AI Agent Overview
Your AI agent is the heart of Airmailer. It reads customer emails, understands their questions, searches your documents for relevant information, and crafts helpful, on-brand responses automatically.
How Your Agent Works
┌─────────────────────────────────────────────────────┐
│ CUSTOMER SENDS EMAIL │
└─────────────────────────┬───────────────────────────┘
↓
┌─────────────────────────────────────────────────────┐
│ AGENT UNDERSTANDS THE QUESTION │
│ │
│ "What's your return policy?" │
│ → Intent: Return policy inquiry │
└─────────────────────────┬───────────────────────────┘
↓
┌─────────────────────────────────────────────────────┐
│ AGENT SEARCHES YOUR DOCUMENTS │
│ │
│ Found: "Return Policy" document │
│ Relevant: 30-day return window, conditions │
└─────────────────────────┬───────────────────────────┘
↓
┌─────────────────────────────────────────────────────┐
│ AGENT GENERATES RESPONSE │
│ │
│ Applies: Your tone of voice │
│ Follows: Your agent rules │
│ Uses: Information from documents │
│ Assigns: Confidence score │
└─────────────────────────┬───────────────────────────┘
↓
┌─────────────────────────────────────────────────────┐
│ APPROVAL CHECK (OPTIONAL) │
│ │
│ Auto Send: Email sent immediately │
│ Draft All: Held for human review │
│ Threshold: Based on confidence score │
└─────────────────────────┬───────────────────────────┘
↓
┌─────────────────────────────────────────────────────┐
│ BRANDED EMAIL SENT TO CUSTOMER │
└─────────────────────────────────────────────────────┘
Agent Capabilities
Your AI agent can:
| Capability | Description | |------------|-------------| | Understand intent | Comprehends what customers are asking | | Search documents | Finds relevant information in your knowledge base | | Generate responses | Creates helpful, accurate answers | | Maintain context | Remembers conversation history | | Apply branding | Uses your tone and follows your rules |
Configuring Your Agent
Your agent's behavior is controlled through three main areas:
1. Documents (Knowledge Base)
The information your agent uses to answer questions.
2. Tone of Voice
How your agent communicates with customers.
3. Rules and Constraints
Specific behaviors and limitations for your agent.
Agent Settings Location
Navigate to Agent in the sidebar to configure:
- Tone of voice
- Agent rules and guidelines
- Response approval settings (approval mode, confidence threshold, escalation notifications)
See Approval Queue for details on configuring human-in-the-loop approval.
How Responses Are Generated
When a customer email arrives:
- Intent Recognition: Agent identifies what the customer wants
- Document Retrieval: Searches for relevant content in your documents
- Context Integration: Considers conversation history
- Response Drafting: Creates a response using retrieved information
- Tone Application: Adjusts language to match your configured tone
- Rule Compliance: Ensures response follows your rules
- Confidence Scoring: Assigns a confidence level to the response
- Approval Check: Routes based on your approval mode setting
- Delivery: Email sent with your brand styling (after approval if required)
Human-in-the-Loop Approval
You can configure your agent to require human approval before sending responses. This provides quality control and oversight for AI-generated emails.
Approval Modes
| Mode | Behavior | |------|----------| | Auto Send | All responses sent immediately | | Draft All | Every response requires approval | | Confidence Threshold | Low-confidence responses require approval |
When to Use Approval
Consider enabling approval for:
- New agent deployments (review quality before trusting automation)
- Sensitive topics (legal, financial, or medical inquiries)
- High-value customers (ensure personalized attention)
- Complex products (where accuracy is critical)
Learn more in the Approval Queue documentation.
Agent Quality
Response quality depends on:
| Factor | Impact | |--------|--------| | Document coverage | More relevant docs = better answers | | Document clarity | Clear, structured content works best | | Tone configuration | Well-defined tone = consistent voice | | Rule specificity | Detailed rules = more predictable behavior |
Best Practices
Optimize Your Documents
- Cover common customer questions
- Keep content current and accurate
- Use clear, structured formatting
- Include specific details (dates, prices, policies)
Set Clear Tone Guidelines
- Be specific about communication style
- Include examples of desired language
- Consider your brand personality
Define Helpful Rules
- Set clear boundaries
- Specify when to involve humans
- Include do's and don'ts
Understanding Agent Limitations
Your agent will:
- Only use information from your documents
- Follow your configured rules
- Maintain your brand voice
Your agent cannot:
- Access external websites in real-time
- Take actions outside email responses
- Access systems you haven't integrated
- Remember information not in documents
Monitoring Agent Performance
Track how your agent performs:
- Activity Dashboard - See success rates
- Activity Feed - Monitor individual interactions
- Audit Log - Review complete history
When Agents Need Help
Sometimes agents can't fully answer a question:
- Information not in documents
- Complex issues requiring human judgment
- Customer explicitly requests human support
Configure your rules to handle these cases appropriately.