Skip to main content

Conversation

The AI agent handles conversations with leads across multiple channels (email and SMS). Understanding how conversations work helps you configure and monitor agent behavior effectively.

How Conversations Work

Incoming Messages

When a message is received:
  1. Agent identifies the lead: Matches the message to an existing lead or creates a new one
  2. Processes the message: Analyzes the content and any attachments
  3. Determines response: Decides what action to take based on the message
  4. Executes tools if needed: Uses tools to retrieve information or perform actions
  5. Generates response: Creates an appropriate response based on context

Outgoing Messages

When the agent sends a message:
  • Autopilot mode: Message is sent immediately
  • Listen mode: Message is drafted for admin review before sending

Documents in Conversations

If documents are passed in:
  1. Documents are processed: OCR, classification, and validation
  2. Checklist is updated: Documents are matched to checklist items
  3. Data is extracted: Information is pulled from documents
  4. Response is generated: Agent acknowledges receipt and provides feedback

Message Handling

New Leads

If the message is from a new lead:
  • A new lead is created automatically
  • The agent introduces itself and explains the process
  • The agent requests required information based on the checklist

Existing Leads

If the message is from an existing lead:
  • The agent retrieves the lead’s current status
  • The agent continues from where the conversation left off
  • The agent updates checklist items based on new information

Multi-channel Coordination

The agent coordinates across channels:
  • Email and SMS: Messages can come from either channel
  • Unified context: Agent maintains context across all channels
  • Channel selection: Agent chooses the best channel for follow-ups

Escalation

If agent message could possibly violate lending policy, submit feedback to the agent. The system can be configured to escalate certain situations to human agents.