AI Employee

Important: AI Employees are autonomous agents that can handle conversations, access tools, retrieve knowledge, and perform actions on behalf of your organization. They learn from your data and adapt to your business needs.

​

🔢 Table of Contents

1. What is an AI Employee?
2. Creating Your First AI Employee
3. Configuring AI Employee Settings
4. Adding Knowledge with Datastores
5. Connecting Tools
6. Deploying Your AI Employee
7. Monitoring and Optimization
8. Best Practices

​

1. What is an AI Employee?

​

Overview

• Have natural conversations with users
• Answer questions using your custom knowledge base
• Perform actions through connected tools
• Qualify leads and collect information
• Transfer to human agents when needed
• Work 24/7 across multiple channels

• Natural Language Understanding: Comprehends user intent and context
• Knowledge Retrieval: Searches datastores for accurate information
• Tool Usage: Calls APIs, submits forms, generates leads
• Multi-Channel: Works on web chat, WhatsApp, Messenger, Telegram, and more
• Handoff to Humans: Seamlessly transfers complex cases to your team
• Continuous Learning: Improves from conversations and feedback

​

Common Use Cases

• Answer FAQs automatically
• Troubleshoot common issues
• Guide users through processes
• Escalate to human agents when needed

• Qualify inbound leads
• Schedule demos and meetings
• Answer product questions
• Collect contact information

• HR assistance for employees
• IT helpdesk automation
• Policy and documentation lookup
• Onboarding support

• Product recommendations
• Order tracking
• Return and refund assistance
• Shopping guidance

​

2. Creating Your First AI Employee

​

Quick Start

1. Navigate to AI Employees in the sidebar (or visit /agents)
2. Click “Create AI Employee” button
3. Fill in basic information:
   • Name: Descriptive name (e.g., “Customer Support Bot”, “Sales Assistant”)
   • Description: Purpose and responsibilities
   • Avatar: Optional profile image
4. Click “Create”

​

Initial Configuration

• Overview: Basic info and quick stats
• Settings: Core configuration options
• Tools: Connected integrations and capabilities
• Analytics: Performance metrics and insights
• Debug: Conversation logs and troubleshooting

​

Naming Best Practices

• “Customer Support Agent”
• “Lead Qualifier - Sales”
• “HR Assistant Bot”
• “Technical Support AI”

• Generic names like “Bot 1” or “Agent”
• Names without context
• Overly long names

​

3. Configuring AI Employee Settings

​

General Settings

• Display name shown to users
• Can be changed anytime
• Keep it professional and clear

• Internal notes about the AI Employee’s purpose
• Helps team members understand its role
• Not visible to end users

• Profile image representing the AI Employee
• Appears in chat interfaces
• Recommended: 200x200px, JPG or PNG

​

Instructions (System Prompt)

1. Role Definition:

   You are a friendly customer support agent for Acme Corp.
   Your goal is to help customers quickly resolve issues.

2. Tone and Personality:

   Use a professional yet warm tone.
   Be patient and empathetic.
   Keep responses concise and helpful.

3. Capabilities:

   You can:
   - Answer questions about our products
   - Check order status
   - Process returns
   - Escalate to human agents for complex issues

4. Limitations:

   You cannot:
   - Process payments directly
   - Access customer credit card information
   - Make promises about delivery dates

5. Guidelines:

   Always greet users warmly.
   Ask clarifying questions if needed.
   Provide step-by-step guidance for technical issues.
   Offer to connect to a human agent if you cannot help.

You are Alex, a customer support specialist for ZappWay. 
You help users with questions about our AI automation platform.

PERSONALITY:
- Friendly, professional, and patient
- Use clear, simple language
- Show empathy for user frustrations

CAPABILITIES:
- Answer questions about features and pricing
- Guide users through setup processes
- Troubleshoot common issues
- Recommend relevant documentation
- Collect information for complex issues

PROCESS:
1. Greet the user warmly
2. Understand their question or issue
3. Search the knowledge base for accurate information
4. Provide clear, step-by-step solutions
5. Confirm the issue is resolved
6. Offer additional help if needed

WHEN TO ESCALATE:
- Technical bugs or errors
- Billing or payment issues
- Custom enterprise requests
- User explicitly requests human agent

Always end conversations by asking if there's anything else you can help with.

​

Model Selection

​

🧠 GPT Models

• GPT-5.1 - Chat
• GPT-5.1
• GPT-5 Nano
• GPT-5 Mini
• GPT-5
• GPT-4o Mini
• GPT-4o

​

🔶 Claude

• Claude 4.5 Sonnet
• Claude 4.5 Haiku
• Claude 4 Sonnet
• Claude 3.5 Sonnet
• Claude 3.5 Haiku

​

🔵 DeepSeek

• DeepSeek R1
• DeepSeek V3

​

🟡 Google Gemini

• Google Gemini 3 - Pro
• Google Gemini-Pro 2.5
• Google Gemini-Flash 2.5
• Google Gemini-Flash 2.0

​

🔷 Meta Llama

• Meta: Llama-4 Scout
• Meta: Llama-4 Maverick

​

🟣 Perplexity

• Perplexity: Sonar-Pro

​

⚫ xAI Grok

• xAI: Grok 4.1 - Fast
• xAI: Grok 4

​

🟤 Amazon

• Amazon: Nova-Premier

​

🟢 Dolphin / Mixtral / Mistral

• Mixtral 8x22B
• Mixtral 8x7B
• Dolphin Mixtral 8x7B

Note:

• GPT-5.1: Most capable, best for complex tasks (slower, higher cost)
• GPT-5 Mini: Balanced performance and speed
• GPT-5.1 ‘Chat’: Fast and cost-effective for simple tasks

• Start with GPT-5 Mini for most use cases
• Use GPT-5.1 for highly nuanced conversations
• Use GPT-5.1 ‘Chat’ for high-volume, simple queries

​

Temperature Setting

• 0.0 - 0.3: Very consistent, deterministic (good for factual Q&A)
• 0.4 - 0.7: Balanced creativity and consistency (recommended)
• 0.8 - 1.0: More creative and varied (good for conversational AI)

• Customer Support: 0.3 (needs consistent, accurate answers)
• Sales Assistant: 0.6 (balance between personality and accuracy)
• Creative Writer: 0.9 (more varied, engaging responses)

Important:

• GPT-5 models do not support manual temperature adjustment, as they come configured with automatic temperature recognition..

​

Visibility Settings

• Public: Anyone with the link can chat
• Private: Only authenticated users in your organization

​

4. Adding Knowledge with Datastores

​

Why Add Knowledge?

• AI relies only on general training data
• May provide outdated or incorrect information
• Cannot answer company-specific questions

• Answers based on your actual documentation
• Always up-to-date with your content
• Provides accurate, relevant information

​

Connecting a Datastore

• At least one datastore created with content
• See Datastore Documentation for setup

1. Open AI Employee settings
2. Navigate to Tools tab
3. Click “Add Tool”
4. Select “Datastore Tool”
5. Choose datastores to connect:
   • Click “Select Datastores”
   • Choose one or multiple datastores
   • Click “Save”
6. Tool is now active

• Product Documentation datastore
• FAQ datastore
• Policy Documents datastore

​

Datastore Tool Configuration

• Max Results: How many chunks to retrieve (default: 5)
• Similarity Threshold: Minimum relevance score (0-1)
• Priority: Which datastore to search first (if multiple)

• Connect focused, relevant datastores
• Keep datastores updated regularly
• Use 3-5 datastores max per AI Employee

• Connecting 10+ datastores (slows responses)
• Including irrelevant content
• Duplicate information across datastores

​

5. Connecting Tools

​

Available Tools

​

Datastore Tool

• Search knowledge bases for information
• Required for: Answering company-specific questions
• See: Datastore Tool Guide

​

HTTP Tool

• Make API requests to external systems
• Required for: Integrating with custom services
• See: HTTP Tool Guide

​

Form Tool

• Present and collect structured data via forms
• Required for: Lead capture, ticket submission
• See: Form Tool Guide

​

Request Human Tool

• Escalate conversation to human agent
• Required for: Complex issues needing human touch
• See: Human Handoff Guide

​

Mark as Resolved Tool

• Mark conversation as completed
• Required for: Clean conversation management
• See: Mark as Resolved Guide

​

Lead Generation Tool

• Automatically create and qualify leads
• Required for: Sales automation
• See: Lead Generation Guide

​

Adding Tools

1. Go to AI Employee → Tools tab
2. Click “Add Tool”
3. Select tool type from list
4. Configure tool settings
5. Click “Save”

• Endpoint URL: API endpoint to call
• Method: GET, POST, PUT, DELETE
• Headers: Authentication and content type
• Body Template: Data to send

​

Tool Usage Examples

Tools: 
- Datastore Tool (Help docs)
- Request Human Tool (Escalation)
- Mark as Resolved Tool

Tools:
- Datastore Tool (Product info)
- Form Tool (Lead capture)
- Lead Generation Tool
- HTTP Tool (CRM integration)

Tools:
- Datastore Tool (Product catalog)
- HTTP Tool (Order lookup API)
- Form Tool (Return requests)

​

6. Deploying Your AI Employee

​

Deployment Options

​

Web Chat Widget

• Website support
• Product pages
• Landing pages
• Documentation sites

​

Messaging Platforms

​

E-commerce Platforms

​

API Integration

​

Testing Before Deployment

• Test conversations with AI Employee
• Verify datastore retrieval works
• Check tool executions (forms, APIs)
• Test human handoff flow
• Try edge cases and error scenarios
• Confirm responses match instructions
• Test on mobile devices (for web chat)

1. Go to AI Employee detail page
2. Use Test Chat interface (right panel)
3. Try various user questions and scenarios
4. Check debug logs for tool usage
5. Refine instructions based on results

​

7. Monitoring and Optimization

​

Analytics Dashboard

• Total Conversations: Number of chats handled
• Messages Sent/Received: Activity volume
• Resolution Rate: % of conversations resolved without human
• Average Response Time: How fast AI responds
• Tool Usage: Which tools are called most often
• Knowledge Base Hits: Datastore search frequency

​

Debug and Logs

• Full conversation transcripts
• Tool calls and responses
• Datastore searches and results
• Error messages
• Processing time for each step

1. Go to AI Employee → Debug tab
2. Select a conversation from list
3. Review step-by-step execution
4. Identify issues (wrong tool, bad retrieval, etc.)
5. Adjust configuration accordingly

​

Conversation File Upload

• PDF documents
• Images (JPG, PNG)
• Text files
• Word documents

​

Answer Sources

• Builds trust (citations visible)
• Helps users find original docs
• Improves transparency

​

Message Suggestions

• “Check order status”
• “Return an item”
• “Speak to a human”
• “Product recommendations”

1. AI Employee settings
2. Message Suggestions section
3. Add suggested messages
4. Save

​

Continuous Improvement

• Review conversation logs
• Check resolution rates
• Identify common failure points
• Update datastores with new content

• Analyze analytics trends
• Refine system instructions
• Add new tools if needed
• Update knowledge base comprehensively

• Major instruction overhaul
• Evaluate model performance
• Consider fine-tuning
• Review ROI and impact

​

8. Best Practices

​

Instruction Writing

When a user asks about pricing, always mention our 7-day free trial first,
then provide pricing tiers with feature breakdowns.

Help users with pricing questions.

Example conversation:
User: "How much does it cost?"
You: "Great question! We offer a 7-day free trial to get started. After that, we have three plans:
- Growth: $29/month for up to 1,000 conversations
- Pro: $99/month for up to 10,000 conversations
- Enterprise: Custom pricing for unlimited conversations
Which plan sounds right for you?"

DO NOT:
- Make promises about delivery dates
- Offer discounts without approval
- Share confidential information
- Provide medical advice

​

Knowledge Base Organization

• Create focused datastores by topic
• Example: “Billing FAQ”, “Product Docs”, “Policies”

• Remove outdated information promptly
• Add new content regularly
• Version documentation clearly

• Use clear headings
• Break information into digestible sections
• Include keywords users might search for
• Avoid jargon without explanations

​

Tool Configuration

• More tools = more complexity
• AI must decide which tool to use
• Start minimal, add tools as needed

• Verify API endpoints work
• Check authentication
• Test error handling
• Monitor API rate limits

​

User Experience

"Hi! I'm Alex, your AI assistant. I can help you with:
✓ Product questions
✓ Order tracking
✓ Account issues

For billing or technical bugs, I'll connect you to our team."

"I'm not sure I can help with that. Would you like me to:
1. Connect you to a human agent
2. Search our help center
3. Try rephrasing your question"

"I found your order #12345. Would you like me to:
- Check its status
- Process a return
- Contact support about it"

​

Security and Privacy

• Don’t include passwords or API keys in instructions
• Use environment variables for credentials
• Limit datastore access to necessary information
• Comply with GDPR/privacy regulations

1. AI Employee settings
2. Rate Limiting section
3. Set limits per user/IP

​

📞 Support & Resources

​

Getting Help

• Help button in dashboard
• Live chat (available on Pro+ plans)

• Tools Overview
• Human Handoff
• Fine-Tuning
• Debug Guide

• Widgets Documentation
• Integration Guides
• API Reference

​

Feedback

• Use feedback button in dashboard
• Email: [email protected]
• Include:
  • AI Employee ID
  • Conversation ID (if applicable)
  • Expected vs actual behavior

​

✅ Quick Reference

​

Essential Actions

​

Configuration Checklist

✓ Name and description set
✓ Avatar uploaded
✓ Instructions written clearly
✓ Model selected (GPT-5 recommended)
✓ Temperature configured (0.6 recommended)
✓ Datastore tool connected
✓ Knowledge bases populated
✓ Additional tools added as needed
✓ Tested with sample conversations
✓ Rate limits configured
✓ Deployed to target channels

> **Important:**
> - **GPT-5** models do not support manual temperature adjustment, as they come configured with automatic temperature recognition..

---