← Glossary

Inference

The process of using a trained AI model to make predictions on new, unlabeled data.

Inference is the operational phase where a trained model processes new returns to predict their categories. Unlike training (which happens once and is computationally expensive), inference is fast and happens continuously as new returns arrive. For real-time applications like return classification, inference must complete in milliseconds to fit into checkout or support workflows.

Related terms

  • AI Classifier — A machine learning model that automatically categorizes inputs—like return reasons or customer feedback—into predefined groups.
  • Machine Learning Model — A mathematical system trained on historical data to make predictions or classifications on new data.
  • Training Data — The labeled examples used to teach an AI model how to categorize returns correctly.
  • Confidence Score — A number between 0 and 1 indicating how certain the AI model is about its classification decision.

Want to see this in practice?

Start a trial