Get started with Agent Intelligence

Complete these setup and initial configuration steps to start using Agent Intelligence.

Before you begin

Role required: admin or ml_admin.

About this task

Agent Intelligence involves training a machine-learning algorithm to make predictions based on your past record data. The training process requires sending record data to a training service in the nearest data center and is therefore unavailable to customers who have data sovereignty requirements.

Procedure

  1. Request Agent Intelligence on a sub-production instance.
  2. Request Agent Intelligence reports on a sub-production instance.
  3. Export the production records for which you want Agent Intelligence solutions to provide predictions to a sub-production instance. For example, export 12 months of incident records to a sub-production instance.
  4. Review the default solution definition records to determine if the filter, input fields, and output field are sufficient to predict your incident or task records. If necessary, create a solution definition for each record set you want to predict on the sub-production instance.
  5. Train the solution definition records you want to test on the sub-production instance.
  6. Enable the Default Task Based Prediction business rule on the sub-production instance. If you have created custom solution definition records, update the business rule with the solution definitions you want to enable.
  7. Test the solution predictions by either creating test records or importing more records from production.
  8. Review the prediction reports to determine the accuracy and coverage of your solution and individual classes.
  9. If necessary, update the solution definition filter to include more or different training records.
  10. Retrain and retest any updated solution definition records.
  11. When you are satisfied with the accuracy and coverage of your solutions, Request Agent Intelligence on your production instance.
  12. Recreate any custom solution definition records and business rules to your production instance, and retrain the solutions.