Day 11: Customer support AI router – ticket classification, prioritization, auto-response

🔗 Knowledge graph – Day 11 builds on everything

Day 1

Prompt engineering – classification prompts, sentiment analysis, response generation.

Day 2

Zapier logic – simple ticket → Slack alerts (basic version).

Day 3

Make multi-step – routers, aggregators, error handling for complex flows.

Day 4

OpenAI API – core engine for classification, sentiment, response generation.

Day 5

Lead qualifier – similar prompt structure adapted for tickets.

Day 6

Support use cases – ticket triage was a key example.

Day 7

3 builds – you built a support router as one option.

Day 8

Lead qualifier – reusing BANT-style prompts for ticket urgency.

Day 9

Sales assistant – email generation similar to ticket responses.

Day 10

Content engine – knowledge base articles for auto-responses.

Shared link: Day 10 generated knowledge base content. Day 11 uses that content for auto-responses. Day 8's BANT prompts are adapted for ticket urgency. Day 9's email prompts are reused for responses.

🎯 What is an AI support router?

📌 Your 24/7 tier-1 support agent

An AI support router automatically:

Classifies incoming tickets (billing, technical, account, general)
Prioritizes by urgency (critical, high, medium, low) using sentiment and keywords
Answers common questions instantly from your knowledge base
Routes complex issues to the right human agent with context

Analogy: Think of it as an emergency room triage nurse. When patients (tickets) arrive, the nurse quickly assesses symptoms (reads email), decides who needs immediate care (priority), sends minor cases to a generalist (auto-response), and directs critical cases to specialists (human agents).

🏗️ System architecture – complete support pipeline

1. Trigger

New email to support@ or contact form

2. Classify

Category + priority + sentiment

3. Knowledge base

Search for relevant answers

4. Router

Auto-answer vs human

5. Auto-response

Send answer + ticket created

6. Human route

Zendesk/HubSpot + Slack alert

⚙️ Build the AI support router (Make.com)

Trigger: New support email

Sources:

Gmail (support@ label)
Contact form (webhook)
Zendesk / HubSpot ticket creation

Practice example: Create a Gmail filter that adds label "support-test" and use that as trigger.

AI Classification (OpenAI call #1)

Prompt (Day 1 style):

{
  "messages": [
    {"role": "system", "content": "You are a support ticket classifier. Analyze the email and return JSON with: category (billing/technical/account/general), priority (critical/high/medium/low), sentiment (angry/frustrated/neutral/positive), and a one-line summary."},
    {"role": "user", "content": "Subject: {{subject}}\n\nBody: {{body}}"}
  ]
}
                        

Practice example: Use this email: "I was charged twice for my subscription! Fix this immediately." Expected: category=billing, priority=critical, sentiment=angry.

Knowledge base search (using Day 10 content)

Option A: Use OpenAI to find best answer from KB:

{
  "messages": [
    {"role": "system", "content": "You have access to this knowledge base. Find the most relevant article to answer the customer's question. Return the article title and answer."},
    {"role": "user", "content": "KB: {{kb_articles}}\n\nQuestion: {{summary}}"}
  ]
}
                        

Option B: Use vector search (advanced) – but for practice, maintain a Google Sheet of FAQ answers.

Router: Auto-answer or human?

Rules:

Auto-answer if: priority low/medium AND category general/faq AND sentiment not angry
Human if: priority critical/high OR category billing/technical complex OR sentiment angry

Practice: Add a router with these conditions using Make's router module.

Auto-response path

Generate friendly reply (OpenAI call #2):

{
  "messages": [
    {"role": "system", "content": "Write a helpful reply to this customer. Use the knowledge base answer. Thank them, answer clearly, and offer next steps. Sign as 'Support Team'."},
    {"role": "user", "content": "Original question: {{summary}}\n\nAnswer from KB: {{kb_answer}}"}
  ]
}
                        

Then:

Send email via Gmail
Create ticket in Zendesk/HubSpot with status "Solved"
Log to Google Sheets

Human route path

For tickets needing human attention:

Create ticket in Zendesk/HubSpot with priority, category, summary
Send Slack alert to appropriate channel (#support-urgent, #support-billing)
Include all context: email, classification, suggested actions

Slack message:
🚨 *New urgent ticket*
From: {{from}}
Category: {{category}} | Priority: {{priority}} | Sentiment: {{sentiment}}
Summary: {{summary}}
[View in Zendesk](link)
                        

Feedback loop (optional)

After ticket is solved by human, add a step to update the knowledge base if it was a new question.

Advanced: Send to Day 10's content engine with prompt: "Create a new FAQ article from this ticket resolution."

📚 Support prompt library – reuse Day 1 skills

Classification

"Classify this support email. Return JSON: category (billing/technical/account), priority (1-5), sentiment (negative/neutral/positive)."

Urgency detection

"Does this email indicate the customer is blocked or cannot work? Return true/false."

Angry customer

"Draft a calm, empathetic response to an angry customer. Acknowledge frustration, apologize, and explain next steps."

KB answer

"Using this knowledge base article, answer the customer's question in 2-3 sentences. Be friendly."

Ticket summary

"Summarize this email thread in 2 sentences for a support agent."

Language detection

"Detect language of this email (use ISO code). If not English, route to human."

Shared link: These prompts adapt Day 1 frameworks (role, task, context, format) for support.

📖 Knowledge base integration – using Day 10 content

Your Day 10 content engine can generate FAQ articles. Here's a simple KB structure:

Google Sheet "Knowledge Base":
- Category (billing, technical, account)
- Keywords (comma separated)
- Question
- Answer
- Article ID
            

In Make, use "Search rows" to find matching KB articles based on category + keyword match.

Practice: Create 5 KB articles using Day 10 prompts. Then test with sample tickets.

7 hands-on practice exercises

📝 Exercise 1: Classification only

Task: Build a Make scenario that takes an email, calls OpenAI to classify (category, priority, sentiment), and logs to Google Sheets.

Test with 5 emails: billing complaint, technical issue, account question, angry customer, simple FAQ.

📚 Exercise 2: KB search

Task: Create a KB sheet with 5 articles. Build a search that finds the best match based on category + keywords.

🔄 Exercise 3: Router logic

Task: Add router conditions (auto vs human) based on priority and sentiment.

✉️ Exercise 4: Auto-response

Task: Generate and send auto-response for auto-route tickets. Use Gmail draft mode.

🔔 Exercise 5: Slack alerts

Task: For human-routed tickets, send formatted Slack message to a test channel.

📊 Exercise 6: Reporting

Task: Create a weekly report (scheduled) showing #tickets by category, avg response time, auto-solve rate.

🔄 Exercise 7: Connect to Day 10

Task: When a human solves a new issue, trigger Day 10 to generate a KB article draft for approval.

📄 Client documentation – AI support router

🎫 AI Support Router – System Overview

What it does: Automatically classifies, prioritizes, and responds to support tickets 24/7.

Input: Emails to [email protected] (or contact form).

Classification: Category (billing/technical/account/general), Priority (critical-low), Sentiment.

Auto-response rate: Typically 40-60% of tickets answered instantly using your knowledge base.

Human routing: Critical/high tickets + angry customers sent to Slack + Zendesk with full context.

Knowledge base: Google Sheet you can update anytime. New articles can be added via Day 10.

Reporting: Weekly email with ticket stats and auto-solve rate.

Pricing: $1,800 setup + $300/mo (includes AI costs for up to 2,000 tickets/month)

⚠️ Pitfalls & how to avoid them

False positives on priority: Test classification prompts with 20+ real emails. Adjust prompt examples.
Auto-answering the wrong thing: Always include "If unsure, route to human" in router logic.
Ignoring language: Add language detection – non-English should go to human (unless you have multi-lingual KB).
No feedback loop: When human solves a new issue, capture it and add to KB.
Cost surprise: Set max_tokens low for classification (50-100). Use gpt-3.5-turbo.

📚 Resources

Day 11: You built a complete AI support router

✔ Ticket classification (category, priority, sentiment)
✔ Knowledge base search using Day 10 content
✔ Intelligent router (auto-answer vs human)
✔ Auto-response generation
✔ Slack alerts for urgent tickets
✔ 7 detailed practice exercises
✔ Connected to Days 1-10