Company logoOpen Ticket AI

Classification AI tasks

Zero-shot classification — sync values, write descriptions, set the workflow.

Classification AI tasks

A classification Agent sets a single ticket attribute — queue, priority, type, service, or a dropdown DynamicField. Classification is zero-shot: sync target values from the ticket system, write short descriptions, and the local LLM picks the match.

Studio creates one Agent kind per target. Native targets are fixed; only DynamicField classification requires specifying a field name.

Classification AI task detail

Create a task

  1. Open AI Tasks.
  2. Click + New Agent and select:
    • Queue classification
    • Priority classification
    • Type classification
    • Service classification
    • Dynamic field classification
    • or Try an example / Beispiel ausprobieren for a demo (department, municipal queues, administration priority, or university queues)
  3. Studio opens the Agent detail page (Inactive by default).

Target and LLM

Under AI Task settings:

  1. Prediction target — fixed for native targets. For Dynamic field classification, enter the exact DynamicField name.
  2. LLM size — Studio recommends a size; the exact model name appears below.

Sync values and describe them

  1. Click Sync values from ticket system to pull exact names without overwriting existing descriptions.
  2. Select a value and add a short AI description in your ticket language.
  3. Don’t predict excludes values the AI must never choose (e.g. raw inbox queues).
  4. Remove deletes a value from the task (not the ticket system).

Accurate descriptions ensure reliable classification. Names must match the ticket system exactly.

Workflow settings

Under Workflow settings, configure execution triggers and low-confidence handling. Save with Save workflow settings.

SettingMeaning
Only when state isOptional: process only tickets in this state (e.g. new)
Only when … isOptional: process only when the target attribute matches a specific value
Also run when … is emptyRun when the attribute is empty
StrictnessMinimum AI confidence: Lax (60%), Normal (80%), or Strict (90%)
Set a fallback value?Use a default replacement when AI confidence is too low
Fallback valueSelect the replacement value (visible when fallback is enabled)
Flag for human review?Mark low-confidence tickets for agents via HumanReviewNeeded DynamicFields

Set Strictness first, then define low-confidence handling. Without a fallback, human review stays enabled.

Custom context goes in Extra information (model settings). Under Advanced, optionally enable Write article note using placeholders like {result_value} and {confidence}.

Tickets with an existing confidence DynamicField for this attribute are skipped.

Activate

  1. Set at least one eligibility filter (state, attribute, or empty state).
  2. Resolve missing values: Predict cannot be enabled for values missing from the ticket system.
  3. Activate — Runtime begins executing the task.
  4. Deactivate stops execution while preserving configuration.

Archiving hides the task from the list without deleting it.

Tips

  • Use one active Agent per attribute (e.g. one queue Agent).
  • Re-run Sync after changing ticket system values.
  • Monitor prediction accuracy and fallbacks on the Dashboard.