have seen in every successful construction-SME rollout:

• Transparency. Don't build up fear of surveillance. AI is a tool, not a control instrument against the workforce.
• Training. 3–5 hours of initial training, split across two sessions. Short on theory, long on practice.
• AI champions per department. One person in estimating who genuinely likes the tool, one in site management, one in admin. Those three become the internal point of contact.
• Clear monitoring boundaries. What is logged, what isn't? In writing, transparent, agreed with the workers' representation.

What does it cost to not have a secure AI strategy?

The risks of rolling out AI without governance are not theoretical. They are concrete, frequently four-digit in Swiss francs, and sometimes existential:

• Trade secrets inside a public model. Once entered, hard to take back. Costing models that made the difference for years become common knowledge overnight.
• EU AI Act violations on EU projects. Anyone working for a German or Austrian client falls under EU AI Act obligations. Breaches can cost up to 7% of annual turnover.
• Data protection breaches under FADP. HR files, payroll, health data of employees in a public chat, that is a reportable event for the FDPIC.
• Reputation damage. A single incident in which it becomes clear that client data ended up in an uncontrolled model ends every trust relationship with the general contractor.
• Inefficient tool landscape. Seven tools, seven logins, no shared data. The hidden hours add up.
• Knowledge loss. Whoever does not document systematically today has a problem in five years that can no longer be trained out.

And the biggest risk: your competition is professionalising, while you are experimenting. The first Swiss construction groups have already taken the step. We documented one example in our piece on going from Shadow AI to governed intelligence.

Bottom line: AI in construction is not a tool topic. It is a governance topic.

Public LLMs are powerful. But unmanaged use in a construction company is dangerous, not because the models are malicious, but because the data flows become opaque.

The question is no longer: "Are we using AI?" It is: "Are we using it securely, in an integrated way, and strategically?"

Next step

If you want to find out how secure your current AI usage really is, where Shadow AI already lives inside your company, and what an integrated AI platform for your construction SME could look like, book a no-strings 30-minute demo of hegyon.ai.

In that half hour you get:

• An assessment of your current tool landscape
• An evaluation of your compliance risks under FADP and EU AI Act
• A concrete implementation roadmap for your construction company

Swiss hosting. Swiss law. Swiss SME focus. Book the demo via the button at the top of every page, or directly at hegyon.ai/en/contact.