Efficiently train AI Signals using Labels, multi-labeling, and progressive training rounds — and upgrade a legacy AI Signal to Signal AI Studio without losing history.This guide covers efficient training of AI Signals — Labels and AI Models, multi-labeling, and progressive training rounds. Your CSM (Customer Success Manager) is available to help.
Labels and AI Models
An AI Model can have 1-9 Labels, each representing something to monitor (for example, Scheduled Appointments, New Leads). The Signal Definition Form evaluation output helps determine a label’s name and definition. Labels are created prior to Round 1 and can’t be changed later. Each Label can be individually activated as a Signal — not all must be. Labels don’t count against your Signal allocation until activated. Multi-labeling: train multiple Labels simultaneously to save time and build more robust, inter-related models. Contextually related Labels don’t have to be trained together — it’s optional.Labeling calls (Round 0 — cold start)
Search relevant calls via Thematic Search using broad phrase sets. Search terms don’t change after each call unless you type a new phrase. You must label a minimum of 20 true and 20 false examples for one Label, and 10 true and 10 false for each additional Label, to unlock Train AI Model. Thematic Search focuses on themes, intent, and context rather than exact keywords. It works best with expanded, detailed phrases (for example, “I’m looking to book an appointment”). Use multiple search terms for true and false examples, and update the phrase if you’re not finding enough relevant calls (for example, 10 falses in a row). View Call Details jumps straight to the full call record. When labeling, enter a search term, click Search, then View Call Details. When verifying labels, View Call Details appears directly — no thematic search is needed there, since calls surface automatically.Active learning (Round 1+)
The system surfaces calls automatically from your Round 0 search terms. AI predicts True, False, or Not Sure per Label, and you verify accuracy. To proceed, verify 40 total calls combined for at least one Label (for example, 10 true/30 false, 25 true/15 false, or 20 true/20 false) — we recommend a minimum of 5 true examples (for example, 5 true/35 false). This enables Train AI Model to process the round.Signal accuracy and suggestions (Round 3+)
Real-time predicted accuracy shows after verifying 5 or more calls per Label. The goal is a minimum of 80% accuracy within a range of 10 points or less (for example, 80-90% or 85-95%). Once the threshold is met, a prompt suggests “This Label may be ready for Signal Activation,” with a “Signal Activated” badge and convenience link once activated. Activate or edit Signals directly from the Manage Signals page. Selecting a Label Round shows relevant details, such as predicted accuracy. If more training happens after activation, you must manually select the latest round when editing the Signal.Upgrade Legacy Signals to Signal AI Studio Signals
Upgrade older AI Legacy signals to newer AI Studio signals while keeping historical data. This applies only to English-only signals — not foreign language signals. Before you begin: you need an existing AI Legacy signal, plus an equivalent AI Studio model already created and saved (for example, “Appointment Scheduled Legacy” upgrading to “Appointment Scheduled Studio”).- On the Manage Signals page, find the Legacy signal (shows “AI Legacy” in the Type column) and click its name.
- Click Upgrade Signal (only shown for AI Legacy signals). A pop-up explains the changes — click Upgrade. The current Legacy Model is replaced by the selected Studio model for all new and future calls; historical calls stay linked to the old configuration.
- On the config page (the AI Studio signal form), use the Model dropdown to select the equivalent Studio model — we recommend one performing the same function. The original signal name and info carry over automatically. Click Save.