Procedures define how your Care AI Agent responds to user queries. Proper configuration is essential to ensure accurate intent recognition, logical and goal-driven responses, safer testing and deployment, and easier long-term maintenance.

This guide outlines the best practices for organizing, configuring, writing, and managing your Care AI Agent procedures.

1. Organizing Multiple Procedures

As your procedure library grows, maintaining a clean structure is critical for your team during mapping and maintenance.

Folder Hierarchy

  • Structure Limits: Each brand supports one level of folders (nested folders are not supported), with a maximum of 20 procedures per folder.
  • Categorization: Group procedures logically by User Segment (e.g., VIP Players, New Users) or Functional Category (e.g., Payment Issues, Account Security).
  • Best Practice: Avoid creating generic folders such as "Misc" or "General," as these become difficult to manage at scale.

Procedure Naming

Internal names are not visible to the AI but are vital for the organization.

  • Use clear, descriptive internal names that include the platform, context, or scope.
  • Example: Instead of naming a procedure just "Refund", use "Refund - In-Game Currency (iOS)". This reduces mapping errors and improves clarity when multiple workflows exist for similar issues.

2. Trigger Configuration

Triggers determine exactly when a procedure is activated. Choosing the right trigger is essential for accurate AI routing.

  • When to Use Intents: Intents are the most accurate trigger method and should be your preferred choice when available. Use them for standard, clearly defined queries, frequently occurring issues, and scenarios requiring high-accuracy routing.
  • When to Use Topics: Use Topics for broader categories, niche use cases, or queries not covered by standard intents.

Note: When configuring a Topic, the Description field is mandatory and must be highly specific. Do not write vague descriptions. The description must include specific signals, example phrases, and keywords the AI should look for to significantly improve recognition accuracy.

  • Bad Description: Billing related issues
  • Good Description: User asks about double charges, credit card failures, or receipt validation for diamond packs.

3. Writing Procedure Instructions

Procedures should be structured for clarity, logic, and goal completion.

  1. Think in Goals, Not Just Actions: Avoid creating a flat list of 15–20 disconnected steps. Instead, structure your procedures using Objectives (Goals) that contain related Steps (Actions). Grouping steps logically under clear objectives improves readability, testing, and long-term maintenance.
    Example Structure:
    • Objective 1: Validate User Identity
      • Step 1: Ask for Player ID.
      • Step 2: Check if the ID exists.
    • Objective 2: Process Refund
      • Step 1: Check purchase date.
      • Step 2: Refund if the purchase is within 48 hours.
  2. Write Clear, Plain-English Steps: Use simple, natural language and avoid technical or overly complex phrasing. Write conditional logic naturally to improve both AI execution and human review.
    • Example: Instead of using complex logic formatting, simply write: If the player is a VIP, escalate immediately. Else, ask for a screenshot.

4. Managing Procedure States

Proper state management prevents unintended AI behavior and ensures safe, controlled releases.

Editing a Live Procedure

Editing a live procedure is not allowed.

To edit a live procedure, follow these steps:

  1. Duplicate the live procedure.
  2. Make your changes in the duplicate.
  3. Move the duplicate to Testing.
  4. Test it using a dedicated Testing Care AI Agent.
  5. Once validated, move the procedure to Live and map it to your Live Care AI Agent.

Testing and Draft Behaviors

  • Draft Delinking: When a procedure is moved from the Testing state back to Draft, it is automatically delinked from all Care AI Agents.
  • Testing Separation: Do not map a testing procedure to multiple Care AI Agents. Maintain a dedicated Testing Care AI Agent and map only Testing state procedures to it. Once fully validated, move the procedure to Live to map it to your Live Care AI Agent to ensure clean separation between environments.