Managing AI fields
Everything you need day to day is on the AI Fields list. One row per field, with the columns it produces, the table it reads, whether it is active, when it last ran and its status. Search by name, or filter by table and status.
Click the arrow at the start of a row to see the columns that field adds, with their type and whether they are visible.
Row actions
| Action | What it does |
|---|---|
| Active toggle | Pauses and resumes the field. A paused (Suspended) field is skipped by the scheduled pipeline and its columns are removed from your table; the answers already produced are kept, so resuming does not pay for them again. Unavailable on a Draft, and while a run is in progress. |
| Run now | Starts a run immediately, without waiting for the scheduled pipeline. |
| Edit | Reopens the wizard on the existing configuration. |
| Delete | Removes the field, its columns and the answers it produced. |
Each run can buy answers, so a field can be refreshed by hand once an hour. When the button is unavailable, hover it: it tells you why, either a run already in progress or how long is left.
Editing a field
Editing reopens the same five steps. What happens when you save depends on what you changed.
Changing the filter or the run logic is free. They decide which records get answered, not what the answer is, so every answer you already have stays valid.
Answers belong to the question that produced them. Edit the question, the columns it reads, or the outputs, and every answer already produced stops counting:
- the columns read empty until a run refills them;
- that run buys every answer again, at full price;
- activation locks until a new test passes.
The confirmation dialog says so before you save. If anything depends on those columns — audiences, signals, AI models — they see empty values in the meantime, so plan the edit rather than discovering it.
There is one piece of good news here. The old answers are not deleted, they are set aside. Change the question back to exactly what it was and they become live again, at no cost.
When a run fails
The field goes to Error and the reason is shown on it. The next scheduled run retries on its own, so a transient problem clears itself.
A run that partly fails still completes. Records that could not be answered are counted, and their columns stay empty for those records. The run history shows how many records were asked, how many were answered, how many were reused from earlier runs, and how many failed.
Records missing a value in any column the question reads are skipped rather than half-asked. If a whole field produces nothing, that is usually why: check how populated those columns are in the column picker.
Two fields on one table
A table can carry several AI fields at once, and they do not interfere. Each adds its own columns.
One rule: a question cannot read another AI field's output. Those columns appear in the picker with a lock. If you need an answer that builds on another answer, put both questions in one field, or write a Field Builder calculated field on top of the output.
See also
- Creating an AI field
- External Data Enrichment — the sibling feature, for values that exist in someone else's database.