Tutorials
Tutorial: automated customer-support assistant
Classify inbound WhatsApp messages, draft a response, reply, and hand uncertain or sensitive cases to a human.
Automation buildersCustomer support teamsIntermediateAvailable
What this guide covers
- This supported pattern combines Message Received, Classifier, Condition, AI Response, Send Text, Human Handoff, and End Flow.
- Prerequisites: Connected Sender, clear support categories, approved policy, and a staffed handoff process.
- Result: Routine tests receive an appropriate reply; escalation tests pause automation and appear in Chat for takeover.
- Limit: AI must not make unreviewed high-impact decisions; tests run AI nodes for real.
What this is and when to use it
This supported pattern combines Message Received, Classifier, Condition, AI Response, Send Text, Human Handoff, and End Flow.
| Use it when | Expected outcome |
|---|---|
You want first-line support with human escalation. | Routine tests receive an appropriate reply; escalation tests pause automation and appear in Chat for takeover. |
Before you begin
- Connected Sender, clear support categories, approved policy, and a staffed handoff process.
Prerequisites
Connected Sender, clear support categories, approved policy, and a staffed handoff process.
Complete the task
- Create a workflow with Message Received for text.
- Add Classifier categories such as order status, product question, and human help.
- Route human-help or uncertain paths to Human Handoff.
- For allowed categories, configure AI Response with bounded instructions and available message context.
- Send the answer with Send Text.
- End successful and escalated paths explicitly.
- Test each category and activate.
Confirm it worked
- Routine tests receive an appropriate reply; escalation tests pause automation and appear in Chat for takeover.
Common problems
| Symptom | What to check |
|---|---|
AI invents policy | Add stricter instructions and route unsupported questions to a human. |
Handoff alert is missing | Verify the selected alert channel and connected account/recipient. |
Current limitations
AI must not make unreviewed high-impact decisions; tests run AI nodes for real.
Product boundary
AI must not make unreviewed high-impact decisions; tests run AI nodes for real.
Where to go next
Related documentation
AI features, privacy, and human reviewUse Yanola Agent, AI Response, Classifier, Extract Info, Detect Mood, and AI-assisted Function drafts responsibly.ClassifierSort messages into categories automatically. Exact configuration, outputs, execution behavior, errors, limits, and examples for the Classifier workflow node.Human HandoffStop execution and hand off the conversation to a human. Exact configuration, outputs, execution behavior, errors, limits, and examples for the Human Handoff workflow node.