What is an AI field?
A CRM field a model fills in from a prompt instead of a person typing it, refreshed as the record changes.
AI field — An AI field is a CRM field whose value is generated by a model from a prompt over the record's data — a one-line summary, an industry classification, a risk flag — and refreshed when the record changes. It adds structured information without manual entry.
Why it matters
A normal CRM field holds whatever a person typed into it. An AI field holds whatever a model produced by running a prompt over the rest of the record: a one-line summary of the account, an industry classification inferred from the company name and description, a risk flag based on recent activity. Because the model can rerun the prompt whenever the underlying data changes, the field can stay current without anyone manually re-entering it.
The distinction that matters is between fields that classify or summarise, which tolerate being approximately right, and fields that state a fact, which do not. An AI-generated industry tag that is occasionally imprecise is still useful for rough segmentation. An AI-generated "annual revenue" presented the same way as a manually verified number is a guess wearing the appearance of data, and a rep who treats it as fact in a proposal is trusting something nobody checked. An AI field should be visually distinct from a manually entered one, so nobody mistakes an inference for a verified fact.
How an AI field is generated
- 1
A prompt is defined for the field
What to generate and from which parts of the record.
- 2
The model runs the prompt over the record
Producing a summary, classification or flag.
- 3
The value is stored on the field
Marked as AI-generated, distinct from manually entered fields.
- 4
The record changes
New activity, a new note, an updated stage.
- 5
The field is refreshed
The prompt reruns so the value reflects the current record.
The mistake to watch for
Questions
- How is an AI field different from a custom field?
- A custom field is defined by an admin but filled in by a person or an integration. An AI field is filled in by a model running a prompt over the record, and can refresh itself automatically as the record changes.
- Can an AI field be edited by a person?
- It should be. A person who spots a wrong inference needs to be able to correct it, and that correction should be visible as different from the model's own output.
- Should an AI field be trusted the same as a manually entered one?
- No, not for facts that must be exact. It should be labeled as AI-generated so a rep knows it is an inference, not a verified entry.