AI Automation Platforms Guide

User: "When someone fills out my Typeform, add them to
Mailchimp and send a Slack notification to the sales channel"

Zapier Central:
✓ Identified trigger: Typeform submission
✓ Identified actions: Mailchimp add subscriber, Slack message
✓ Created Zap with proper field mapping
✓ Ready to test

# Traditional mapping
contact_name = typeform_response.fields[0].answer

# AI-assisted mapping
# Zapier AI recognizes "Full Name" field regardless of position
contact_name = ai_map(typeform_response, "contact name")

Make AI Module:
├── Text Generation (OpenAI/Anthropic)
├── Image Analysis
├── Document Processing
├── Sentiment Analysis
└── Entity Extraction

// n8n AI Agent configuration
{
  "agent": "conversational",
  "model": "gpt-4",
  "tools": [
    "web_search",
    "calculator",
    "custom_api"
  ],
  "memory": "buffer_window",
  "max_iterations": 10
}

Trigger: New lead in CRM

AI Analysis Step:
├── Analyze company (size, industry, location)
├── Score lead quality (1-100)
├── Identify product interest
├── Detect urgency signals
└── Generate personalized talking points

Routing Logic:
├── Score > 80: Enterprise sales team
├── Score 50-80: Mid-market team
├── Score < 50: Marketing nurture
└── High urgency: Immediate alert

1. Trigger: HubSpot - New Contact
2. Action: OpenAI - Analyze Contact
   Prompt: "Analyze this lead and return JSON with
            score (1-100), team (enterprise/midmarket/nurture),
            urgency (high/medium/low), and talking_points (array)"
3. Path A (score > 80): Slack - Enterprise Channel
4. Path B (score 50-80): Slack - Midmarket Channel
5. Path C (score < 50): Mailchimp - Add to Nurture List

Trigger: New blog post published

Processing Pipeline:
├── Extract key points (AI)
├── Generate social media variations (AI)
│   ├── Twitter thread
│   ├── LinkedIn post
│   └── Instagram caption
├── Create email newsletter snippet (AI)
├── Generate SEO meta description (AI)
└── Schedule distribution

Output:
├── Buffer: Schedule social posts
├── Mailchimp: Queue newsletter
├── WordPress: Update meta tags
└── Slack: Notify marketing team

Trigger: New support ticket

AI Classification:
├── Category detection
├── Sentiment analysis
├── Urgency scoring
├── Language detection
├── Related KB article matching
└── Suggested response generation

Routing:
├── Technical + High Urgency: Engineering on-call
├── Billing: Finance team
├── Feature Request: Product team
├── Negative Sentiment: Customer Success
└── Self-serviceable: Auto-respond with KB link

// AI Classification Node
const classification = await this.helpers.requestWithAuthentication.call(
  this, 'openaiApi', {
    method: 'POST',
    url: 'https://api.openai.com/v1/chat/completions',
    body: {
      model: 'gpt-4',
      messages: [{
        role: 'system',
        content: `Classify support tickets. Return JSON:
          {
            "category": "technical|billing|feature|general",
            "sentiment": "positive|neutral|negative",
            "urgency": 1-10,
            "suggested_response": "string",
            "kb_articles": ["article_id_1", "article_id_2"]
          }`
      }, {
        role: 'user',
        content: ticketContent
      }]
    }
  }
);

Trigger: Recording uploaded to cloud storage

Processing:
├── Transcription (Whisper/AssemblyAI)
├── AI Analysis:
│   ├── Meeting summary (3-5 sentences)
│   ├── Key decisions made
│   ├── Action items with owners
│   ├── Follow-up questions
│   └── Sentiment by participant
└── Distribution

Actions:
├── Notion: Create meeting notes page
├── Asana: Create tasks for action items
├── Slack: Send summary to channel
├── Calendar: Schedule follow-ups
└── CRM: Update relevant opportunities

Triggers (Multiple):
├── RSS: Competitor blog posts
├── Twitter: Competitor mentions
├── Google Alerts: News mentions
├── LinkedIn: Company updates
└── G2/Capterra: New reviews

AI Processing:
├── Relevance scoring (0-100)
├── Category classification
├── Sentiment analysis
├── Key insight extraction
├── Competitive impact assessment
└── Recommended response

Actions:
├── Airtable: Add to intelligence database
├── Slack #competitive: High-relevance alerts
├── Weekly: AI-generated summary report
└── CRM: Flag affected opportunities

My Apps → Add Connection
├── CRM (HubSpot, Salesforce)
├── Email (Gmail, Outlook)
├── Communication (Slack, Teams)
├── Storage (Google Drive, Dropbox)
└── Marketing (Mailchimp, ActiveCampaign)

Create Zap →
1. Choose Trigger App (e.g., Typeform)
2. Choose Trigger Event (New Entry)
3. Connect Account
4. Test Trigger
5. Add Action (e.g., Slack)
6. Configure Message
7. Test Action
8. Turn On

Scenarios → Create New Scenario
1. Add Trigger Module (watch)
2. Add Processing Modules
3. Add Action Modules
4. Configure Data Mapping
5. Set Scheduling
6. Activate

Add Module → AI → OpenAI
├── Operation: Create Chat Completion
├── Model: GPT-4
├── System Prompt: Define AI behavior
├── User Message: Map from previous modules
└── Parse Response: Extract structured data

# Requirements: Docker or Node.js 18+
# Minimum: 2 CPU, 4GB RAM, 20GB storage

# Docker installation (recommended)
docker volume create n8n_data

docker run -d \
  --name n8n \
  -p 5678:5678 \
  -v n8n_data:/home/node/.n8n \
  -e N8N_BASIC_AUTH_ACTIVE=true \
  -e N8N_BASIC_AUTH_USER=admin \
  -e N8N_BASIC_AUTH_PASSWORD=secure_password \
  n8nio/n8n

# docker-compose.yml
version: '3'
services:
  n8n:
    image: n8nio/n8n
    ports:
      - "5678:5678"
    environment:
      - N8N_HOST=n8n.yourdomain.com
      - N8N_PORT=5678
      - N8N_PROTOCOL=https
      - NODE_ENV=production
      - WEBHOOK_URL=https://n8n.yourdomain.com/
      - GENERIC_TIMEZONE=America/New_York
      - DB_TYPE=postgresdb
      - DB_POSTGRESDB_HOST=postgres
      - DB_POSTGRESDB_DATABASE=n8n
      - DB_POSTGRESDB_USER=n8n
      - DB_POSTGRESDB_PASSWORD=secure_password
    volumes:
      - n8n_data:/home/node/.n8n
    depends_on:
      - postgres

  postgres:
    image: postgres:15
    environment:
      - POSTGRES_USER=n8n
      - POSTGRES_PASSWORD=secure_password
      - POSTGRES_DB=n8n
    volumes:
      - postgres_data:/var/lib/postgresql/data

volumes:
  n8n_data:
  postgres_data:

Settings → Credentials → Create New
├── OpenAI API
├── Anthropic API
└── Custom LLM (for self-hosted models)

✓ Good: One workflow per business process
✗ Bad: Mega-workflow handling multiple unrelated processes

Example:
├── Lead Qualification Workflow
├── Lead Routing Workflow
├── Lead Nurture Workflow
└── Lead Conversion Workflow

Every workflow should have:
├── Try/Catch blocks for API calls
├── Retry logic with exponential backoff
├── Failure notifications (Slack, email)
├── Logging for debugging
└── Fallback paths for critical processes

# Structure prompts for consistent output
system_prompt = """You are a data classification assistant.
Analyze the input and return ONLY valid JSON in this format:
{
  "category": "one of: sales, support, billing, other",
  "priority": "one of: high, medium, low",
  "summary": "one sentence summary"
}

Do not include any text outside the JSON object."""

AI Response → Validation Step:
├── Check JSON validity
├── Verify required fields
├── Validate enum values
├── Confirm data types
└── Fallback if invalid

AI Cost Controls:
├── Use cheaper models for simple tasks (GPT-3.5)
├── Reserve expensive models for complex analysis
├── Implement caching for repeated queries
├── Set budget alerts
└── Monitor usage daily

Security Checklist:
├── Use OAuth where available
├── Rotate API keys quarterly
├── Limit credentials to necessary permissions
├── Audit access regularly
└── Use environment variables (n8n)

Sensitive Data Handling:
├── Don't log PII in automation
├── Encrypt data at rest
├── Use secure webhooks (HTTPS only)
├── Implement IP allowlists
└── Review AI prompts for data leakage

Task Automation Savings = (Manual Time) × (Hourly Cost) × (Frequency)

Example: Lead routing
- Manual time: 5 minutes per lead
- Hourly cost: $30
- Monthly volume: 500 leads
- Monthly savings: (5/60) × $30 × 500 = $1,250

Error Cost Savings = (Error Rate Reduction) × (Cost per Error) × (Volume)

Example: Data entry
- Previous error rate: 5%
- New error rate: 0.5%
- Cost per error: $50 (correction time)
- Monthly volume: 1,000 entries
- Monthly savings: (4.5%) × $50 × 1,000 = $2,250

Diagnosis Steps:
1. Check API credentials are valid
2. Verify OAuth tokens haven't expired
3. Test API endpoint directly
4. Check rate limits
5. Review firewall/IP restrictions

Resolution:
1. Verify source data structure
2. Check for null/undefined values
3. Validate data types
4. Test with sample data
5. Add data transformation step

When AI returns unexpected results:
1. Review prompt clarity
2. Add explicit format instructions
3. Include few-shot examples
4. Validate JSON structure
5. Implement retry logic