The way AI models work is that they make predictions on how likely it is that a certain intent is contained in a message.

Each prediction is composed by 2 elements:

  • The object of the prediction: in Jatana these are Macro Intents or a Custom Intents.
  • The confidence score: a score from 0 to 100 that represents how confident the AI is about the correctness of its prediction.

The confidence threshold acts as a checkpoint to make sure that the automation is triggered only when the confidence score is above the confidence threshold that you have setup for a specific automation.

In short:

If confidence score < threshold => the automation doesn't run
if confidence score > threshold => the automation run

NOTE: we recommend starting with a threshold of 80% and gradually reduce over time.

Did this answer your question?