Skip to content

AI Predict Value – letting AI choose a value of a field

The command "AI Predict Value" lets an AI assistant choose one of the allowed values of a field – for example a category, a priority or a responsible team. This page is for administrators who classify records automatically.

1. At a glance
Purpose Pick a value from the allowed values of a field, based on a description
Typical use Categorise incoming tickets; suggest a priority; assign a team
Where in the editor Toolbar menu "AI" → "AI Predict Value"
Needs a user interface No
Needs a current record Yes
Related AI Assist, Business rules

The allowed values come from validation rules with valid values on the field (see Business rules). Without such a rule there is nothing to choose from.

The command only fills variables. Write the result into the field with Update Entity Record afterwards.

2. Settings

All settings are on the tab "General", below "Display Name", "Description" and "Execution Condition" (see Actions).

Setting Meaning Default / Required
"Assistant" The AI assistant that makes the choice Required
"Target Field" The field whose allowed values are offered. The list shows the fields of the entity the command works on that have a validation rule. Required
"Prompt" Expression for what the choice is based on – typically the describing fields of the record Required
"Predicted Value Variable" Receives the ID (key) of the chosen value, not its display text. Empty when nothing fits. Required
"Explaination Variable" Receives the reasoning of the assistant – useful to show the user why a value was chosen Optional

The command also has a setting that forces the assistant to pick one of the values even when none fits well. It is off by default and is not offered in the editor; only the AI assistant that edits the configuration can set it (see Introduction). When it is off, the assistant may answer that no value fits.

3. Behaviour

  1. If no assistant is set, it no longer exists, no target field is set or there is no current record, the command is skipped.
  2. nexxcite collects the valid values of all active validation rules of the entity on the target field. Rules whose condition does not hold for the current record contribute no values.
  3. The prompt, the values with their descriptions and the instructions of the assistant are sent to the AI provider. The assistant sees nothing else.
  4. The ID of the chosen value is written into "Predicted Value Variable", the reasoning into "Explaination Variable".

If the request fails, "Predicted Value Variable" is empty and "Explaination Variable" receives the reason. There is no separate error variable; the action continues.

4. Examples

Categorise a new ticket. The field "Category" has a validation rule with the valid values Hardware, Software, Network and Access, each with a short description.

  1. "AI Predict Value" with "Target Field" Category, "Prompt" = subject and description of the ticket, "Predicted Value Variable" CategoryId, "Explaination Variable" Reason.
  2. An Update Entity Record command sets "Category" to CategoryId, with the "Execution Condition" that CategoryId is not empty.

Run this in a business rule after the ticket was created, so that users see the category when they open the ticket.

Good descriptions of the valid values improve the choice considerably – write them for the AI as you would for a new colleague.

5. Limitations

  • Only fields with validation rules with valid values can be used.
  • An empty result does not tell whether nothing fitted or the request failed – read "Explaination Variable".
  • The prompt and the valid values are sent to the AI provider. AI choices can be wrong; let people correct them.