Skip to content

Comparison

ChatGPT for your team, or a custom AI assistant on your own data?

Many Swiss SMEs start with ChatGPT or Microsoft Copilot, and for good reason. Sooner or later the question comes up whether a general tool is enough, or whether you need an assistant that knows your own documents, systems and rules. This page compares both honestly, including the cases where we would advise you not to build anything.

At a glance

CriterionChatGPT or CopilotCustom AI assistant
Getting startedSame day: sign up, invite your team, startAbout 4 weeks for a working pilot
Cost modelLicence per user, roughly USD 20 to 30 per user and month for business plans (approximate list prices)One-off build, pilot from CHF 4,900, then running costs such as hosting and Care from CHF 290 per month
Knowledge of your businessOnly what users paste, upload or connect, depending on the planYour manuals, contracts, price lists and policies, kept up to date
SourcesAnswers can sound certain without showing where they come fromEvery answer shows the documents it is based on
IntegrationsGeneric connectors, mainly for Microsoft or Google environmentsBexio, Abacus, SAP, your CRM or your own systems
Data locationProvider's cloud, often outside Switzerland; business plans do not train on your data by defaultHosting in Switzerland or the EU on request, no training on your data
Who can use itYour licensed employeesEmployees, and customers on your website if you want

What ChatGPT and Copilot do well

General AI tools are very good at everyday knowledge work: drafting an email, rewording a letter, translating a text into French, summarising a long report or collecting ideas for a presentation. They cost little to try, your staff learn them quickly, and if your company already works in Microsoft 365, Copilot sits directly in Outlook, Word and Teams.

If your team mainly needs help with writing and thinking, a general tool is often all you need. We say this in our free consultation as well, because building something custom for these tasks would be money spent without a real benefit.

Where a general tool reaches its limits

  • It does not know your price list, your internal processes or last month's contract changes unless someone feeds them in every time.
  • Answers built on general knowledge can be outdated or simply wrong for your case, and it is not always clear where they come from.
  • It cannot look up an order in Bexio or update a contact in your CRM without additional setup.
  • Every employee writes prompts differently, so the quality of the result depends on who asks.
  • You cannot put it on your website in a controlled way to answer customer questions in four languages.

What a custom assistant adds

A custom assistant uses the same kind of language models underneath. The difference is everything around the model: it searches your own documents and systems, picks the relevant passages and answers with sources, so anyone can check the answer in seconds. It follows the rules you set, respects who is allowed to see what, works in German, French, Italian and English, and hands over to a person when it is unsure.

It can also prepare actions, for example a reply to a customer or an entry for accounting. A person approves what the AI prepares before anything leaves the company. This approval step is part of the design, not an afterthought.

Costs over time

With a general tool, you pay per user, so costs grow with your headcount. With a custom assistant, you pay once for the build and then for running it: hosting, model usage and maintenance. Which option is cheaper depends on how many people use it and how often the same kind of question comes up. We do not claim that custom is always cheaper. For a team of five who mainly write emails, it rarely is.

Data protection and control

Both options can be used in line with the Swiss Data Protection Act if they are set up properly. The business plans of the large providers offer contractual guarantees, but your data is processed in their cloud, often outside Switzerland, and you depend on their terms. With a custom assistant you decide where it runs, which documents it may use and who sees which answers.

Our verdict

If your team needs a flexible helper for writing, translating and summarising, start with ChatGPT or Copilot and set clear rules for what data may be entered. If the same questions about your own documents come up every day, if answers need sources, or if customers should get help on your website, a custom assistant is the better choice. In many companies the right answer is both, each for what it does best.

Questions and answers

Often yes, or comparable ones. We choose the model to suit the task and your hosting requirements, and some open models can run entirely in Switzerland. What makes the assistant useful is the connection to your data, the rules and the testing around the model.