An AI chatbot that answers from your company’s own knowledge.

When the same questions keep returning

Product information, procedures and terms are scattered. Experts repeatedly answer the same questions and new employees do not know where to look. We choose an audience and the knowledge scope the assistant should make available.

The answer starts with its sources

We review documentation and identify current material and its owners. Knowledge retrieval, including RAG, is selected for the documents and questions involved. Connecting a folder does not resolve contradictions between documents.

• Knowledge scope and approved sources
• A chat interface suited to its audience
• Answers with source references
• Roles and document access controls
• Handoff to the appropriate person

What we verify before launch

We test common questions, knowledge gaps, outdated information and attempts to access restricted data. The assistant should indicate insufficient evidence and offer a human handoff. We agree on quality thresholds and escalation cases.

Knowledge needs maintenance

We define how material is added or retired and how changes are checked against answers. Measurement covers quality, usefulness and escalated cases. Customer support and internal procedures may need separate sources and permissions.

Customer chatbot: only information cleared for external use

Sources are approved offer descriptions, public instructions and FAQs. The chatbot has no access to internal margins, team notes or other customers' data. For the chatbot to give the status of an individual case, the customer must sign in. A ticket number alone is not enough.

Illustrative dialogue (an example without client data): customer: “How do I file a complaint?” Chatbot: “The approved guide lists the steps and required information. Here is the source. Would you like to speak to support?” Customer: “Will my complaint be accepted?” Chatbot: “I do not make that decision. I can pass your question to a person.”

We agree the support handoff channel, staff availability and the information needed for a request. When a source is missing, a promise is disputed, a question is out of scope or the customer asks for a person, the chatbot hands the conversation to support staff. When we hand a case to a person, we attach the conversation record in line with GDPR and the agreed data protection rules.

Internal assistant: knowledge matched to the employee's role

Sources are current procedures and documents with named owners. Access follows the employee's account and role. A document unavailable to the user must not appear in an answer or its references. We separate public and internal knowledge and check attempts to exceed permissions.

Illustrative dialogue (an example without client data): employee: “How do I hand off a complaint?” Assistant: “The procedure available to your role says to send the request to the process owner. Here is the relevant passage and document.” Employee: “Two documents give different steps. Which should I use?” Assistant: “The sources conflict. I will pass the question to the procedure owner.”

When a current source or permission is missing, the assistant does not fill gaps by guessing. It routes the question to the knowledge owner or authorised team. Before launch we agree answer acceptance, source updates and ownership of open requests.

What we need from you

For the first call bring recurring questions and where answers currently live. Appoint someone who can identify current documents and a decision maker to approve access and the pilot. You do not need to rewrite the whole knowledge base yourself. Your time commitment depends on source count and contradictions; we agree on it after reviewing the material, alongside ongoing tool costs.

How we define the scope

For a quote, we need sample documents, questions, user groups and the intended interface. Knowledge preparation, chat implementation, integrations and usage or maintenance costs are separated.