Why Microsoft Copilot Disappoints Many SMEs, Especially in Construction
Microsoft 365 Copilot is a powerful productivity layer in Word, Excel, Outlook and Teams. When Swiss SMEs expect a corporate copilot for real business processes, a gap opens up, and in construction it becomes especially visible.
The expectation is bigger than the product promise
There's a recurring moment across many Swiss SMEs over the past twelve months. Management sits through a Microsoft presentation, watches Copilot summarise slides, draft emails, explain Excel tables, take Teams meeting notes, and signs the licence agreement. Six weeks after rollout the sobering feedback comes in from the workforce: "Yes, it writes emails. But my cost estimates, my subcontractors, my quotation process, it still can't find any of that."
This isn't an attack on Microsoft. Microsoft describes the product exactly that way itself: Microsoft 365 Copilot is a productivity layer inside the M365 apps, grounded in Microsoft Graph, in existing permissions, and in a properly prepared data foundation. Microsoft explicitly recommends a readiness assessment, data governance, adoption training and ongoing usage monitoring before rollout. The disappointment lives in the gap between that product promise and the SME expectation of "a copilot for everything".
What Microsoft Copilot actually is
Copilot is excellent at what it was built for: a Microsoft 365 environment. It summarises Word documents, explains Excel tables, drafts Outlook replies and takes Teams meeting notes, precisely, quickly, with whatever permissions the user already has in the M365 tenant. Inside the Microsoft ecosystem it is a real productivity gain.
Why it lands as "not enough" in many SMEs
In a Swiss SME, business-critical information rarely sits exclusively in M365. It lives in an ERP, in a CRM, in industry-specific systems, in cost-estimation files, in sub-folders on a network drive, in WhatsApp chats with subcontractors, in the heads of the foreperson and the site manager. Copilot sees none of this, not because it's a bad tool, but because it isn't built for it.
The five typical disappointments in everyday SME use
Talk to several Swiss construction, trustee and trades SMEs about their first six months with Copilot, and the same five points come up every time.
1. No complete business context
Copilot can only use what it can see. If your quotation system lives in an Excel file on the network drive, if the costing logic sits in an internal tool, if subcontractor terms are in a separate database, Copilot stops at the door. It helps with phrasing, but it doesn't know what surcharge you apply to this particular client, or why.
2. AI doesn't fix a permission mess
Copilot uses the permissions that already exist in SharePoint, OneDrive and Teams. If those permissions are historically grown, sloppy or overly generous, and that's the reality in most SMEs, Copilot suddenly surfaces content that employees were technically allowed to see but had never bothered to search for. Oversharing becomes visible without anyone meaning it to. AI doesn't turn a bad permission structure into a good one.
3. External systems take real integration work
ERP, CRM, industry software, accounting, QR-Bill platforms, these systems are rarely native to Microsoft Graph. Microsoft offers connectors, but every connector is its own integration, licensing and maintenance topic. "Let's connect Copilot to Bexio" sounds like two weeks of work; in reality it's a project with requirements, tests, training and ongoing costs.
4. Governance and training are underestimated
If employees get Copilot but no clear policy on what they may or may not enter, the exact risks that the revised Swiss FADP protects against start showing up. Microsoft itself stresses training, welcome emails and usage monitoring in its readiness documentation. That work doesn't appear in the licence price, and isn't budgeted for in most SMEs. Cutting that corner gets paid back later in audit, reputation or compliance terms.
5. Swiss legal and data-protection questions stay yours
Even a big vendor doesn't release you from your obligations as a data controller under the revised FADP. Cloud processing, cross-border data transfers, data-protection impact assessments for high-risk processing, surveillance limits in the employment relationship, all of that stays with you. The FDPIC's published positions on generative AI make that clear. For anyone working with EU clients, the AI literacy obligations under the EU AI Act apply on top, in force since 2 February 2025. More on that on our EU AI Act page.
Why this becomes especially visible in construction
In Swiss construction the five points turn into tangible pain, because critical data is particularly unlikely to sit cleanly in M365.
Costing, quotations and Excel reality
Most Swiss construction firms cost in Excel. Templates grown over years, individually maintained, with subcontractor rates in one tab, surcharges in another, project experience in a third. A Copilot that opens this file inside M365 can phrase things, but it doesn't know the logic. It doesn't know you apply a 12% risk premium to this particular client because the last project generated three variations. That story isn't in the cell.
Site know-how and house rules
One point becomes increasingly clear in Swiss construction: answers about construction get better when only vetted, current and industry-specific sources are used, not the open internet. A pure app-copilot without a controlled knowledge base hits its limits exactly here. It doesn't know your internal SUVA processes, your defect-notice templates, your warranty workflow with a specific general contractor. It knows an average pulled from the internet.
Back office, HR and reporting
The first productive AI gains in a construction SME usually don't sit on the site, they sit in administration: incoming QR-Bill invoices, expense photos, time-tracking queries, salary statements, supplier approvals. Those processes live in accounting software, in industry-specific modules, sometimes still in paper workflows. A copilot that stays outside those systems helps you write the response email, not the actual process. What this can look like differently we wrote out step by step in Rolling out AI in a construction company.
What Swiss SMEs actually need
The answer isn't "Microsoft is wrong, buy something else". It's more nuanced: Copilot is strong where it was built, inside M365. For the processes that live outside M365, an SME needs a different layer.
An employee copilot, not just an app-copilot
The difference is conceptual. An app-copilot lives inside the applications (Word, Excel, Outlook) and helps you use those applications more productively. An employee copilot lives inside the business processes, it knows your quotations, your subcontractors, your employees, your suppliers, and supports you where you actually lose time. Both have their place; they aren't competitors.
Work context, roles, audit, Swiss hosting
A well-built employee copilot combines four properties: it knows your work context across business modules (ERP, CRM, HR, wiki); it respects roles and permissions without inheriting them from historical M365 growth; it logs every action in a tamper-resistant audit chain; and it hosts your data in Switzerland under Swiss law. Those properties are documented openly on the hegyon.ai Trust Center, including the sub-processor list and DPA on request.
Microsoft Copilot vs. hegyon.ai
Here is the sober comparison, not to talk one down and the other up, but to make clear what each tool was built for:
| Comparison point | Microsoft Copilot | hegyon.ai |
|---|---|---|
| Primary context | Strong inside the Microsoft 365 context, grounded in Microsoft Graph | Work context across business modules and assistants |
| Permissions | Uses existing M365 permissions; quality depends on a clean rights structure | RBAC, audit trail and data classification as product properties |
| External business systems | Connectors or agents per system needed, their own project work | Tightly coupled with operations modules (sales, purchasing, banking, HR, CRM) |
| Model strategy | Microsoft-controlled model environment | Multi-LLM orchestration: OpenAI, Anthropic, Google, Mistral per task |
| Swiss fit | Data residency and effective configuration tenant- and plan-dependent | Swiss hosting at Hostpoint Zurich, Swiss-process fit, Swiss jurisdiction |
| Perceived value | Productivity in Word, Excel, Outlook, Teams | Employee copilot close to ERP, CRM and process work |
The honest reading: a well-configured Swiss SME will use both layers. Copilot for the M365 productivity, an employee copilot for the processes that live outside that. The question isn't "Microsoft or hegyon", it is "which layer for which pain".
How to test AI in 6 to 12 weeks without a bad purchase
Anyone who wants to genuinely evaluate Copilot or an employee copilot should plan six to twelve weeks, not for a tool evaluation, but for a controlled pilot.
Clarify data and roles
Before a single account is handed out, a one-page AI policy belongs on the table: allowed data classes, prohibited data classes, responsible owners, logging policy. This isn't bureaucracy, it's the precondition for training to make any sense. Skip it and you're building on sand, whether the tool is Copilot or anything else.
Pick two construction use cases
Not "AI for everyone". Instead, two concrete processes where time is measurably lost today. Ideally: one office process (quotation text, meeting minutes, payables review) and one knowledge process (subcontractor terms lookup, defect notice preparation, SUVA workflow explanation). That way you see both the efficiency and the governance sides.
Train employees and create a live FAQ
90-minute modules per role with real use cases for the role. Site management trains on real quotations, costing trains on anonymised tender examples, the back office trains on real invoice layouts. An internal FAQ is maintained live, not written for the launch and then never touched again. More on the human side in Rolling out AI without losing your people.
Measure, scale or stop
After 30 days, measure: which use cases are running in production? Which follow-up questions come up? Which error patterns repeat? From there, three options, scale, adjust, or honestly stop. The worst option is "keep going because we already have licences". An honest stop decision is cheaper than a permanently mediocre rollout.
Next step for Swiss SMEs
If you already run Microsoft Copilot, don't switch it off, it's a real productivity layer inside M365. But it shouldn't be the only AI layer in your business. For the processes that live outside M365, and in a Swiss SME that's most of them, you need an employee copilot that knows your work context.
If you want to see what that looks like in the everyday of a construction SME, quotation text, accounts-payable workflow, payroll, knowledge management in one interface, Swiss hosting, audit chain, multi-provider LLM, book a 30-minute demo. You don't get a slideshow; you get a real platform with real data.
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.