Genesys Cloud - Main

 View Only

Sign Up

  • 1.  Agent Copilot: Context Usage and Reporting Features

    Posted 18 days ago

    Hi everyone,
    I'm currently exploring the Agent Copilot capabilities in Genesys Cloud and I have a couple of questions:

    1. Use of conversational context during search
      When Agent Copilot performs a search to suggest a response to the agent, does it also use the context of the ongoing conversation to enrich the query and provide more relevant suggestions?
      In other words, does AgentCopilot automatically analyze the interaction so far to improve the accuracy of its recommendations?

    2. Reporting on Copilot interactions
      Are there any reports or dashboards in Genesys Cloud that allow detailed analysis of:

      • individual Agent Copilot interactions,
      • which suggestions were generated,
      • and any performance metrics of the model?

    I'm trying to better understand how to monitor the effectiveness of Agent Copilot in our organization and how to optimize the content used by the AI.

    Thanks in advance for your support!


    #AICopilot(Agent,SupervisorAdmin)

    ------------------------------
    Vittorio Iessi
    Senior Consultant
    ------------------------------


  • 2.  RE: Agent Copilot: Context Usage and Reporting Features
    Best Answer

    Posted 18 days ago

    Hi Vittorio,

    Yes, Agent Copilot can leverage Content Search against the Knowledge Base and combine it with Copilot suggestions to gain more context from the ongoing interaction. By analyzing the conversation so far, Copilot can enrich the search and provide more relevant, context-aware recommendations to the agent, rather than relying only on isolated keywords.

    From a monitoring and reporting perspective, there are two main options:

    • Agent Copilot Performance (native UI)
      This provides a high-level view of Copilot adoption and usage, including how often suggestions are generated and used.

    • Copilot Query API (/copilot/query)
      This allows you to customize data extraction and analyze Copilot usage in much more detail - for example, by interaction and by user - enabling deeper visibility into how Copilot features are being used and their effectiveness.

    Using the native performance view for a macro perspective and the API for deeper, customized analysis is usually the best approach to monitor Agent Copilot effectiveness and continuously optimize the content powering the AI.

    We are using this same approach in a real production environment, and I shared some practical insights from a high-volume operation case here, which may be useful as a real-world reference:

    Building a Scalable Agent Copilot Strategy: Lessons from a Fintech Operation

    Hope this helps!



    ------------------------------
    Mateus Nunes
    Tech Leader Of CX at Solve4ME
    Brazil
    ------------------------------

    Original Message


  • 3.  RE: Agent Copilot: Context Usage and Reporting Features

    Posted 18 days ago

    Hi Mateus,

    thank you for your replay!

    I have another question about the content search.

    Fallow the simple example:

    Q: "When can I cancel an order?"

    A: " You can cancel an order after the status is in progress"

    Q: "How much time I have?"

    In this case in the second question the answer is not provided by the Agent Copilot.

    Instead if i Ask something like:

    Q: "How much time I have to cancel an order?"

    The Agent Copilot give me the correct answer.

    To better analyze this scenario I'd like to Know how Agent Copilot work, if it use the conversation history or not.

    Thanks in advance

    Best regards



    ------------------------------
    Vittorio Iessi
    Senior Consultant
    ------------------------------

    Original Message


  • 4.  RE: Agent Copilot: Context Usage and Reporting Features

    Posted 18 days ago

    Hi Vittorio,

    Great question - this example illustrates the behavior very well.

    Today, Agent Copilot (via Content Search) primarily uses the context of the current message, not the full conversational history across multiple separate messages.

    In your example:

    • When the customer asks "How much time I have?", the question by itself is ambiguous and does not contain enough context for Copilot to confidently match it to a knowledge article.

    • When the customer asks "How much time do I have to cancel an order?", all the required context is present in a single message, so Copilot can correctly retrieve and suggest the relevant content.

    It's also important to note that:

    • A single message may contain multiple intents or questions, and Content Search can return multiple related articles as long as they are semantically connected to the content of that same message.

    • The quality and structure of knowledge article curation plays a major role - especially how well articles are written to reflect real customer phrasing and variations.

    Additionally, Copilot's behavior is influenced by the Confidence Threshold:

    • A higher confidence threshold results in fewer but more accurate suggestions.

    • A lower confidence threshold increases the number of suggestions, but may surface content that is less relevant to the actual message context.

    In short, Copilot uses the context of the current utterance, not the accumulated context of previous messages in the conversation. For multi-turn scenarios like the one you described, the best results typically come from ensuring that key context is present in the same message or from improving knowledge article phrasing and confidence tuning.

    Hope this helps clarify the behavior.



    ------------------------------
    Mateus Nunes
    Tech Leader Of CX at Solve4ME
    Brazil
    ------------------------------

    Original Message


Related Content