Analysis Agent

Analysis Agents produce structured data about an interaction that can be consumed by other products or tools. The structured data can inform next steps during an experience or be used to evaluate performance after an experience is concluded.

 

Agent Methodology

  • A conversation (or subset of a conversation) is sent to the agent.

  • The agent captures metrics or meta data about the interaction, in a structured data format.

  • Depending on your configuration, the agent will either fire at least one tool or produce JSON formatted output.

 

Channel Compatibility

The Analysis Agent is channel agnostic (e.g., compatible for inbound or outbound voice, SMS, etc.).

 

Use Cases

Analyze call recordings to determine efficiency of the interaction, trends, potential gaps, etc.

For example, your agent might review a transcript to capture call center interaction data (a general performance score, problem resolution skills, etc.) and send that structured data to SmartAnalytics to easily view trends across all company call centers.

 

Configurations

  • System Prompt

  • Output Structure

    • Not available with any other Agent Type

    • Either "Tools" or "JSON Output"

  • Parameters

 

Tips and Best Practices

  • Model “intelepeer: gpt-4o-mini…” is the only option supported for Analysis Agents.

  • Output Structure:

    • If you select the “JSON Output” you must define the schema that is produced by your agent.

      • When using this agent in conjunction with a flow in SmartFlows, where you are sending data via an External Web Call action, we recommend typically using the Output Structure “JSON Output”.

    • If you select the “Tools” you must define at least one tool.

      • Make sure your tools are well built. If your tools are poorly written, your agent will not perform as expected.

      • When using this agent to review a transcript for example, we recommend using the “Tools” Output Structure.

      • We recommend including only one tool in your agent, with multiple attributes/properties.

        • For example, you could use the “evaluate_conversation” tool with properties for “performance_evaluation”, “performance_score”, “problem_resolution_evaluation”, “problem_resolution_score”, etc.

 

For more information, check out How to Build an Analysis Agent.