Skip to main content
Define your AI Signals, label a sample of calls, train the model across rounds, then activate and validate the resulting Signal.
Defining your AI Signals helps you determine which business outcomes you want to identify, streamline, or resolve using insights and data from Signal AI Studio. Clearly identifying the AI Signals your business needs — such as appointment lead, appointment booked, or new prospect — helps you define your desired outcomes. Your Invoca team works with you to identify the most essential indicators or phrases that occur during your calls, to provide the most useful data supporting growth opportunities in your market.

​
Step 1: Complete the Signal Definition Form

Your Invoca CSM (Customer Success Manager) can provide hands-on guidance during this process. We also recommend watching the “Signal AI: Signal Definition Form” course in the Invoca Academy for a more in-depth description of the form’s requirements. Completing the Signal Definition Form is an essential step — the information you provide helps identify the primary questions or subjects you want your data to reflect when using Signal AI Studio. We recommend reviewing the “Create Your Signal Strategy” resource in the Invoca Academy before filling out the form.

​
Step 2: Label your calls

  1. Create your labels based on the Signal definitions approved during the definition process.
  2. Label your calls by entering thematic search phrases to identify relevant calls, marking each as True or False.
  3. Train your AI model once you complete your first round of labeling — this teaches the AI to recognize patterns and start making predictions from your calls.

​
Step 3: Verify and train your labels

  1. Verify your labels by reviewing the AI predictions made for each label. Approve the ones that are accurate.
  2. Train the AI model to improve prediction accuracy in the next round. Predicted accuracy starts displaying in Round 3, and activation suggestions are made based on training progress.
  3. Verify your calls. At this point, thematic search is no longer used — calls are surfaced automatically, and AI actively predicts True or False. Click Train AI Model between each round.

​
Step 4: Activate and validate your labels

  1. Activate your label after a minimum of 3 rounds of training and reaching the predicted accuracy threshold.
  2. Validate and review how your Signals perform in production by spot-checking the accuracy of your previously established True/False cues.

​
Training your Signal in Signal AI Studio

Training splits into two phases: Labeling Calls and Verifying Labels. Labeling Calls phase
  • Use thematic search to enter terms that surface relevant calls.
  • For each call, determine which labels are True, False, or Not Sure, based on transcript analysis.
  • Label calls until you’ve reached a minimum of 20 true and 20 false examples for at least one label.
  • Label 10 true and 10 false examples for additional labels before progressing to the next round.
Verifying Labels phase
  • Review calls surfaced automatically based on your prior thematic search criteria.
  • AI starts making predictions of True, False, or Not Sure.
  • Confirm or correct each AI prediction based on transcript analysis — this is “verifying labels.”
Next steps: after reaching the required labeled/verified call count in each round, click Train the AI Model to apply what’s been learned and improve accuracy. This repeats until the label reaches suggested accuracy levels for activation as a Signal — around 80% predicted accuracy, with a range difference of 10% or less. Each training round takes about 1 hour. The number of rounds needed varies depending on your use case.

​
Activate and validate your Signals

After Round 3 and verifying a few calls, an accuracy bar displays real-time predicted accuracy for each label. Signal AI Studio indicates when it’s time to activate a label as a Signal based on at least 80% accuracy for each label, within a narrow range (for example, an under-10-point difference, such as 85-95%). We recommend verifying about 30 calls per round to confirm consistent accuracy before activation, and that predictions are consistently correct (correcting less than 10% of predictions).

​
Formally validating your Signals after activation

Your CSM can submit a request for Invoca’s Analytics Services team to initiate the formal validation process. Your team scores and audits calls using the worksheet provided.

​
Reviewing your last call

The Review Last Call feature lets you return to a previously labeled call, correct labeling mistakes, or confirm a call is marked accurately (True/False/Not Sure). This feature applies to all Signal AI Gold customers labeling calls in Signal AI Studio. Benefits:
  • Quickly return to a previous call and correct labeling mistakes.
  • Improves overall data quality and AI model accuracy.
To edit a call via Review Last Call: if you’ve finished labeling a call and selected Save & Next Call, you can return to it and edit incorrect labels.
  1. On the next call, select Review Last Call.
  2. Select the True/False/Not Sure labels you want to update.
  3. Select Save & Next Call to save your edits.

​
Archiving AI models

Review your models regularly to confirm all active labels are up to date and relevant. If a model is no longer relevant, hide or archive it using the Archive Models capability. A clean, concise backlog makes it easier to find specific AI models using the Search for AI Model search bar — enter a keyword to see all matching models in the list. You can filter to view All, Active, and Archived models. In Archived, you can restore a previously archived model. While viewing Active models, select the three-dot icon on a model to archive it. Keep in mind:
  • Any deployed AI Signals connected to an archived model stay active and functional.
  • Once a model is archived, further training in Studio is disabled until it’s manually restored.
  • New AI Signals can’t be created using an archived model.

​
Where to go next

Last modified on September 23, 2026