🎓 Teaching e Knowledge Base - Crafter.ai

This is the complete guide to building your chatbot's intelligence through the Teaching Step, which includes guided surveys, custom Q&A, Flow Designer, and knowledge base management.

🎯 Overview of the Teaching Step

The Teaching Step is the heart of your chatbot. Here you define:

• Base Expertise: Fundamental skills through guided surveys
• Custom Expertise: Personalized and specific Q&A
• Flow Designer: Complex conversations and advanced workflows
• Knowledge Base: Documents and external resources
• Improve: Analysis and performance optimization

📋 Base Expertise (Guided Survey)

What is the Guided Survey

The survey is a smart questionnaire that analyzes your business and automatically generates an initial knowledge base for the chatbot.

Survey Categories

1. Business Information

Typical questions:

• Industry
• Company size
• Target customers
• Main sales channels

Example response:

Industry: Clothing e-commerce
Size: SME (10-50 employees) 
Target: Young adults 18-35 years
Channels: Website, Social media, Marketplace

2. Products and Services

Typical questions:

• Main product/service catalog
• Price ranges
• Distinctive features
• Sales processes

Best Practices for Responses:

• ✅ Be specific: "We sell t-shirts, jeans, and casual accessories"
• ✅ Include details: "Price ranges €20-150, focus on sustainability"
• ❌ Avoid generalities: "We sell clothes"

3. Customer Support

Typical questions:

• Current support channels
• Service hours
• Most common issues
• Average response time

4. Business Processes

Typical questions:

• Sales workflow
• Return/refund management
• Company policies
• Standard procedures

How to Effectively Complete the Survey

Preparation

1. Gather information: Existing FAQs, business documents, support statistics
2. Involve the team: Consult sales, marketing, customer service
3. Think about use cases: What questions do customers ask most often?

During Completion

• Complete answers: Use full sentences, not single words
• Concrete examples: Include real examples of products/services
• Natural language: Write as you would speak to a customer
• Operational details: Hours, prices, specific procedures

Example of a Complete Response

❌ Incorrect: "We sell shoes" ✅ Correct: "We sell sports and casual shoes for men and women, from brands like Nike, Adidas, and our private label. Price range €50-200. Specializing in running and urban lifestyle. Available sizes 36-48, with a free home trial service."

🎯 Custom Expertise (Personalized Q&A)

Structure of Q&A

Main Question

The question that the user might ask exactly.

Best Practices:

• Use the customer's natural language
• Include regional/slang variations
• Think about how a real customer would phrase the question

Utterances (Variations)

Different ways to ask the same question.

Complete example:

Main question: "How much does shipping cost?"

Utterances:
- Shipping cost
- Delivery price
- How much do I pay for delivery?
- Shipping rates
- Transport costs
- Shipping fees
- Delivery cost
- How much does the courier cost?

Answer

The response that the chatbot will provide.

Elements of a good answer:

• ✅ Complete information
• ✅ Friendly language
• ✅ Call-to-action when appropriate
• ✅ Updated information

Example of a structured response:

Shipping costs:
• 📦 Standard shipping (3-5 days): €4.90
• 🚀 Express shipping (1-2 days): €9.90
• 🆓 FREE shipping for orders over €50

We deliver throughout Italy with express courier. 
Want to see our products? 👉 [Catalog Link]

Types of Advanced Q&A

Q&A with Variables (use flow designer)

For personalized responses with placeholders.

Question: "When will my order arrive?"
Answer: "Hello {{user_name}}! Your order #{{order_number}} 
          will arrive {{estimated_delivery_date}}. 
          You can track it here: {{tracking_link}}"

Conditional Q&A (use flow designer)

For different responses based on context.

IF current_time IN opening_hours:
  "Our operators are available now! Would you like to be transferred?"
ELSE:
  "Our operators are available from 9 AM to 6 PM. 
   Can I help you or would you prefer to leave a message?"

Q&A with Rich Media

Include images, videos, links.

Question: "How do I assemble product X?"
Answer: "Here is the tutorial video for assembling product X:
          📹 [YouTube Video]
          📄 [PDF Instructions]
          🛠️ Need help? Contact our technical support!"

🌊 Flow Designer (Advanced Conversations)

What is the Flow Designer

The Flow Designer is a visual editor for creating complex conversations with conditional logic, data collection, and integration with external systems.

Basic Elements of the Flow

User Nodes (Input)

• User Says: The user says something
• User Replies: The user responds to a specific question

Bot Nodes (Output)

• Bot Replies: The bot responds with a message
• Bot Asks: The bot asks a question
• Bot Stops: The bot ends the conversation

Logical Nodes

• Condition Check: Checks a condition
• Fork Condition: Forking based on multiple conditions

Action Nodes

• Bot Calls API: Call to external services
• Bot Runs Code: Execute custom code
• Mail Lead: Send email with collected data

Example of a Complete Flow: Appointment Booking

graph TD
    A[User Says: "I want to book"] --> B[Bot Asks: "For which service?"]
    B --> C[User Replies]
    C --> D{Check Service Available}
    D -->|Available| E[Bot Asks: "What day do you prefer?"]
    D -->|Not Available| F[Bot Replies: "Service not available"]
    E --> G[User Replies: Date]
    G --> H[Bot Calls API: Check Calendar]
    H --> I{Slot Available?}
    I -->|Yes| J[Bot Asks: "Name and phone?"]
    I -->|No| K[Bot Replies: "Slot not available"]
    J --> L[User Replies: Data]
    L --> M[Mail Lead: Send Booking]
    M --> N[Bot Replies: "Booking confirmed!"]

Configuring Advanced Nodes

Bot Calls API

For integrations with external systems.

Configuration:

{
  "url": "https://api.example.com/bookings",
  "method": "POST",
  "headers": {
    "Authorization": "Bearer {{api_token}}",
    "Content-Type": "application/json"
  },
  "body": {
    "service": "{{selected_service}}",
    "date": "{{selected_date}}",
    "customer": "{{customer_name}}"
  }
}

Condition Check

For complex conditional logic.

Examples of conditions:

• {{current_time}} > "18:00" → Out of hours
• {{user_location}} == "Milan" → Service area
• {{order_value}} >= 100 → Free shipping
• {{user_type}} == "premium" → Access to advanced features

Slot Management

To collect and maintain information during the conversation.

Example of Slot:

Slot: customer_info
├── name (required)
├── email (required, email validation)
├── phone (optional, Italian format)
└── notes (optional, max 200 characters)

Best Practices for Flow Design

1. Planning

• Map the workflow: Before starting, draw the flow on paper
• Identify critical points: Where can problems arise?
• Fallback paths: Always plan alternatives in case of error

2. User Experience

• Clear messages: Every question should be understandable
• Limited options: No more than 3-4 choices at a time
• Constant feedback: The user should always know where they are

3. Error Management

• Timeout handling: If the user does not respond
• Invalid input: Handling invalid inputs
• API failures: What to do if an external service does not respond

4. Testing

• Happy path: Test the main flow
• Edge cases: Test limit scenarios
• Error paths: Test all failure points

📚 Knowledge Base and Document Upload

Supported Formats

Text Documents

• PDF: Up to 50MB, automatic text extraction
• DOCX/DOC: With formatting preservation
• TXT: Plain text, any size
• RTF: Rich Text Format

Structured Documents

• Excel/CSV: For tabular data (FAQs, catalogs)
• JSON: For structured data and API responses
• XML: For feeds and structured data

Web Content

• HTML: Saved web pages
• URL Scraping: Automatic import from websites
• RSS/Atom: Automatic feeds

Upload Process

1. Document Preparation

Optimization for AI:

• ✅ Clear structure: Use headings, subheadings, lists
• ✅ Natural language: Avoid excessive technical jargon
• ✅ Updated information: Check dates and prices
• ✅ Remove irrelevant content: Footers, headers, advertisements

2. Upload and Processing

1. File selection → 2. Upload → 3. AI Processing → 4. Review → 5. Publishing

Processing times:

• PDF 10 pages: ~2-3 minutes
• DOCX 50 pages: ~5-8 minutes
• Excel 1000 rows: ~3-5 minutes

3. Review and Optimization

After uploading, check:

• Correct extraction: Was the text interpreted correctly?
• Categorization: Did it end up in the right folder?
• Relevance: Is the information useful for the chatbot?

Automatic Content Management

Auto-Categorization

AI automatically categorizes content:

📄 "Returns Guide.pdf" → 📁 Customer Support
📄 "Spring Catalog.xlsx" → 📁 Products
📄 "Shipping FAQ.docx" → 📁 Orders and Shipping

Automatic Updates

• Monitoring URLs: Periodic checks of websites
• Version control: Tracking document changes
• Automatic sync: Updating knowledge base

Duplicate Detection

The system automatically identifies:

• Duplicate content
• Obsolete information
• Conflicts between different sources

🔍 Improve Section (Optimization)

Analysis of Missed Conversations

What are "Misses"

Conversations where the chatbot could not respond adequately.

Misses Dashboard

Displayed metrics:

• Total number of misses
• Temporal trends
• Most frequently unanswered questions
• User satisfaction scores

Resolution Workflow

1. Review conversation → 2. Identify knowledge gaps → 3. Create Q&A/Flow → 4. Train → 5. Test

Conflict Analysis

What is a Conflict

When two or more Q&A could answer the same user question.

Types of Conflicts

• Semantic overlap: Similar questions, different answers
• Ambiguity: A question can have multiple interpretations
• Contradictions: Conflicting information

Conflict Resolution

Strategies:

1. Merge: Combine similar Q&A
2. Clarification: Add clarification questions
3. Priority: Define priority among answers
4. Context: Use conversation context

Analytics and Optimization

Key Metrics

• Success Rate: % of resolved conversations
• User Satisfaction: User ratings
• Response Time: Average response time
• Handover Rate: % of transfers to humans

Optimization Loop

graph LR
    A[Data Analysis] --> B[Identify Problems]
    B --> C[Implement Solutions]
    C --> D[Training]
    D --> E[Deploy]
    E --> F[Monitor Results]
    F --> A

🎯 General Best Practices

Content Strategy

1. Quality over Quantity: Better a few perfect Q&As than many mediocre ones
2. User-Centric: Always from the user's perspective
3. Iterative Improvement: Continuous improvement based on data
4. Consistency: Consistent tone of voice and information

Training Strategy

1. Regular Training: At least weekly for active chatbots
2. Incremental Updates: Small frequent updates vs. large refactors
3. A/B Testing: Test different versions of responses
4. Backup Strategy: Always keep previous versions

Maintenance Routine

• Weekly: Review new misses and conflicts
• Monthly: Complete performance analysis
• Quarterly: Strategic review of the knowledge base
• Yearly: Complete review of content and workflows

✅ Complete Teaching Checklist

Base Expertise

• Survey completed 100%
• All questions with detailed answers
• Updated business information
• Documented business processes

Custom Expertise

• At least 20 main Q&As created
• Utterances added for each Q&A
• Organization in logical folders
• Coverage of main topics

Knowledge Base

• Key documents uploaded
• Processing completed without errors
• Categorization verified
• Updated content

Flow Designer

• Main workflows implemented
• Testing all paths
• Error management configured
• API integrations tested

Optimize & Improve

• Initial analysis of misses completed
• Main conflicts resolved
• Baseline metrics established
• Monitoring process activated

🔗 Next Steps

After completing the Teaching:

1. [Preview and Testing]
2. [Advanced Flows]
3. [Analytics Setup]

---

⏱️ Estimated time for complete Teaching: 4-8 hours for a basic chatbot
🎯 Goal: A chatbot that correctly answers the main questions of your business