Custom Metrics

Custom metrics let you define your own evaluation criteria for conversations, on top of the built-in system metrics. Each custom metric is defined per project and is automatically evaluated on every conversation in that project.

For a full description of each metric type, see the Metrics Glossary.

Managing custom metrics

The Custom Metrics screen shows all metrics defined for the currently selected project. From here you can:

Create a new metric — define a new evaluation criterion for this project
Duplicate a metric — create a copy of an existing metric to modify it. Once the new version is ready, activate it and deactivate the old one to preserve historical results.
Activate or deactivate a metric — only active metrics are evaluated on incoming conversations
Delete a metric — permanently remove a metric and its results

ℹ️ When a new custom metric is added to a project, all users with access to that project receive a notification.

Creating a custom metric

When creating a metric, you choose a type that determines how conversations are evaluated:

Type	How it works	Output
String Match	Checks for specific keywords or phrases	True / False
Regex Match	Matches a text pattern	True / False
LLM Judge	AI evaluates against custom pass/fail criteria	True / False
Categorization	AI classifies the conversation into one of your categories	Category name

Each metric also has a name and optional description to help your team understand what it measures.

Viewing metric results

Once a metric is active, its result appears on every conversation in the Conversation List and in the conversation detail view. For LLM Judge metrics, the reasoning behind each result is also stored and visible in the detail view.