Ask a question
On the intranet, in Teams or in the portal – in natural language.
RAG chatbots: Company knowledge as an assistant for staff and support.
A chat assistant answers questions based on approved company sources and shows which documents the answer is based on.
In short: RAG means answers from your own documents, with the source. The assistant searches for relevant content and drafts an answer from it. Sources, permissions and the handling of missing results must be checked.
Starting point
Where time is lost today.
- The same questionsNew staff keep asking the same things.
- Searching the manualSupport leafs through manuals and folders.
- Documented, but can’t be foundThe knowledge exists – it’s just that nobody finds it.
- General chatbotsPublic assistants don’t know your internal rules.
- Unclear sourcesWhether an answer is up to date and backed by evidence remains unclear.
How it works
What the solution does, step by step.
Check permissions
Only sources this person may see are considered.
Search sources
Find relevant passages in the approved documents.
Assemble the context
The passages found as the basis for the answer.
Answer
Drafted from the context, not made up.
Show sources or escalate
Reference to the documents; if uncertain, a note or handover to a person.
Before – after
What changes in day-to-day work.
Examples from the workflow – no percentages, as only a measurement in your business can provide those.
| Situation | Today | With the solution |
|---|---|---|
| Onboarding | questions to the colleague at the next desk | answer from the manual, with the source |
| Support research | leaf through the manual | relevant passage in seconds |
| Update status | unclear | source and version visible |
| Missing knowledge | plausible answer made up | uncertainty shown, escalation |
Benefit
What this use case improves.
Faster onboarding
New staff find answers themselves.
Less searching
Questions instead of folder paths.
Consistent use of knowledge
Everyone gets the answer from the same source.
Traceable sources
Answers should be verifiable against the sources cited.
In action
What working with it looks like.
Sample images from our visual library – the interfaces shown are illustrations, not client data.
Responsibility
What stays with people.
Maintaining the sources, assessing answers when in doubt and any action that follows the answer. The assistant provides evidence; it does not decide.
The solution prepares, sorts and hands over – your staff make decisions and grant approvals, following rules you set.
Systems & integrations
Where the solution connects.
For example – what exactly gets connected depends on interfaces and permissions and is checked in the AI potential assessment.
- Intranet
- Microsoft Teams
- Slack (where planned)
- Website / portal
- SharePoint
- DMS
- Knowledge bases
Building blocks & products
The technology behind it.
Solution building blocks are the technical components of the Swiss AI Platform, combined by process.
RAG / knowledge systems
The knowledge layer underneath.
View building blockAI agents
If the assistant should also prepare actions.
View building blockDocument AI
Prepare sources.
View building block
Control, data & operation
Control stays part of the system.
Depending on the product and interfaces, we run the solution on your premises, in a Swiss cloud or as a combination. Before the pilot, we clarify data flows, access rights and compliance requirements. Background: Private AI and On-premise & hybrid (running on your own servers or a mix of both).
| Topic | Implementation |
|---|---|
| Data | Internal documents and users’ questions – internal to confidential. |
| Approvals | Information only; actions only through agents with their own approval. Escalation to people when uncertain. |
| Roles & permissions | The source systems’ permissions apply to each individual user. |
| Logging | Questions, sources used and answers are traceable; gaps in knowledge are visible. |
| Operating model | Running on your own servers or in a private cloud for the knowledge layer; the chat channel is checked in the project. |
Industries
Where this use case is typical.
- Healthcare
- Manufacturing
- Accounting & fiduciary
- cross-industrythe workflow is not tied to a single industry
Standard approach
How a trial run (pilot) can start.
The pilot does not cover the whole process straight away. We pick a clearly defined part, set approvals and success criteria and test it with real data and the people who work with it. This approach is the same for every use case.
- Choose a processA clearly defined part of the workflow that takes up time today and can be measured.
- Clarify data and systemsWhich data, which interfaces, which permissions – and what is missing.
- Implement a defined workflowWith defined approvals and success criteria, not the whole process at once.
- Measure in real operationWith real data and the people who work with it; then expand or stop.
Related use cases
Related pages.
Frequently asked questions
Frequently asked questions: RAG chatbots
What is the difference from knowledge management?
Knowledge management describes the underlying knowledge layer. The RAG chatbot is a specific chat interface on top of it.
Can the chatbot show sources?
That should be a core part of the solution.
What happens if the knowledge is missing?
The assistant should show uncertainty and not make up plausible-sounding answers.
Which companies is this use case suitable for?
Companies with documented but hard-to-access knowledge and recurring questions from staff or in support – typically in onboarding, policies and product knowledge.
Next step
Check this task for your business.
Describe your current workflow in two sentences. We check the benefit, data situation, integration effort and risks and show whether a product, a solution building block or a custom solution makes sense.