Skip to main content

What labels are

Labels are the possible outcomes of a conversation — payment promise, refusal, unavailable, voicemail, etc. They let you classify each call and later analyze the campaign’s results in reports.
The initial set of labels is defined together with Inceptia when the use case is set up, but it isn’t fixed: from the Labels tab you can create new ones and edit the existing ones, as long as your user has permissions on labels. If you don’t see the New button and can’t open a label to edit it, ask your Customer Admin for that permission; if they don’t have the option to grant it, the request goes on to Customer Success.

Where to see them

The bot’s Labels tab shows the full table of labels available for that bot. At the top there’s a search box by name and filters by Type of contact and Groups: Table of the bot's labels with a search box, filters by type of contact and group, and the New button
A label being Retries doesn’t mean the case will be called indefinitely: how many times it’s retried is set by the campaign’s Retries parameter, not by the label. That limit is counted per phone number, not per case — a case with several enabled phone numbers can receive more calls than the configured number. See Retries.

How they’re assigned

Labels are assigned by the Classifier agent, which runs post-conversation: once the call ends, it analyzes the full context of what was said and assigns the label that applies. On the canvas it’s the red agent.
For classification to work, the agent’s instructions have to reference the labels it can assign and explain when each one applies. A label that isn’t referenced in the prompt is never assigned — it’s one of the most common mistakes when configuring a bot.

Referencing labels in the prompt

You don’t type labels by hand: you insert them from the Classifier agent’s Instructions view, where each reference becomes a fixed piece inside the text. Classifier agent's Instructions view with the labels panel to the right of the field

The available labels panel

To the right of the instructions field there’s a panel with the labels available for the bot, grouped by group and sorted alphabetically, with a search box by name.

Inserting a label

There are two ways, and you can reference the same label as many times as you need:
  • Dragging it from the panel to the point in the text where you want it.
  • Typing /label in the editor, which opens a search box by name.

How the reference behaves

The reference looks like a chip with the label’s name and behaves like a single character: the cursor doesn’t go inside it, a single press of delete removes it whole, and it has an icon to remove it with the mouse. Hovering over it — both over the chip in the text and over the label in the panel — shows its Id, which is still needed to configure the agent’s post_conversation_action tool (see Agent tools).
Chip for the Recurring Call label showing result_id = 13 on hover
The Id never appears written in the text, but it’s what reaches the bot: what the agent receives is result_id = N. If you copy the field’s content into an external editor you’ll see that format instead of the label’s name.
If the prompt references an Id that no longer exists in the catalog, it isn’t shown as a valid chip: it stays flagged in the text so you can fix it.

Creating or editing a label from the prompt

The panel has a New button and, on each label in the list, an icon to edit it. Both open the Labels tab in a new browser tab, so you don’t have to interrupt writing the prompt to look up an Id or fix a name. The form lets you define the name, the group it belongs to, the contact type and what happens to the case when the bot uses it — the three options in the Result column.
With the field empty, the editor shows a sample template in gray so you have somewhere to start. Clicking anywhere in the field places the cursor and replaces the template with whatever you type.

Example instruction for the classifier

A classification prompt starts by defining the role, states explicitly which tool it has to call, and sets the rules before getting into the criteria for each result:
From there on come the criteria for each result, and that’s where label references come in: in the editor you see them as chips with the name, while what the bot receives is result_id = N.
It’s worth including an explicit rule that classification can only be based on what was literally said in the conversation — that the model shouldn’t infer commitments that weren’t mentioned, and that silence never amounts to a commitment. It helps keep the classifier from “hallucinating” results.