Metrics

​

Understanding Coval Metrics

Metrics are essential tools for evaluating Voice and Chat Agents, acting as judges that assess call quality. They analyze call transcripts and/or recordings to provide evaluations based on specific criteria. Metrics help identify strengths and areas for improvement. Regular analysis of these metrics enables organizations to make data-driven decisions, enhancing agent performance and user satisfaction.

​

Types of Metrics

The metrics that you define can be run on both simulated calls and live-calls that you push to Observability.

The most important metrics to assess agent performance are:

• LLM-Judge metrics: Define custom prompts to evaluate your agents based on your prompt, your required steps to be performed, and desired outcomes
• Binary Tool Call metrics: Check if your tool calls (functions) have been performed correctly
• Categorical metrics: Define a set of categories/topics to filter topics of your conversations (good for exploratory call analysis)
• Audio metrics (voice-agents-only): Audio performance
• Audio LLM Judge: While LLM Judge metrics use the transcript as an input, Audio-LLM Judge metrics take the audio as an input. You can define testing prompts to ask questions on things like sentiment analysis
• Custom: If you have your own metrics that you want to upload to the Coval platofrm to run next to our built-in metrics, let us know.

​

Create a Metric

Add a Display Name: Give your metric a clear, descriptive name.

Select Manager Type: Choose what type of metric you want to define.

Question: Define the specific goal you want your agent to achieve. For evaluating the agent, always refer to the “assistant”

Description: Provide an internal description for better clarity and context.