“Where is the template for maintenance contracts?” “What warranty did we promise that customer last year?” “What was the procedure for a complaint again?” In every business, the answers are buried in manuals, emails and folder structures – and searching for them costs time. An AI assistant that knows your own documents finds them in seconds.
Why ChatGPT alone is not enough
General AI models know a lot about the world but nothing about your business. Ask them about internal details and they guess – convincingly. The solution is called RAG (Retrieval-Augmented Generation): answers supplemented by looking things up.
How RAG works – simply explained
- Import documents: manuals, PDFs, Word documents, wiki pages or tickets are split into short sections.
- Index the content: each section is stored so that the AI can later search by meaning – not just by exact words.
- Ask a question: an employee asks in plain language.
- Look it up: the system finds the most relevant sections in your documents.
- Answer with sources: the AI writes an answer based only on those sections and names the documents it came from.
The benefit: answers can be verified, and new documents are available immediately – without retraining the model.
Typical uses in an SME
- Internal handbook: retrieve processes, templates, policies and IT guides by asking a question.
- Support and service: technicians find solutions from previous cases and vendor documentation.
- Quotes and contracts: quickly find earlier offers and contract clauses.
- Onboarding: new employees ask the assistant instead of interrupting colleagues.
What to watch out for
- Permissions: the assistant may only show content the person asking would be allowed to see anyway. Payroll lists don’t belong in an assistant for everyone.
- Data quality: outdated or contradictory documents lead to wrong answers. Cleaning up beforehand pays off.
- Source references: only answers that point to the document are trustworthy in daily use.
- Storage location: your documents are confidential. A RAG system can run entirely on your own hardware or in a Swiss data centre – see local AI for SMEs.
Start small
Begin with one clearly defined area, such as the internal handbook or support guides. After a few weeks you will see which questions are asked and where documents are missing. The assistant can then be expanded step by step and connected, for example via n8n, to email, the ticket system or ERP.
Conclusion
An AI assistant with your own documents makes scattered company knowledge accessible – traceable and without data having to leave the company. We plan and run such assistants on your servers or in Switzerland. More under automation & AI.
Frequently asked questions
What does RAG mean?
Retrieval-Augmented Generation: the AI first searches your documents for relevant passages and then writes its answer based only on them.
Does the AI have to be trained on our data?
No. With RAG the model stays unchanged. Your documents are only searched, and new documents are available immediately.
Can the assistant still give wrong answers?
Yes, but the risk is much lower than with general chatbots. With source references, every answer can be checked quickly.
Which file formats can be included?
Typically PDF, Word, text, web pages, wiki content and emails. Scanned documents first need text recognition.
