Metrics
Metrics in EfficientAI are the way to describe success criteria and the list of metrics you want to track while evaluating your voice agents.
What metrics are
Metrics are the scoring rules applied to completed runs. They define what "good" means for your agent, and let you compare changes in prompts, providers, and configurations with consistent criteria.
EfficientAI supports four metric families:
| Family | Measures | Examples |
|---|---|---|
| LLM-evaluated conversation | Whether the agent behaved correctly in dialogue | Instruction following, professionalism |
| Acoustic | Raw signal quality of audio | Jitter, shimmer, pitch variance |
| AI voice quality | Perceived naturalness and consistency | MOS-style scores, consistency signals |
| Custom | Product-specific requirements | boolean, enum, number_range, text, rating |
Custom metrics are useful when a business rule matters to your workflow and should be scored on every run.
Metric types
There are two core parts in metrics:
- Single metric
- Categorisation labels
Single-metric
Use a single metric when you want one direct score or pass/fail signal for a specific behavior.
Categorisation labels
Use categorisation labels when you want to break performance into structured buckets for easier analysis and reporting.
Next: Evaluator and Evaluation Suite
Community & contact
- Found a bug or have a feature request? Open a GitHub issue.
- Join our Discord for faster replies!