Understanding the Breakout Tool

The Breakout Tool is a versatile feature in HALO that serves multiple purposes, including live agent handovers, routing between agents, and analytics tracking. This article will provide a comprehensive overview of the tool, its use cases, and best practices for implementation.

What is the Breakout Tool?

The Breakout Tool is a mechanism in HALO that allows AI agents to "break out" of their current flow when they encounter a situation they cannot handle. This tool is commonly used for:

1. Handover to Live Agents: Triggering a live chat handover to customer service agents.

2. Routing Between Agents: Redirecting the conversation to the root agent or another specialized agent.

3. Fallback Scenarios: Handling situations where the AI cannot provide an answer, based on predefined fallback prompts.

4. Analytics Tracking: Measuring the effectiveness of AI agents by tracking how often they break out of their flows.

How Does the Breakout Tool Work?

The Breakout Tool operates differently depending on the agent that calls it. Here's a breakdown:

1. Root Agent Behavior

• When the Root Agent calls the Breakout Tool, it directly triggers the handover to a live agent or initiates the fallback flow (depending on your configuration in the Style Tab).

• The Root Agent acts as the central decision-maker, determining whether to escalate to live support or redirect to another AI agent.

2. Non-Root Agent Behavior

• When a Non-Root Agent (e.g., a specialized agent like "Order Agent") calls the Breakout Tool, the flow is first redirected to the Root Agent.

• The Root Agent then decides the next step:

  • If the Root Agent also calls the Breakout Tool, the handover to a live agent is triggered.

  • If the Root Agent sends a textual response or redirects back to the Non-Root Agent, the handover flow is not triggered, potentially causing inconsistent behavior.

Common Issues and Solutions

Issue: Inconsistent Handover to Live Agents

When a Non-Root Agent calls the Breakout Tool, the handover may not always be triggered, leading to inconsistent behavior.

Solution: Force the Root Agent to Be Handoff-Only

To ensure consistent handover behavior:

1. Go to the Agent Settings for your Root Agent.

2. Configure the Root Agent to be Handoff-Only. This means the Root Agent can only:

   • Call the Breakout Tool.

   • Redirect to another AI agent (e.g., Knowledge Agent, Order Agent).

3. Avoid adding textual responses or complex logic in the Root Agent when handling Breakout calls.

This setup ensures that any Breakout call from a Non-Root Agent will consistently result in a live agent handover.

Alternative: Using the Live Chat Handover Tool

For more direct control over handovers, you can use the Live Chat Handover Tool instead of the Breakout Tool. Here's how:

1. Import the Tool: Import the Live Chat Handover Tool from the tool templates in HALO.

2. Configure Parameters: Set up the tool with all relevant parameters (e.g., live chat platform, agent group, etc.).

3. Set Handover Type: In the Style Tab, change the Handover Type to "Live Chat Handover Tool."

4. Trigger from Any Agent: Use this tool to trigger handovers directly from any agent, including Non-Root Agents.

This approach eliminates the dependency on the Root Agent for handovers, providing more flexibility.

Best Practices for Using the Breakout Tool

1. Define Clear Boundaries for Agents:

   • Use the Breakout Tool to route questions outside the scope of a specialized agent back to the Root Agent.

   • Ensure the Root Agent has clear logic for handling Breakout calls.

2. Leverage Fallback Prompts:

   • Configure fallback prompts in the Style Tab (field: "What should we do when we don't know the answer?") to handle unanswered questions gracefully.

3. Monitor Analytics:

   • Use Breakout and Complete Tool calls to track how often agents escalate or fail to answer questions.

   • Analyze this data to improve agent performance and training.

4. Test Handover Scenarios:

   • Regularly test handover flows to ensure consistent behavior, especially when using multiple agents.