Defining goal

The goal defines the purpose of your AI skill. It is a high-level statement that guides the AI's behavior and clarifies its role in assisting the human agent.

• Focus on agent assistance: Clearly articulate how the AI will help the human agent. The goal should always reflect the AI's role as a support tool for the agent, not as a direct customer-facing entity.

Example: "You are a helpful and polite assistant that will help agents answer queries regarding lost baggage and recommend necessary actions."

• Keep it concise and action-oriented: A short, clear goal helps the AI maintain focus.
• Align with skill capabilities: Ensure the goal is realistic and achievable based on the Knowledge Base content and actions defined for the skill.

For detailed steps on setting the goal, refer to the Create and manage AI Assistant skills article.

Crafting instructions

Instructions provide detailed guidance to your AI skill on how to process information and generate suggestions. This section differentiates your AI Assistant skill from an AI Agent, as these instructions are for the AI to assist the agent.

• Define skill’s persona (as an assistant to the agent): Clearly state the skill's role and expertise as an assistant to the human agent.

Example: "You are an expert AI assistant for agents handling billing inquiries. Your role is to analyze customer conversations and provide agents with the most relevant information and actions to resolve billing questions."

• Outline tasks and decision flow: Break down the overall task into specific, sequential steps from the AI's perspective. Guide the AI on what to look for in the conversation and what type of suggestion or action to provide.

Example: "First, listen for the customer's primary issue regarding their lost baggage. Then, if flight details are needed, suggest the agent use the 'Retrieve Flight Details' action. If a claim needs to be filed, suggest the agent use the 'Create Lost Baggage Claim' action."

• Specify suggestion formatting: Instruct the AI on presenting information clearly and concisely for the agent.

Example: "Suggestions should be presented in clear, bulleted points for easy readability. Action suggestions should clearly state the action name and its purpose for the agent."

• Reference actions clearly: When the AI should suggest an action, explicitly reference the configured action name.

Example: "If the customer mentions 'fraudulent charge,' suggest the agent use the <Initiate Fraud Investigation> action."

• Plan for error handling and fallbacks: Instruct the AI on how to respond when it cannot provide a confident or relevant suggestion.

Example: "If you cannot provide a confident suggestion for the agent's current context, inform the agent by stating: 'No relevant suggestion available. Please refer to the knowledge base or consult a supervisor.'"

• Add guardrails (scope of assistance): Remind the AI to stay within its defined scope.

Example: "Your suggestions must always be focused on assisting the agent with the customer interaction. Do not attempt to answer non-billing questions or interact directly with the customer."

For detailed steps on providing instructions, refer to the Create and manage AI Assistant skills article.

Structuring Knowledge Bases

The Knowledge Base is the factual foundation for your AI skill. Its organization should prioritize information that helps an agent respond or act effectively.

• Agent-centric content: Prioritize information agents frequently need to explain, troubleshoot, or act upon. Focus on content that is directly useful for an agent's workflow.

Example: “For a "Refund Policy" article, include not just the policy text but also common customer questions and practical steps for processing a refund.”

• Organize content logically: Use categories within your KB to group related information. This helps both the AI and agents navigate efficiently and improves suggestion relevance.
• Ensure accuracy and consistency: Verify that all information is accurate and up to date. Avoid conflicting or outdated content.
• Optimize document quality:
  • Clarity: Use plain language.
  • Conciseness: Be direct; AI and agents need quick answers.
  • Structure: Use headings, subheadings, bullet points, and numbered lists for readability and to help the AI extract key information.
  • File size: Consider breaking down very large documents into smaller, topic-specific ones to improve retrieval speed and relevance.
• Regular review and update: Continuously review your KB content to ensure it remains relevant and accurate. Update information whenever policies, products, or processes change.

For detailed steps on linking a knowledge base to a skill, refer to the Create and manage AI Assistant skills article.

Defining actions

Actions define specific tasks the AI skill can suggest or perform. When configuring actions, consider their presentation to the agent and their impact on workflow.

• Clearly define action objectives: The Action name and Action description should be clear, concise, and immediately understandable to the agent.

Example: Action name: "Retrieve Order Status"; Action description: "This action will fetch the current status of a customer's order using their order ID."

• Minimize complexity: Keep individual actions simple and focused. Break complex, multi-step processes into smaller, distinct actions.
• Accurately describe user inputs: For each user input (slot), provide a clear description to help the AI accurately identify and collect needed information.
• Choose appropriate fulfillment mode: Educate agents on the differences between the fulfillment modes:
  • Unmoderated mode: For routine, low-risk actions requiring no agent review (example, logging a simple interaction detail). Ensure robust error handling.
  • Moderated mode: For actions requiring agent verification, input, or approval (example, submitting a form, confirming sensitive data). This empowers the agent and ensures accuracy.

For detailed information, agents can be referred to the Understand and manage AI-suggested actions article.

• Design for agent workflow: Consider how the action will appear in the Agent Desktop. Use the Card Layout configuration to influence information presentation.

For detailed steps on configuring actions, refer to the Configure actions for AI Assistant skills article.