Evaluate results of churn prediction

Published December 06, 2021

Modules

Difficulty

Selected Clients

---

Industry

Automotive

Finance

Books

Construction

Cosmetics

Electronics

Fashion & Apparel

FMCG

Home & Furniture

Jewelry

Retail

Telco

Media

Multi

Travel

You can create a dashboard to monitor churn predictions.

The scenario in this example is a continuation of Predict churn, but with some tweaks it can be used for other predictions.

Prerequisites

• The prediction from Predict churn is ready.
• You are familiar with expressions, metrics, and dashboards.

Process

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

1. Create metric to calculate the conversion of customers who were at different risks of churn.
2. Create dashboard.

Create metric

In this stage, you create two metrics to calculate the conversion of customers who were at different risks of churn.

1. Create new metric.
2. As the metric type, select Formula metric.
3. As the first node, add a product.buy event:
   1. Set the aggregator to Count.
   2. Set Occurrence type to All.
   3. Add a contact filter with the following event conditions:
      • Action: snr.prediction.score
      • modelId equal to the model ID of your prediction.
        
        Tip: The model ID can be copied from the menu in the Prediction list.
      • score_label equal to Low

      

4. As the second node, add a snr.prediction.score event:
   1. Set the aggregator to Count.
   2. Set Occurrence type to All.
   3. Set the following conditions:
      • modelId equal to the model ID of your prediction.
      • score_label equal to Low
5. Set the mathematical operator between the nodes to division.
6. Encase the nodes in brackets.
7. Multiply the result by 100.

   

8. Save the metric.
   Result: This metric shows the percentage of customers who were at low risk of churn and made a purchase.
9. Repeat steps 1-8 to create a second metric, but replace the value of score_label with High.
   This will create a metric showing the percentage of customers who were at high risk of churn and made a purchase.

Create dashboard

In this stage, you create a dashboard that counts customers at risk of churn and shows the results of the metrics created in the previous stage. You can use this information to assess the effectiveness of churn prediction.

1. Go to Analytics > Dashboards > Add dashboard.
2. Add a segment to the dashboard:
   1. In the menu above the canvas, click .
   2. Click the widget that appears.
   3. In the panel on right, from the Segmentation drop-down list, select Create new.
      Result: The segmentation editor opens.
   4. Add the segment with customers whose probability of churn is high:
      1. Click Choose filter and select the snr.prediction.score event.
      2. Add the following condition: modelId equals the ID of the model of the prediction you want to analyze.
         
         Tip: The model ID can be copied from the menu in the Prediction list.
      3. Add the following condition: score_label parameter equals high.
      4. Enter a meaningful name for the segment.
   5. Add a new segment by clicking the plus icon next to the current segment.
   6. Repeat step d for the new segment, but change the condition to score_label parameter equals low.
   7. Click Save & Exit.
      Result: You are returned to the dashboard creator.
3. Add the metrics to the dashboard:
   1. Click .
   2. Click the widget that appears.
   3. From the Metric drop-down list, select one of the metrics that you created previously.
   4. On the Style tab, change Data format to %.
   5. Repeat steps 3a-3d to add the other metric.
4. Click Save dashboard.
   Result: The dashboard becomes available for viewing and sharing. You can change the date range when viewing the results.

Check the use case set up on the Synerise Demo workspace

You can check the dashboard configuration directly in Synerise Demo workspace.

If you don’t have access to the Synerise Demo workspace, please leave your contact details in this form, and our representative will contact you shortly.

Read more

• Dashboards
• Metrics
• Predictions