Automate Customer Support with Mintlify Documentation & Zendesk AI Agent

Description

This n8n workflow automates support ticket handling with AI-driven classification, response generation, and safety checks. Responses are based solely on your Mintlify documentation, ensuring accuracy, consistency, and reduced manual effort in customer support.

✅ Trigger: New Ticket Received
The workflow is triggered whenever a new support ticket is created.

🔍 Check for Assignee
If the ticket is already assigned to a human agent, the bot does nothing and exits.
If the ticket is unassigned, the bot continues processing.

🔢 Bot Response Count Check
The workflow checks how many times the bot has already responded to this ticket.
If the bot has replied more than 3 times, it stops and waits for a human to take over.
This prevents endless loops and flags potentially complex cases for review.

🧠 AI-Based Ticket Categorization
An AI model analyzes the ticket content and classifies it into one of the following categories:
🧾 Billing → Sends a predefined billing-related message.
📢 Advertising → Automatically deletes the ticket.
🚨 Fraud → Sends a predefined fraud-related message.
❓ Other → Proceeds to generate a dynamic response.

🤖 Mintlify Integration
For tickets categorized as "Other", the customer’s question is sent to the Mintlify API, which returns a documentation-based answer.

✍️ AI Response Formatter
The raw response from Mintlify is passed to an AI model that:
Summarizes and rewrites the answer in a clear, friendly tone
Limits the response to 120 words
Adds conversational elements like “Hi,” “Thanks,” and a proper closing

🛡️ AI Confidence Filter
A second AI model reviews the formatted response to ensure it sounds confident and accurate.
It looks for uncertainty phrases like:
“I’m not sure”
“I don’t have enough information”
“It depends…”
If the response is flagged as uncertain, the workflow stops and waits for a human agent to respond.

📤 Send Response & Update Ticket
If the response passes the confidence check:
The reply is sent to the customer
The ticket status is updated to “Pending”