Showing results for 
Show  only  | Search instead for 
Did you mean: 
This is a verified product documentation article. For case-based resolutions articles, please reference the Knowledge Base section of Invoca Community.

Your Signal AI insights improve by learning from the data in your own phone calls. When you provide your Signal AI with training data, the predictive power of that Signal is reflected in your Signal AI accuracy score. 

Here's how to view the accuracy score of your Signal AI insights:

  1. Login to your Invoca account. From your network-level dashboard, select Signal in the gray menu ribbon.
  2. In the list below, select a Signal with the type "AI (Custom)". 

In the Live Predictive Model tile, your Signal AI is measured with an accuracy range instead of a single score. We show that accuracy range to demonstrate how well your Signal AI is categorizing calls within the training data you’ve provided, and how well we can expect it to perform with incoming, unclassified calls once it’s been deployed in your Invoca account. 

  • The percentages at either end of your accuracy range show how accurately your Signal AI model has predicted different outcomes from your phone calls in your training data. 
  • The size of that accuracy range correlates with how confident we are in the precision of those predictions going forward, with a narrower range demonstrating better precision.

By showing both measurements, we can also better represent the improvement of your Signal AI predictions when you provide more training data. For example, imagine you create a first version of your Signal AI using 500 phone calls for training data. Your Signal AI learns from 400 of those calls and sets aside the other 100 to test the pattern it finds. Of those 100 calls, it accurately predicts whether all 100 of them were sales, representing a 100% accuracy score. You then add another 750 phone calls to update your Signal AI to a new version — but in that next data sample, it only predicts 700 out of 750 calls correctly.

Your Signal AI didn’t get less accurate from version to version — in fact, it used the data you provided to recognize stronger patterns in your data. The training data in your first sample may have been more uniform, or easier to predict than your second sample, which resulted in a lower accuracy score, but more precise predictions overall.

You can learn more about the data science behind your Signal AI accuracy, including how the accuracy scores at both ends of your accuracy range are calculated, in this blog post


Need more help?

Don't see what you are looking for? You can ask the Community or contact support.