Use Global Rules in your Agent Settings to define how your assistant behaves across all their interaction. This includes its language, task execution, its management of difficult scenarios, and its compliance with your business requirements. By configuring rules effectively, you can ensure a durable and consistent customer experience.

​
What are rules?

Rules guide the behavior of your assistant by providing specific instructions on how it should respond to users, perform tasks, and manage sensitive scenarios. This section is crucial for:

  • Defining the assistant’s tone, language, and mannerisms.

  • Ensuring compliance with organizational policies.

  • Managing edge cases or high-risk interactions.

​
Types of rules

​
1. Behavior and interaction guidelines

Set rules for how the assistant interacts with users, including:

  • Tone: Specify whether the tone should be formal, casual, empathetic, etc.

    • Example: “Always remain polite and professional, even when users are frustrated.”

  • Language Style: Define language-specific preferences (e.g., avoiding jargon, using simple terms).

    • Example: “Use simple, clear language suitable for non-technical users.”

  • Consistency: Ensure the assistant adheres to company branding and messaging.

​
2. Task execution

Define how the assistant should perform specific tasks:

  • Clear Instructions: Be explicit about the actions required.

    • Example: “If a user asks about refunds, provide the refund policy link and offer to send it via SMS.”

  • Response Limitations: Restrict the assistant to specific knowledge areas or tasks.

    • Example: “Only provide information about your account’s services; do not answer general queries outside this domain.”

​
3. Content restrictions

Establish boundaries for what the assistant can and cannot say:

  • Sensitive Topics: Prohibit responses on certain subjects.

    • Example: “Do not discuss politics, religion, or personal opinions.”

  • Accuracy: Instruct the assistant to avoid guessing or fabricating responses.

    • Example: “If unsure, direct the user to a human agent or a verified source.”

​
Best practices

  1. Be specific and actionable

    • Use precise language to avoid ambiguity.

    • Example: Instead of “Be helpful,” use “Answer all questions about billing within two sentences and provide follow-up actions if needed.”

  2. Provide examples

    • Include examples of expected user interactions and appropriate responses to guide the assistant’s behavior.

    • Example:

      • User: “What is your refund policy?”

      • Assistant: “You can find our refund policy here: [link]. Would you like me to text this link to your phone?”

  3. Plan for edge cases

    • Address high-risk scenarios (e.g., emergency calls, user frustration).

    • Example: “For emergency-related queries, immediately request the user terminate the call and contact 911.”

  4. Test and iterate

    • Regularly review the assistant’s performance and refine rules based on real-world interactions.

​
Example rules for common scenarios

  • Handoff to human agent

    • “If the user explicitly asks for a human agent or uses language suggesting dissatisfaction, trigger a call handoff.”

  • Sensitive query handling

    • “For any query outside the assistant’s domain, respond: ‘I’m sorry, I can’t assist with that. Please contact our support team for further help.’”

  • Consistency in responses

    • “Always begin responses with ‘Thank you for reaching out!’ to maintain a friendly tone.”

Make sure you regularly monitor and update rules to ensure they align with your business goals and user expectations.

RAGIntroduction