RIFT - billing leak detection RIKO - the AI front desk RISE - social media (coming soon) Is it for me? Security Philanthropy Blog Login Cart Free leak report
← Back to Help centre RIKO

How RIKO escalates a conversation

What a visitor sees when RIKO hands a chat to your clinic, the ten escalation triggers, per-clinic keywords, and the safety gate that cannot be turned off.

This article explains what happens when RIKO decides a visitor needs your clinic rather than the bot: what the visitor sees, what you receive, the ten kinds of trigger, the per-clinic settings, and the safety response for crisis messages.

What escalation means in RIKO

RIKO never brings a person into the chat. When a message needs a human, the bot stops answering from the knowledge base, asks the visitor for their name and email, and passes the message to your clinic so you can follow up by email or phone. In the History tab the conversation is marked Escalated. The visitor is told your team will be in touch at the email they gave; nobody joins the chat window.

What the visitor sees

  1. The visitor sends a message that trips one of the triggers below.
  2. The bot replies: "I can connect you with our team - just share your name and email and someone will follow up with you."
  3. The visitor types their name and email (for example "Sam Jones, sam@example.com"). If they reply without an email address, the bot asks again: "No worries - just drop your name and email and I'll pass it straight to the team."
  4. The bot confirms: "Thanks Sam Jones! A team member will be in touch at sam@example.com shortly."

The bot does not go back to normal answering in that session until the contact details are given.

What your clinic receives

Depending on your plan and settings, an escalation reaches you in up to three ways:

Whatever alerts are set up, the full conversation is always in the History tab, where you can filter by the Escalated chip.

The ten escalation triggers

RIKO checks every visitor message against these triggers, in this order. The first match wins.

  1. Asking for a person - "speak to a human", "real person", "not a bot", "get me someone". Critical.
  2. Safety or medical emergency - self-harm, suicide, "medical emergency", "life-threatening". Critical, and handled by the safety gate described below.
  3. Legal matters - lawsuits, lawyers, negligence, compensation, ombudsman, discrimination, harassment. High.
  4. Refund or payment disputes - demanding a refund, chargebacks, "give me my money back", fraudulent charges. High.
  5. Formal complaints - "I want to complain", "lodge a complaint", "completely unacceptable", "extremely dissatisfied". High.
  6. Strong negative language or abuse - fury, insults, swearing, "this is a joke", "worst service". High. Can be switched off per clinic.
  7. Your keyword triggers - words or phrases your clinic has asked to escalate on. High.
  8. Your sensitive topics - subjects your clinic wants kept away from the bot. High.
  9. Conversation too long or going nowhere - the chat reaches 8 turns, or the bot has failed to answer 2 times. Medium.
  10. Low confidence - the knowledge base has only a weak match for the question. Low: the bot still answers, but is told to say it is not sure and to give your contact email rather than guess.

Triggers 1 to 9 start the name-and-email flow above. Trigger 10 does not; it only makes the answer more cautious.

The safety gate (cannot be turned off)

If a message contains self-harm, suicide or medical-emergency language, RIKO treats it as a crisis before doing anything else - before booking, before contact collection, before looking anything up. The visitor immediately receives a message saying it sounds like they might be going through something really difficult, that the bot is an AI and cannot provide crisis help itself, but that people who can are available right now:

The message ends by telling the visitor the team has been notified so someone can follow up, and that they are not alone. At the same time your clinic is alerted straight away (email alert, Slack and webhook if set up), without waiting for the visitor to leave their details. The bot then asks for the visitor's name and email as in any escalation. The conversation is marked Escalated with reason "safety concern".

This response is built into every RIKO widget and there is no setting to disable it. A clinic website is exactly where someone in distress may reach out, and the bot must never answer such a message as if it were an ordinary enquiry or pull it into a booking conversation.

Per-clinic escalation settings

Each clinic has its own escalation profile with these settings:

These settings do not have controls in the dashboard. To add keywords or sensitive topics, turn negative-sentiment escalation off, change the thresholds or set the notification email, send the details to support via the Support button in the dashboard or email support@rapidintelligence.com.au and we will apply them to your account.

When the bot simply does not know

Not knowing an answer is not an escalation. If your knowledge base has nothing relevant, the bot replies "I'm sorry, I don't have enough information to answer that. You can contact [your business] directly at [your Contact Email]." These chats are marked No answer in the History tab. Filling the gap in the Knowledge tab is the fix. See Building the RIKO knowledge base.

Escalation and plans

Escalation triggers 1 to 6, 9 and 10 run on every plan. Slack alerts and the webhook are Pro features. Custom keywords and sensitive topics are applied by support on request for any plan.

Still stuck?

Ask RIKO Help, the chat bubble in the corner of this page - it answers from these articles. If it cannot help, use the support form in your dashboard or email support@rapidintelligence.com.au. Please do not include patient names or details.