Technical Documentation

Vidora

1 views 0

Along with many third party integrations, Zephr offers an integration with Vidora – a machine learning pipeline platform that allows companies to build and use complex machine learning models. Zephr customers use Vidora in a variety of ways, including propensity scoring for users.

When used within Zephr rules and decisions, information such as propensity scores allow you to provide more custom, data driven journeys for your end users, based upon the outcome that you think they will best respond to. This could include offering subscription discounts to users who seem less likely to subscribe, or offering additional content to loyal users.

Configuring Vidora with Zephr

To being your configuration, navigate to Settings > Plugins > Vidora within your Zephr Admin Console. If Vidora isn’t available within your Plugin list, email support@zephr.com.

Within the configuration page, you’ll need to enter your Vidora API Details. These include an API Key, and – if you’re using JWT tokens with Zephr – a JWT Claim for the Vidora User ID. Here you can also set your timeout length, which dictates when Zephr will timeout on the call to Vidora if a score is not being returned in a timely fashion.

Vidora Mapping in Zephr

Once you’ve added your API credentials, click into the Mapping tab to map your Vidora User Models and Vidora Score Banding to Zephr. In this section you’ll see two options – Vidora Model Naming and Vidora Score Banding.

Vidora Model Naming allows you to give easy to remember names to your Vidora Models, instead of using the 16-character model ID. This will mean it is easier to see which Vidora Model is being used within your Zephr Rules and Decision Points.

Vidora Score Banding allows you to set ‘bands’ for your Vidora Scores. Bands are helpful if you wish to group a series of Vidora scores as a user segment. The example below shows this. As you will see, the mapping provides three bands for scoring – a score between 0 and 0.25 bands users into the Low Propensity group, 0.25 – 0.75 is classified as Medium Propensity, and 0.75 – 1 is classified as High Propensity. This mapping is set as a JSON object, and should be done for each of the Vidora Models named above.

{
    "abcdefgh12345678": [
       {
          "from": 0,
          "to": 0.25
          "band": "Low Propensity"
       },
       {
          "from": 0.25,
          "to": 0.75,
          "band": "Medium Propensity"
       },
       {
          "from": 0.75,
          "to": 1,
          "band": "High Propensity"
       }
    ]
}

Once you’ve completed your mapping, click Save.

Zephr Rules Based Upon Vidora Scores

Although this will change in the future, at present there are no drag and drop options within the Zephr Rule Builder for checking a User’s Vidora Score. Instead, you will need to use the Custom Decision option, found under Advanced > Custom Decision within the Zephr Rules Builder Palette.

An example of this is the below code, which would be used within a Custom Decision, looking for Users who fall into the Low Propensity band mentioned in the mapping above:

return (blaize.secondPartyUserData['abcdefgh12345678'] == 'Low Propensity')

When replicating this in your own Custom Decision, use the following template and adapt the [model name] and [band name] as needed:

return (blaize.secondPartyUserData['[model name]'] == '[band name]')

If you are not using bands, and instead wish to make a Decision based upon whether a Vidora Score for a particular model is present, use the following, replacing [model name] as needed:

return !!blaize.secondPartyUserData['[model name]'];

Entering these Custom Decisions will result in Yes and No nodes branching off of the decision, as per other Decision Points. Once your Custom Decision is saved, continue your rule as usual. When the Rule is complete, save and publish your Rule as normal.