A threshold determines when the AI's prediction is confident enough for automatic action. When the model's confidence score falls below the threshold, the classification is sent to human agents for review. Setting thresholds is a tradeoff: lower thresholds automate more cases but allow more errors; higher thresholds improve accuracy but reduce automation benefits.
← Glossary
Threshold
The minimum confidence score required for the AI to auto-assign a classification versus flagging for human review.
Related terms
- Confidence Score — A number between 0 and 1 indicating how certain the AI model is about its classification decision.
- AI Classifier — A machine learning model that automatically categorizes inputs—like return reasons or customer feedback—into predefined groups.
- False Positive — When the AI incorrectly assigns a return reason that doesn't match the actual reason.
- False Negative — When the AI fails to identify the correct return reason and assigns an incorrect one.