Predict churn

Published December 06, 2021
Selected Clients
carrefour logo

Predicting likelihood to churn has been always considered one of the most important problems for Marketers. You can do it now by using the Prediction module!

Churn prediction
Churn prediction



In this use case, you will go through the following steps:

  1. Create prediction target based on aggregate.
  2. Create prediction.

Create prediction target

In this stage, create an aggregate and use it in an expression that will serve as the target for the prediction model.

  1. Go to Analytics iconAnalytics > Aggregates > New aggregate.
  2. Enter the name of the aggregate.
  3. As the aggregating function, select Count.
  4. As the event to be aggregated, select transaction.charge.
    Tip: Events may have different labels between workspace, but you can always find them by their action name (in this step, it’s transaction.charge).
  5. Using the date picker in the lower-right corner, set the time range to Relative time range > Custom > last 30 days.
  6. Save the aggregate.
  7. Go to Analytics icon Analytics > Expressions > New expression.
  8. Enter the name of the expression.
  9. From the Expressions for dropdown list, select Attribute.
    Predictions work only with attribute expressions.
  10. In the formula creator, click the Select node and from the drop-down list select Function > Greater.
  11. As the first argument, select the aggregate you created earlier.
    Aggregates are added as customer attributes.
  12. As the second argument, select Constant with a value of 0.
  13. Save the expression.

Create prediction

  1. Go to Prediction icon in left menu Predictions > New prediction.
  2. In the upper-right corner, enter a name for the prediction.
  3. In the Prediction type section, click Define.
  4. Select Classification and click Apply

Create and select audience

In this section you decide which segment of the customers should be taken into account while making a prediction. For every individual in the segment, Synerise produces a single prediction.

Selecting the audience, and specifically its size, is a tradeoff between reach, duration of the calculation, and costs of data points produced.

Segmentations can be very complex and the possibilities of building the conditions are practically unlimited. In this example, a simple segmentation will include customers who visited your website at least once in the last 30 days.

  1. In the Audience section, click Define.
    You can use existing segmentations. This example shows how to create a new one.
  2. Click Choose segmentation > Create new.
  3. Enter a segmentation name and click Next step.
  4. Click Choose filter.
  5. Select the page.visit event.
  6. Using the date picker in the lower-right corner, set the time range to Relative time range > Custom > last 30 days.
  7. Click Create segmentation.
    Result: The segmentation is saved as the audience of the prediction and also becomes available in the Analytics module for other uses.
  8. Click Apply.

Select prediction target

  1. In the What would you like to predict? section, click Define.
  2. Click Select expression and select the expression created earlier.
  3. Click Apply.

Select inputs

In this section, you set up input features based on which the prediction model will be trained.

It is possible to select feature inputs manually, but we recommend using the automatic selection, as explained below. Our algorithms evaluate feature relevance in context of the prediction target and are, in most cases, more effective than manual selection.

  1. In the Model inputs section, click Define.
  2. Click Add feature > Automatically.
    Result: The list is populated with input features.
  3. Click Apply.

Configure additional settings

The additional settings define how often re-calculations are made and the content of events produced by the prediction.

  1. In the Settings section, click Define.
  2. From the How many days in advance do you want to make a prediction list, select 30 days.
    The time must correspond to the time range selected earlier in the prediction target.
  3. For the purposes of this example, the Set up recurring prediction calculation check box is cleared (this is the default value)
  4. In How would you like to display results, select 2-point scale.
    The algorithm detects the importance of a prediction.
  5. As the number of features to show as event parameters, set 5 by using the slider.
    The parameters will be included in a snr.prediction.score event that can be used, for example, in Analytics and Automation.
  6. Enter a user-friendly name for the prediction score.
    The name is shown as the value of the scoreName parameter in the snr.prediction.score event.
  7. Click Apply.
  8. To finish and calculate the prediction, click Save & Calculate.

The prediction results are saved as snr.prediction.score events in customer profiles.

What’s next

You can use the prediction results in your work, for example to Create an automation to reduce churn or Evaluate results of churn prediction.

A more advanced example of using a segmentation created from a churn prediction is described in Promote discounted items to customers at risk of churn.

Read more


We are sorry to hear that

Thank you for helping improve out documentation. If you need help or have any questions, please consider contacting support.



Thank you for helping improve out documentation. If you need help or have any questions, please consider contacting support.

Close modal icon Placeholder alt for modal to satisfy link checker