The Performance page in your Freddy AI Agent's Analyze tab is a key tool for understanding its engagement metrics and performance outcomes. This article will explain these sections, what the key metrics mean, and how to use them to optimize your AI Agent's performance.

Knowledge and workflow usage

Below the overview strip, you will find detailed breakdowns of the knowledge source content and Workflows that the AI Agent is using to respond to conversations. 

Knowledge usage

Each row corresponds to a specific knowledge source, such as:

• Files: Documents like PDFs from which the AI agent can extract information.
• URLs: External web links provided to the AI agent as a knowledge source.
• Solution articles: FAQs or solution articles in your Freshchat knowledge base.
• Q&As: Predefined question-and-answer pairs that are configured within the AI agent.

For each content source, the following details are provided:

• Title: The name or URL of the source.
• Source Type: Whether the content is derived from a URL, file, Solution article or Q&A.
• Sent: The number of times this content has been sent in a conversation.
• Helpful: The number of users who marked this response as helpful.
• Not Helpful: The number of users who marked this response as unhelpful.

Note: You can filter the list by source type and search for a specific source.

Workflow usage

The Workflow usage tab gives you details on:

• Workflow: The name of the Workflow used by AI agent(s).
• Triggered: The number of times a specific skill was activated.
• Helpful: The number of upvotes a skill received from customers.
• Unhelpful: The number of downvotes a skill received from customers.

Note: You can sort the values in either ascending or descending order.

Actionable insights to improve your AI Agent performance

Ensure the content in your Files, URLs, Q&A pairs, and Solution articles is clear, concise, and relevant to user queries. Regularly updating and expanding your knowledge base with fresh content is critical to improving resolution rates and reducing agent transfers.

• Low Helpful Marks: If a piece of content has been sent frequently but has received few or no "Helpful" marks, it may indicate that the content is irrelevant or easy to understand. Review the content to ensure it effectively addresses the users’ needs.
• High Agent Transfer Rate: If your agent transfer rate is high, investigate whether the AI Agent encounters questions it hasn’t been trained to answer. You may need to expand the AI Agent's knowledge base or refine its training to improve performance.
• Underutilized Content: If URLs, files, or Q&A pairs are rarely used, ensure the AI Agent is properly configured to pull relevant information. Check whether these resources are needed and whether the AI Agent is correctly referencing them.
• Focus on Resolution Rate: Aim to increase the resolution rate by providing more comprehensive training for the AI Agent. Incorporating additional content from URLs, files, Solution articles, and Q&A can help the AI Agent respond to more user queries independently.