Employees at many Swiss companies have been using AI for some time, often through private free accounts with no control over company data. The real question, therefore, is which platform allows your teams to use AI securely every day. This article compares the two approaches based on the questions that management teams at medium-sized businesses are actually asking.
The key difference lies in the approach. Microsoft 365 Copilot is an assistant within Microsoft applications. It operates in Word, Excel, PowerPoint, Outlook and Teams and uses Microsoft Graph to access your emails, files, chats and calendar appointments. Its strength is precisely this integration into everyday work. Anyone who wants to summarise an email, create minutes from a Teams meeting or generate an initial presentation draft from a Word document can stay within the familiar application.
CompanyGPT takes the opposite approach. It is a standalone browser-based platform, similar to ChatGPT in how it is used, but operated as an isolated corporate environment. The platform brings several AI models together in one interface, integrates your company knowledge through knowledge bases and works regardless of which Office licences an employee has. We explain the fundamentals in our article CompanyGPT: Why Businesses Need Their Own Secure AI Environment Today.
Put simply, the two tools answer different questions. Copilot asks, “How can I get my Office work done faster?” CompanyGPT asks, “How can I give all employees controlled access to powerful AI?”
Microsoft has invested heavily in recent years. The EU Data Boundary has been fully implemented since February 2025, and Copilot is classified as an EU Data Boundary service for EU tenants. Customer data is therefore generally processed within the EU.
Two limitations are particularly important for Switzerland. First, the guarantee depends on the tenant, not on the company’s registered office. The SIDD Institute notes that Swiss subsidiaries of global corporations with a US tenant do not benefit from the EU Data Boundary.
Second, there are defined exceptions, such as during peak loads or for certain models, where requests may be processed outside the EU. According to Computerwoche, Microsoft has announced plans to process Copilot interactions in data centres within individual countries’ borders in the future. The direction is clear, and implementation is under way.
CompanyGPT treats data residency as an architectural principle. Every request is processed in Switzerland or the EU, never in the USA, regardless of the model selected.
No provider trains its models on the content, every use is logged, and customer environments are strictly separated.
Additional safeguards include PII redaction for personal data, protection against prompt injection and end-to-end encryption. Swiss-hosted models such as Apertus handle the majority of everyday requests.
For a long time, Microsoft 365 Copilot primarily used OpenAI models. According to Schneider IT Management, Anthropic models such as Claude have also been available since September 2025, and AI agents from Microsoft and its partners have been available through the Agent Store since early 2026.
The range of models is expanding but remains tied to the Microsoft ecosystem. In its GDPR guide, CodeKlar also notes that requests to Anthropic models may leave the EU Data Boundary.
CompanyGPT was designed from the outset as a multi-model platform. Available models include Apertus, Llama, Mistral, ChatGPT, Gemini and Claude, provided through Swiss or EU infrastructure depending on the model. An orchestration system automatically distinguishes between simple, medium-complexity and complex requests and selects the appropriate model. Simple requests run on cost-effective standard models, while only complex ones use the more expensive leading models. Users can manually select a particular model at any time. New versions of Claude, ChatGPT and Gemini are generally available five to ten days after their release.
In terms of usability, there is no need for concern. Copilot Chat provides a chat interface that feels similar to ChatGPT. Employees ask a question and receive an answer; in the paid version, answers can also draw on their emails and documents. The underlying difference is that Microsoft determines which model responds, and Gemini is not available at all.
From our projects and ongoing discussions with businesses, we know the pattern that tends to follow. Employee A prefers ChatGPT, employee B relies on Claude for writing, and employee C uses Gemini for research. Understandably, the company initially wants to provide just one tool, often Copilot because Microsoft contracts are already in place. But employees’ preferences do not disappear; they simply shift elsewhere.
The Microsoft and LinkedIn Work Trend Index shows where. Worldwide, 78% of AI users bring their own AI tools to work; in Europe, the figure is as high as 80%. For companies, this means that business data ends up in private accounts, without control over security, data usage or retention. The seemingly obvious solution of officially licensing every team’s preferred tools merely replaces one problem with another. Individual subscriptions to ChatGPT, Claude and Gemini quickly add up to substantial costs per employee, and every tool comes with its own security settings, data processing agreement and invoice. In practice, oversight is lost.
CompanyGPT resolves this dilemma differently. All leading models are already available: ChatGPT, Claude and Gemini run alongside Swiss standard models in the same tool, using the same secure infrastructure.
Employees can work with their preferred model without the company having to purchase separate licences or assess every provider’s security settings individually. One contract, one budget with spending limits, one dashboard.
With Copilot, the answer to the same question is different. Users who are satisfied with the models Microsoft provides are well served, and for purely Office-related tasks many users do not care which model is working behind the scenes. However, those who specifically want to use Claude can do so only in scenarios supported by Microsoft. Anyone who wants Gemini needs a separate subscription, returning to the individual-licence approach described above.
Model choice therefore becomes a cost and security issue when teams perform different types of tasks or employees have strong preferences based on their personal use.
Copilot uses Microsoft Graph to access everything for which the individual user has permission. This is its greatest strength and also its best-known risk. Copilot only displays content a user is authorised to access, but it can suddenly expose existing mistakes in sharing permissions.
Anyone who shared a folder containing salary information too broadly years ago will notice it at the latest when the AI quotes from it. The SIDD Institute and other experts therefore recommend reviewing and cleaning up permissions before rollout.
CompanyGPT takes a two-pronged approach. Through its SharePoint interface, it can already find all documents stored there and preserves the Microsoft permissions unchanged. If an employee cannot access a SharePoint area, they cannot access it through CompanyGPT either.
In addition, it provides curated knowledge bases using RAG (Retrieval-Augmented Generation), for individuals, teams or the whole company, with source citations in the chat history and safeguards against hallucinations. Individual employees’ chat conversations are not incorporated into other users’ answers.
Hardly any medium-sized company operates entirely within a single software ecosystem today. A typical technology stack includes an ERP system such as Abacus, SAP, Bexio or KLARA for orders, inventory and finance; Microsoft 365 for email, documents and collaboration; a CRM such as HubSpot for marketing, sales and service; and a project management tool such as ClickUp, Monday or Asana. Team knowledge is therefore distributed across at least four systems. This is precisely where the everyday usefulness of an AI tool is determined.
Copilot is at home in the Microsoft ecosystem. SharePoint, Outlook and Teams are its native environment, and integration is central to the product, including features directly within the applications, such as rewriting a paragraph in Word.
However, Copilot is no longer limited entirely to Microsoft. Third-party systems can be connected through Graph Connectors. Microsoft documents connectors for Asana and Jira, among others, which allow tasks from those tools to be searched and summarised in Copilot.
Two limitations remain. The connectors are designed for reading and indexing. Anyone wishing to take action from Copilot, such as creating a ticket, must build custom agents in Copilot Studio, with separate credit-based licensing. Not every tool is supported either; for example, there is currently no official ClickUp connector.
The same pattern applies to ERP systems. Microsoft documents its own connectors for SAP through Copilot Studio, but knowledge queries are geared towards read access, and setup requires data gateways and custom-built agents. In other words, it is an IT project. For Swiss SME solutions such as Abacus, Bexio or KLARA, no ready-made Copilot connectors exist, so businesses must develop their own integrations using the respective APIs.
CompanyGPT is designed for this mixed environment. Interfaces to HubSpot and project management tools such as ClickUp, Monday and Asana already exist, as does the SharePoint integration, which makes all documents stored there searchable while retaining the same access rights as in Microsoft.
If required, the ERP can also be connected via an interface, making order and customer data from the core system available in chat.
Knowledge from connected systems can be retrieved in chat, and CompanyGPT can also take action based on it, such as creating a new ticket directly in ClickUp. Outlook and Teams will be added as further sources with the October 2026 release.
The pricing models differ fundamentally, and that is exactly what makes the comparison interesting for management teams.
Microsoft charges per user. According to the Copilotenschule pricing overview (as of September 2026), Microsoft 365 Copilot Business costs EUR 18.20 per user per month for companies with up to 300 users (currently a promotional price of EUR 15.60), while the Enterprise version costs EUR 26.00. Both prices are based on annual billing and require an eligible base Microsoft 365 licence.
According to bluebatch, the basic Copilot Chat interface is included at no additional charge with eligible Microsoft 365 licences, but without access to the user’s own emails, documents or Teams chats. The same analysis calculates that total first-year costs, including base licences, tenant preparation, governance and training, are considerably higher than the add-on prices alone. The Fraunhofer FIT study “Is Microsoft 365 Copilot Worth It?” also emphasises the need to assess internal costs and benefits before implementation.
CompanyGPT charges per company. Depending on company size, the subscription costs between CHF 380 (up to 15 employees) and CHF 2,274 (up to 250 employees) per month and covers platform operation, maintenance, security, updates and support.
There is also a one-time setup fee of CHF 2,968 for companies with up to 15 employees and CHF 4,968 for companies with 16 or more employees. Integration of internal knowledge sources and SSO are excluded and billed according to the work involved. The minimum term is twelve months, followed by a three-month notice period. The subscription includes Swiss-hosted standard models and integration of up to 5 GB of company knowledge via RAG. Leading models incur usage-based token costs, with a guideline of CHF 0 to CHF 80 per employee per month depending on intensity of use. Budgets can be capped per company and per person, and a dashboard shows costs in real time.
Consider a deliberately simplified example of a company with 50 employees (the euro and Swiss franc are currently close in value).
At EUR 18.20 per user, Copilot Business costs approximately EUR 10,900 per year, plus the base licences, which are usually already in place, and one-time costs for permission cleanup and training.
CompanyGPT costs CHF 9,576 per year in subscription fees plus a one-time setup fee of CHF 4,968. Token costs are additional: with regular use involving only a few documents (guideline CHF 0 to CHF 10 per person per month), these amount to up to approximately CHF 6,000 annually, and more with intensive daily use. At moderate usage levels, the two solutions are therefore in a similar price range. The structural difference lies elsewhere.
With Copilot, every licensed user costs the same amount, regardless of whether they use the tool every day or never. With CompanyGPT, the company pays a base fee, while variable costs follow actual usage.
Model access is another difference: Copilot does not provide all leading models, while CompanyGPT lets employees choose between ChatGPT, Claude and Gemini according to their preferences and tasks.
Microsoft 365 Copilot is a good choice if your company works extensively within the Microsoft ecosystem, all relevant employees have an eligible M365 licence, and the main benefit sought is greater efficiency in Office work.
An adoption study published in the journal Zeitschrift für Arbeitswissenschaft shows that Copilot provides the greatest added value for clearly structured, text-based tasks and is perceived as user-friendly. A well-organised permission structure is essential for a successful start.
CompanyGPT is a good choice if, as a management team, you want four things at once.
First, risk control. Trade secrets, customer data and NDA-protected content remain in Switzerland or the EU without exceptions, removing the incentive to use shadow AI through private accounts.
Second, predictable costs. You pay per company rather than per person, set budget limits for the company and individual employees, and see expenses in the dashboard before the invoice arrives.
Third, access for the entire workforce instead of isolated groups of licence holders. Employees in production, logistics and field sales can also use AI without each person needing a Microsoft licence.
Fourth, independence. Your teams use the strongest model for each task, draw on knowledge from ERP, HubSpot, SharePoint and project management tools, and receive new model versions within days of release, without being tied to a single provider’s roadmap. You can find more information on the CompanyGPT page.
Start with a brief assessment based on four questions.
1. Which of our data needs protection, from personal information and customer contracts to trade secrets?
2. Which data residency and security requirements apply to us, for example under the Swiss Data Protection Act, customer NDAs or industry regulations?
3. Which systems do our teams actually work in?
4. And how are they already using AI, whether officially or through private accounts?
Anyone who has answered these four questions will usually quickly recognise which of the two approaches is the right fit, or whether a combination of both makes sense.
The good news is that there is little chance of making the wrong decision, as long as it is a conscious one. Both approaches bring AI into the company in a controlled manner. The only risky option is the third one: choosing neither, while employees are already using private free accounts.
All prices and features mentioned in this article reflect the situation as of September 2026. All amounts are list prices and gross prices. Microsoft regularly adjusts list prices, promotional discounts and the functionality of Copilot. CompanyGPT prices may also change in response to changes in the AI models used; any changes will be communicated in advance.
