Skip to main content
Skip table of contents

Feedback in HALO

The Feedback Mechanism in HALO is a vital Quality Assurance feature that allows you to provide constructive feedback on AI-generated answers. By refining HALO's interactions with inquiries, you can improve the accuracy and relevance of the responses provided. This mechanism enables greater insights and control over HALO's functionality.

Tip: Check out our Discover Video on the Feedback Mechanism for more details.

Providing Feedback

In HALO, feedback can be provided through individual conversations. To begin, navigate to the Conversations tab, where you can search for and select a specific conversation to view in full detail. Within this view, you can review all messages exchanged during the conversation and rate HALO's responses by liking or disliking them. An example conversation is displayed below for reference.

image-20250311-121630.png

Types of Feedback

Positive Feedback

  • Like: If you are satisfied with HALO's answer, you can like it. Liked responses help enhance model consistency and may contribute to future fine-tuning.

image-20250311-121849.png

Negative Feedback Options

  • Dislike: If dissatisfied with a response, mark it as disliked and specify the reason:

    • I don't like the answer: Select this option if the answer is mostly correct but unsatisfactory in phrasing or format. Provide detailed feedback on desired modifications.

    • Wrong answer: Use this option if HALO provided an incorrect response, prompting you to review and update the knowledge base.

    • Do not answer: Choose this option to prevent AI from addressing similar questions. Future inquiries may either prompt a transfer to a live agent or remain unanswered.

image-20250311-131608.png

Feedback for "I Don’t Like the Answer"

For "I Don’t Like the Answer," you can provide specific guidance on improving phrasing, structure, or format. After testing the suggested changes, save your feedback to ensure improvements are applied to future similar questions.

image-20250311-122141.png

Tips on Providing Effective Feedback for "I Don’t Like the Answer"

  • Validate correctness: Provide feedback only if the answer is factually correct but unsatisfactory; update the Knowledge base for incorrect answers. Do not reference customer questions in feedback; the system matches feedback to questions.

  • Specify changes: Clearly state desired changes rather than rewriting the entire answer.

  • Avoid Knowledge references: The Feedback Mechanism focuses solely on improving given answers without additional context from the Knowledge base.

  • Use simple, directive language: Examples include "Write GenAI as Generative AI", "Format as a numbered list", "Do not mention our Email Channel", etc.

Adding Knowledge

Enhance HALO's responses by adding missing knowledge directly through the Conversations Screen. Use the “Add Knowledge” button to input new information into your Knowledge base, ensuring comprehensive responses in future interactions.

image-20250311-122756.png

Upon clicking the "Add Knowledge" button, a new window opens where additional knowledge can be entered.

image-20250311-122901.png

Once you click Save, the newly added information is stored in the Knowledge tab located in the left-side menu, under the resource named 'Knowledge'.

image-20250311-123411.png

All added knowledge can be accessed here. By opening this knowledge source and clicking ‘Content,’ you can view all the knowledge bits provided.

image-20250311-123317.png

Viewing and Editing Feedback

The Feedback tab allows you to review all provided feedback. You can see feedback types, details, and use the actions menu to edit or delete past feedback entries.

image-20250311-132900.png

By leveraging HALO's Feedback Mechanism, you can ensure enhanced precision and alignment with your organization's communication goals. Implement these practices to optimize AI response quality and user satisfaction.

JavaScript errors detected

Please note, these errors can depend on your browser setup.

If this problem persists, please contact our support.