Using AI suggestions to categorize incidents
Starting in 9.2, you can use artificial intelligence (AI) suggestions to populate the incident type and incident category fields in the Tickets application. The multi-class classifier (MCC) model analyzes the incident summary and details to provide categorization recommendations.
Before you begin
Before you can use AI suggestions, an administrator must complete the following tasks:
- Ensure that the AI service is available and running
- Configure the MCC model in the AI configuration application
- Define training and inference filters to determine which incidents display suggestions
- Select sufficient historical incidents for training (minimum 20 incidents per category)
- Train the model with historical incident data and verify model accuracy
- Enable the incident type and incident category fields in the Tickets application user interface
Verify with your administrator that AI suggestions are enabled for your incidents. Only incidents that meet the inference filter criteria display AI suggestions.
About this task
AI suggestions are displayed when you enter text in the Summary or Details field and keep the Incident Type or Incident Category field blank. The MCC model provides multiple suggestions that are ranked by confidence level. Manually review and select the suggestion that best describes the incident.
Procedure
- In the side navigation menu, select application to create a new incident or open an existing incident.
-
In the Summary field, enter a brief description of the incident.
You can also enter detailed information in the Details field. The MCC model uses text from either field, or both fields, to generate suggestions.
-
Keep the Incident Type field, the Incident
Category field, or both fields blank.
If you populate one of these fields, AI suggestions are displayed only for the blank field.
-
Review the AI suggestions that are displayed for the blank field.
The suggestions are ranked by confidence level, with the most likely match displayed first. Each suggestion includes a confidence score that indicates how certain the model is about the recommendation.
- Optional:
Review the confidence score for each suggestion.
Each suggestion includes a confidence score that indicates how certain the model is about the recommendation. Higher scores suggest stronger patterns in the training data. Lower scores might indicate ambiguous descriptions or insufficient training data for similar incidents. Consider confidence scores when selecting suggestions, especially for critical incidents.
-
Click the suggestion that best describes the incident.
The selected value is populated in the field. The suggestion widget is removed after you make a selection.
- Optional:
If none of the suggestions accurately describe the incident, manually enter or select the
appropriate value in the field.
You can ignore the AI suggestions and manually categorize the incident at any time. Your manual selections help improve the model's accuracy over time.
- Complete the remaining fields in the incident record and save the record.
Results
The incident is categorized with the selected incident type and incident category. The categorization data is used to identify trends, support regulatory compliance, and improve future AI suggestions.
What to do next
You can change the incident type or incident category at any time by editing the incident record. If you change these fields after you accept an AI suggestion, the model learns from your correction and improves its future recommendations.