An AI assistant that sells: what it must know, and when to hand over

Start with your knowledge, not the model

A chatbot is only as good as what it knows. The language model brings fluent language; your prices, policies, catalogue and FAQs bring the facts. We connect the two with retrieval-augmented generation (RAG): before it answers, the assistant looks up the most relevant passages in your content and builds the reply on them. Answers stay specific to your business, and they are easy to keep current — change the document and the answer changes.

Design the path to a sale

• Answer the question first, then offer one clear next step: a quote, a booking or a product page.
• Ask only for what you need to follow up — usually a name and a contact.
• Send the whole conversation to your team or CRM, so nobody asks the customer the same thing twice.
• Speak the customer's language: one knowledge base can serve several languages.

Know when to stop

A good assistant says "I don't know" instead of guessing. Set clear limits: no legal or medical advice, no prices outside the published list, and a fast hand-over to a person for complaints, refunds and anything urgent. Test with real customer questions before launch, and read the transcripts every week during the first month.

Be open about it

Under the EU AI Act, from 2 August 2026 people must be told when they are talking to an AI, unless it is obvious. One short line in the chat window does it. Link to your privacy notice, and collect no more personal data than the conversation needs.

A character, not a widget

An assistant with its own look and voice is remembered — and remembered assistants get used. We give yours an original art character in the spirit of Jangera, so it sells your brand while it answers.

Sources and further reading

• EU AI Act — Article 50

Terms in this article