Classification Actions

Classification Actions provide AI-powered suggestions for classifying data cards, such as assigning a ticket to the correct category, service, or cost center. This helps reduce manual effort and improves routing accuracy.

When triggered, the AI analyzes the configured context attributes (for example ticket subject or description) and compares them with the available classification options.

Classification Actions can suggest values for reference attributes and for static string attributes with a predefined list of values.

Classification Actions are configured by administrators in the M42 Intelligence administration interface and can be triggered manually by agents or automatically through workflows.

Managing Classification Actions

Administrators can manage classification actions from the Classification Actions tab.

The tab lists all configured actions with the following information:

• Unique name – Internal identifier for the configuration
• Description – Explanation of the action
• User title – Name shown to agents in the UI
• Template – Template where the action is available
• Context attributes – Attributes used as input for the AI
• Classification attribute – Attribute that receives the suggested classification
• Description attribute – Optional attribute used to improve classification accuracy
• Visibility – Determines whether the action is available to users

Administrators can add, edit, or delete classification action configurations from this view.

Creating a Classification Action

To create a classification action:

1. Open the Classification Actions tab.
2. Select Add.
3. Configure the action settings.
4. Save the configuration.

The configuration dialog includes the following fields:

• Unique name – Internal name used to identify the configuration.
• User title – Title displayed to users in the Agent UI.
• Description – Optional description explaining the action.
• Classification system prompt – System-defined prompt used for classification. This prompt cannot be edited.
• Additional admin prompt (optional) – Additional instructions to guide the AI classification.
• Visibility – Controls whether the action is visible to users.
• Template – Template where the classification action will be available.
• Context attributes – Attributes whose values are analyzed by the AI.
• Classification attribute receives the suggested value. Select a single-value reference attribute, which defines the available options, or a single-value static string attribute with a predefined list of values.
• Description attribute (optional) – Attribute containing descriptions of the classification options.

Description Attribute

The Description attribute is optional and can be used to improve classification accuracy.

When configured, the AI evaluates both:

• The name of the classification option.
• The description of the option.

If descriptions are not available for some options, the system classifies based on the option name only.

Only single-value String or Text attributes can be used as description attributes.

Visibility

The Visibility setting controls whether the classification action is available to users:

• Visible – The action can be used in the Agent UI.
• Hidden – The action remains configured but is not available to users.

Deleting a Classification Action

To delete a classification action, select the delete option from the action list and confirm the deletion. Deleted configurations cannot be restored.

Using Classification Suggestions in the Agent UI

When a Classification Action is available for a template, agents can receive AI-generated classification suggestions when viewing or editing a data card.

The suggestion appears in the classification attribute field (for example Service or Category) together with an AI suggestion indicator.

Reviewing the Suggestion

When a suggestion is generated, the UI displays:

• The suggested classification value.
• An AI suggestion label.
• Action buttons to Accept or Deny the suggestion.

Agents can review the suggestion before deciding whether to apply it.

For static string attributes, the suggested value appears in the language of the user.

Additional information about the suggestion may be shown in a tooltip, explaining how the AI arrived at the recommendation based on the ticket content.

Accepting or Rejecting a Suggestion

Agents have the following options:

• Accept – Applies the suggested value to the classification attribute.
• Deny – Rejects the suggestion without changing the attribute.

Agents can also manually select a different value from the attribute dropdown if the suggested value is not appropriate.

AI Badge

When a classification value is automatically set by AI (for example through workflows), an AI badge appears next to the attribute.

The badge indicates that the value was set by AI.

Hovering over the badge shows the message:

AI automatically set this value by analyzing the context.
AI suggestions can be edited. To get more information about AI assisted experience, please contact your administrator.

The badge also appears on a static string attribute that a workflow has filled in.

The badge is visible to all users who have permission to view the attribute.

If the attribute value is modified manually, the AI badge is automatically removed.

Permissions

Classification suggestions are generated only from entities and attributes the user has permission to access. This ensures that sensitive data is not exposed through AI suggestions.

Using Classification Actions in Workflows

Classification Actions can also be executed automatically in workflows using the AI Node. This allows the system to classify data cards without requiring manual interaction from agents.

Configuring Classification in a Workflow

In the workflow editor:

1. Add an AI Node to the workflow.
2. Select Classification as the node type.
3. Choose the Classification Action to execute.
4. Select the target attribute where the classification result will be written.

The selected attribute must be a reference attribute or a static string attribute. It typically corresponds to the classification attribute defined in the action configuration.

Limitations and Requirements

The following limitations apply to Classification Actions:

• Classification Actions are supported only in the Agent UI.
• The classification attribute must be a single-value reference attribute or a single-value static string attribute.
• A reference classification attribute can refer to up to three templates by default. The platform setting ai.classification.maxTemplates controls this limit (default: 3).
• A reference classification attribute that refers to all templates cannot be used.
• The optional description attribute must be a single-value String or Text attribute.
• Classification suggestions are generated only from entities and attributes that the user has permission to access.
• Classification Actions use the language model that is connected to M42 AI Core. Direct provider connections (M42 GenAI (legacy), Azure OpenAI and OpenAI) still work in 26.2 and are retired in 27.1. For the steps to move, see Moving AI provider configuration to AI Core.
• The number of available classification options may be limited depending on system configuration.
• By default, a classification action uses up to 100 options. The platform setting ai.classification.maxOptions controls this limit (default: 100). If the attribute has more options than the limit, the action fails with an error instead of using a partial list.
• To limit the options to relevant values, configure a Matrix42 Query Language (EQL) search filter on the reference attribute. The action then offers only the options that match the filter.
• If the attribute refers to more than one template, a filter can be set for each template.
• The EQL search filter is not applied to a category dropdown attribute that has a parent category attribute. For such an attribute, the options come from the category hierarchy and are still limited by ai.classification.maxOptions.
• The EQL search filter is available for reference attributes. For a static string attribute, keep the number of predefined values within the ai.classification.maxOptions limit, or raise the limit.

Setting names and their availability can change between versions.

Classification results are suggestions, and users can always review or modify the classification before saving the data card.