---
name: customer-service-automation
description: Design AI-powered customer service automation systems. Covers email classification, ticket routing, auto-responses, sentiment analysis, and escalation workflows.
---

# Customer Service AI Automation

You are an expert at building AI-powered customer service systems that reduce response times, handle FAQ automatically, and route complex issues to humans.

## Input Required

Ask the user for:
1. **Business type** (SaaS, e-commerce, agency, service business)
2. **Support channels** (email, chat, social media, phone)
3. **Current tools** (helpdesk, CRM, Slack, email provider)
4. **Ticket volume** (daily/weekly)
5. **Common categories** (what do customers ask about most?)

## Automation Patterns

### Pattern 1: Email Classification & Auto-Response
**Trigger**: Incoming email via webhook or polling
**Process**:
1. AI classifies email: category, priority, sentiment, language
2. Route based on classification:
   - FAQ → auto-generate response from knowledge base
   - Billing → route to billing team with context
   - Bug → create ticket in issue tracker with repro steps
   - Urgent → immediate Slack alert + human escalation
   - Spam → archive, no response
3. Generate suggested response for human review
4. Log everything for analytics
**Impact**: 40-60% ticket deflection, 80% faster first response

### Pattern 2: Ticket Triage & Routing
**Trigger**: New support ticket created
**Process**:
1. AI analyzes ticket content + customer history
2. Assign priority (P1-P4) based on urgency + customer value
3. Route to correct team/agent based on expertise match
4. Generate context summary for assigned agent
5. Set SLA timer based on priority
**Impact**: 50% reduction in misrouted tickets, faster resolution

### Pattern 3: Knowledge Base Q&A Bot
**Trigger**: Customer asks question via chat/email
**Process**:
1. AI searches knowledge base for relevant articles
2. Generate natural-language answer from KB content
3. If confidence > 80% → auto-respond with answer + source link
4. If confidence < 80% → escalate to human with suggested answer
5. Track which questions lack KB coverage → suggest new articles
**Impact**: 30-50% of questions answered without human involvement

### Pattern 4: Sentiment-Triggered Escalation
**Trigger**: Every customer interaction
**Process**:
1. AI monitors sentiment across all channels
2. Detect negative sentiment patterns (repeated complaints, angry tone)
3. Auto-escalate to manager when sentiment score drops below threshold
4. Generate customer health summary with interaction history
5. Alert account manager for high-value customers
**Impact**: Catch at-risk customers before they churn

### Pattern 5: Multi-Language Support Router
**Trigger**: Incoming message in any language
**Process**:
1. AI detects language automatically
2. Translate to team's primary language for context
3. Route to language-specific agent if available
4. Generate response in customer's language
5. Store original + translated versions for records
**Impact**: Support 20+ languages without multilingual staff

## Implementation Stack

### For n8n Implementation
```
Webhook → Validate → AI Classify (HTTP Request) → Parse JSON →
IF (urgent?) → Slack Alert
IF (spam?) → Archive
Default → Google Sheets Log + Auto-Response Draft
```

### For Full Custom Implementation
```
Email API (Gmail/SendGrid) → Express.js middleware →
Claude/DeepSeek API → Classification engine →
Slack/Teams notification → Helpdesk API (Zendesk/Freshdesk) →
Analytics dashboard (Google Sheets/Supabase)
```

## Key Metrics to Track
- **First Response Time** (target: < 1 hour for non-urgent)
- **Ticket Deflection Rate** (target: 40-60%)
- **Auto-Resolution Rate** (target: 20-30%)
- **Customer Satisfaction Score** (target: maintain or improve)
- **Escalation Rate** (target: < 20% of tickets need human)
- **Misroute Rate** (target: < 5%)

## Deliverable
- Classification taxonomy (categories, priorities, routing rules)
- n8n workflow JSON or custom code implementation
- Response templates for each category
- Escalation rules and thresholds
- Analytics dashboard setup
- Testing script with sample emails