Customer support that answers, even when nobody is in the office.
3
Channels on one knowledge base: web chat, WhatsApp, email
1
Handover to a human, with the full history
RAG
Answers from your wiki, documents and FAQs
DE / EN / PT
Languages we run ourselves; more as needed
The system makes the difference, not the prompt.
A support assistant holds up in daily use when it is embedded in knowledge, workflows and escalation. Otherwise you get a prettier FAQ page.
Natural language instead of a decision tree or button menu.
Answers from your documents, FAQs, manuals and CRM context.
Actions such as appointment requests, lead capture or follow-ups.
What has to be in place operationally.
For the assistant to hold up in daily use, we define with you what it closes, what it hands over and what it may access.
- 01
Which enquiries the assistant closes and when people take over.
- 02
Which systems may be read or written.
- 03
Which language, tone and risk rules apply per channel.
Four building blocks that turn a demo into an operation.
Channels, knowledge base, actions and limits together form a service model that relieves your team instead of keeping it busy.
Channels
One assistant for web chat, WhatsApp and email; phone on request.
- One knowledge base and the same rules on every channel
- Continuity when a customer switches channel
- In your customers' languages
Knowledge base
Answers from your wiki, documents and FAQs, not from the open web.
- Every answer with a source
- Gaps are reported
- Maintained by your team
Actions
The assistant triggers the next step in your systems.
- Create the enquiry in the CRM
- Request an appointment
- Trigger a follow-up
Limits
Rules for sensitive questions, escalation and traceability.
- Recognise risk patterns
- Pass on the history
- Fallback path to a human
The technical layer stays clear and controllable.
The assistant does not sit loosely on a model. It gets defined sources, tools and handover points.
Knowledge base
Answers from your wiki or document layer (RAG) instead of open web knowledge.
System connections
CRM, calendar, practice or industry software are connected deliberately.
Data handling
Self-hosted or on controlled infrastructure in Switzerland or the EU, with clear data paths and permissions.
Conversation first, if it is still unclear which enquiries cost the most time.
Not every problem is a support problem. Sometimes visibility is missing first, sometimes a CRM, sometimes the knowledge the assistant is supposed to answer from.
In the conversation we separate whether enquiries are missing, sit unanswered, or take too long.
If the question is visibility, the audit first measures how Google and AI answers read your website.
That way the assistant lands where it takes work off your plate, not beside the business.
If the problem is clearly operational, we go straight into implementation. If it is fuzzy, the conversation settles the order.
The founder of RakenAI
For Goekhan Dogan, brand identity and personal communication belong together. His experience in lead management and international medical tourism informs assistant planning: what information does a prospective customer need, and when should a person take over the conversation?
Meet GoekhanWhat owners ask before an assistant goes live.
What happens when the assistant does not know the answer?
It hands over to a human. You write the escalation rules, and they live in the configuration, not in the model's judgement. That way the cases that need a person reliably reach one.
Can one assistant cover web chat, WhatsApp and email together?
Yes, on one knowledge base. Every channel answers from the same knowledge base under the same rules, and the enquiry lands in the CRM with its history. We connect phone on request; it is not the default channel.
Does the assistant decide anything on its own?
It answers, qualifies and creates enquiries within the limits you set. Medical, legal and commercial judgement calls go to your team with the full history instead of being decided by the system.