Zapier connector

Zapier is a tool that connects between various of your daily apps to create automated workflows based on triggers and actions. With the Peltarion plugin for Zapier, you can easily integrate your deployed Peltarion AI models with your Zaps.


To use AI in Zapier, you first need to have an AI model that does something useful for you. If you don’t know where to start, have a look at our tutorials to find inspiration.

Once you have a trained model, you need to create a Standard deployment from the deployment view. The deployment view will show you the information you need to use the model:

  • All the names (and shapes) of the input and output features of your model

  • The URL and deployment token that will allow Zapier to access your Peltarion model

  • The status of the deployment: make sure it is enabled, or queries will be denied

How to use Peltarion with Zapier

Once you have a deployed model on the Peltarion Platform, head to Zapier to start automating things.

  1. Create a new Zap.

  2. Set up the trigger: that’s the event from one of your connected apps that will initiate the workflow.

  3. In the Add app event box, use Search apps…​ to look for Peltarion. The Peltarion connectore will appear in the results and you can click on it to add it as the app event.

  4. As Action event, select Prediction.

  5. Click Sign in and enter the URL and deployment token from your Peltarion model.
    Zapier calls it an account, but this info only allows to use the specific model that you deployed, this doesn’t give access to your full Peltarion account.

  6. It’s time to Set up action: pick the data from your connected app that should be sent for every input feature that the AI model expects.

  7. When you’re done, Zapier will let you Test the action and continue.
    You will be able to add more triggers and actions to create changes in other connected apps.

API response

[output feature name] is the name of the output feature (as named on the Deployment view on the platform).

  • [output feature name]: The class with the highest probability. That is, the predicted class.

  • [output feature name] + Class probabilities: Probability of each class.

  • [output feature name] + Conviction: A value on how clear the prediction is.
    Math: Conviction = ln (1st candidate / 2nd candidate)
    If the two top predictions are very close, the conviction will be low. That is, the model doesn’t give a clear prediction.
    Usually a value between 0.7-1 is a good threshold, which means that the probability for the highest probability class is 2-3 times higher than the runner-up.
    Example: If the probability for the highest probability is 0.6 and the probability for the runner-up class is 0.3 then the conviction value will be: ln(2/1) = 0.693 ~= .7

In a car damage classification problem with key = "damageType" the response would look like this:

    "damageType": "bumper",
    "damageType Conviction": 0.5465,
    "damageType Class Probabillities": {"bumber": 0.4561, "windscreen": 0.2154, ...}

Working Zap examples

Check the examples on the connector’s page to see how complete Zaps look like and get inspired.

Was this page helpful?