What does Custom Intent accuracy mean?

We calculate the accuracy of our Custom Intent to help you set expectations around how the Intent will perform when tagging your messages.

In simple terms, the accuracy score represents what proportion of the messages the model is likely to classify correctly.

The accuracy score can also indicate that the model is performing poorly and needs more training.

The accuracy scale

Each Custom Intent will display the accuracy score as of the last training round. This 

How to interpret the score:

  • Bad - Intent is likely to misclassify most of the results, further training is essential;
  • Average - Intent is likely to misclassify results, further training is recommended;
  • Good - Intent is likely to classify most results correctly, further training is optional.

You can learn more about Custom Intent training here.

How is the score calculated?

We use an industry-standard technique called cross-validation for assessing the accuracy of our AI models.