Decision Trees

Decision Trees enable CX Assistants to guide helpdesk agents through structured decision-making processes. By connecting Assistants to decision trees, you can ensure agents receive consistent, step-by-step guidance for complex procedures, troubleshooting workflows, and verification processes.

Overview

Decision Trees provide interactive guidance that adapts based on the conversation context:

1. Structured Guidance: Pre-defined decision paths for common scenarios
2. Context-Aware: Analyzes conversation history to provide relevant steps
3. Interactive: Asks clarifying questions when information is needed
4. Vector Search: Uses semantic search to find the most relevant decision tree

How It Works

1. Decision Tree Setup

Create a decision tree that will be populated with structured procedures, troubleshooting steps, or verification workflows. The content is stored in a searchable vector index.

2. Connect Assistant to Decision Tree

Configure an Assistant with a decision_tree reference. This automatically:

• Adds the retrieve_decision_tree tool to the Assistant
• Injects instructions for using decision trees
• Enables step-by-step guidance based on decision tree content

3. Retrieval During Evaluation

When the Assistant receives a question:

1. The LLM calls the retrieve_decision_tree tool with the user's input
2. The tool includes conversation history for context
3. The decision tree system performs semantic search to find relevant procedures
4. Guidance is returned with next steps or clarifying questions
5. The LLM presents the guidance to the agent

Key Features

Context-Aware Guidance

The tool automatically:

• Includes conversation history to understand the current situation
• Extracts relevant information to skip unnecessary questions
• Provides targeted guidance based on available context

Interactive Decision Making

The Assistant will:

• Ask simple, numbered multiple-choice questions when information is needed
• Think out loud to show which steps are being taken and why
• Only ask for confirmation when uncertain
• Request one piece of information at a time

Semantic Search

Decision trees use vector-based search to:

• Find the most relevant procedure for the agent's question
• Handle varied phrasing and terminology
• Retrieve multiple related decision paths when applicable

Conversation Integration

The tool tracks:

• Agent-customer conversation history
• Previous interactions in the current conversation
• Conversation IDs for continuity across multiple tool calls

Use Cases

Customer Verification

Guide agents through customer verification processes:

• Identity verification steps
• Account security checks
• Authentication procedures

Troubleshooting Workflows

Provide systematic troubleshooting guidance:

• Technical issue diagnosis
• Step-by-step resolution procedures
• Escalation criteria and paths

Policy Application

Help agents apply policies consistently:

• Return and refund policies
• Exception handling procedures
• Approval workflows

Escalation Management

Structure escalation decisions:

• When to escalate to a supervisor
• Information required for escalation
• Proper escalation procedures

Comparison with Knowledge Assist

Feature	Decision Trees	Knowledge Assist
Purpose	Structured, step-by-step guidance	Information retrieval
Content Type	Procedures, workflows, decision paths	Documents, articles, FAQs
Interaction	Interactive with questions	Passive information delivery
Context Use	Conversation-aware	Query-based
Output	Next steps and questions	Document excerpts
Use Case	"What should I do?"	"What information exists?"

Prerequisites

Before setting up Decision Trees:

1. Account Configuration: Knowledge Assist must be enabled for your account (deploy_knowledge_assist flag)
2. Decision Tree Content: Have structured procedures ready to add to the decision tree
3. Assistant Configuration: Create or identify the Assistant to connect
4. Vector Index: Decision tree content must be indexed in Elasticsearch

Next Steps

• User Guide: Step-by-step configuration instructions
• Developer Guide: Technical implementation details and API reference