Standard support bots treat an angry user the same as a happy one. This is a recipe for churn. Emotional Intelligence Workflows analyze the sentiment of a message before deciding how to handle it.

The Strategy

We insert a lightweight Sentiment Analysis Step at the very beginning of the pipeline.

Happy/Neutral: Route to standard AI Agent.
Frustrated: Switch AI to "Empathetic Mode" (more apologetic, concise).
Angry/Crisis: Bypass AI entirely and alert a Human Manager immediately.

Implementation

1. The Sentiment Step

This runs on every incoming message. We use a fast, cheap model (like gpt-4o-mini or even a local BERT model) for this check.

Codetext

const sentiment = await cencori.ai.classify(message.content, {
  labels: ["positive", "neutral", "frustrated", "angry"],
  description: "Detect the emotional tone of the user."
});

2. The Routing Logic (Switch Statement)

Your workflow branches based on the result.

Codetext

switch (sentiment) {
  case "angry":
    // ALARM: Immediate Escalation
    await tools.slack.alert(`🚨 ANGRY USER: ${user.email}`);
    await cencori.workflows.transferToHuman();
    break;
 
  case "frustrated":
    // Mode Switch: High Empathy
    await agent.updateSystemPrompt("You are deeply apologetic. Focus on solving the issue quickly.");
    await agent.reply(message);
    break;
 
  default:
    // Standard AI
    await agent.reply(message);
}

3. Long-Term Tracking

You can also track this over time in Adaptive Memory.

If a user is consistently "Frustrated" over 3 days, trigger a churn.risk event.
If a user shifts from "Neutral" to "Positive", trigger an upsell.opportunity event.

Code Example: The "Cool Down" Loop

A sophisticated pattern is the "Cool Down" loop.

User is Angry.
AI acknowledges deeply: "I completely understand why you are upset. That sounds incredibly frustrating."
AI asks specifically: "To fix this right now, I need to know X."
Re-evaluate Sentiment on next reply.
If User cools down -> Resume standard help.
If User stays Angry -> Escalate.