Events
Mar 31, 11 PM - Apr 2, 11 PM
The ultimate Microsoft Fabric, Power BI, SQL, and AI community-led event. March 31 to April 2, 2025.
Register todayThis browser is no longer supported.
Upgrade to Microsoft Edge to take advantage of the latest features, security updates, and technical support.
Transactional churn prediction helps predict if a customer no longer purchase your products or services in a given time window. Transaction churn is useful to find customers who are no longer purchasing products at any time during the defined churn window. To find customers who might cancel their preset ongoing purchases, we recommend using the Subscription churn model.
You must have business knowledge to understand what churn means for your business. For instance, a business with annual events can define their churn measured in years, while a business that caters to weekly sales measures churn in months. We support time-based churn definitions, meaning a customer is considered as churned after a period of no purchases.
For example, Contoso wants to know how engaged customers are for an email campaign dedicated to retention. Contoso's customers visit a storefront on a variable basis, most often around 3-4 times a month. Their transactions are irregular, making it difficult to determine when a customer stops purchasing the Contoso brand. Through the transaction churn model, Contoso can determine the likelihood that customers buy again. They can see the leading patterns that lead to customers leaving the brand, allowing them to adjust other strategies.
Go to Insights > Predictions.
On the Create tab, select Use model on the Customer churn model tile.
Select Transaction for the type of churn and then Get started.
Name this model and the Output table name to distinguish them from other models or tables.
Select Next.
Select Save draft at any time to save the prediction as a draft. The draft prediction displays in the My predictions tab.
Set the Prediction window. For example, predict the risk of churn for your customers over the next 90 days to align to your marketing retention efforts. Predicting churn risk for a longer or shorter period of time can make it more difficult to address the factors in your churn risk profile, but it depends on your specific business requirements.
Enter the number of days to define churn in the Churn definition field. For example, if a customer has no purchase in the last 30 days, they might be considered as churned for your business.
Select Next.
Select Add data for Customer transaction history.
Select the semantic activity type, SalesOrder or SalesOrderLine, that contains the transaction history information. If the activity isn't set up, select here and create it.
Under Activities, if the activity attributes were semantically mapped when the activity was created, choose the specific attributes or table you'd like the calculation to focus on. If semantic mapping didn't occur, select Edit and map your data.
Select Next and review the attributes required for this model.
Select Save.
Add more activities or select Next.
Select Add data for Customer activities.
Select the semantic activity type that contains the data you would like to use. If the activity isn't been set up, select here and create it.
Under Activities, if the activity attributes were semantically mapped when the activity was created, choose the specific attributes or table you'd like the calculation to focus on. If semantic mapping didn't occur, select Edit and map your data.
Select Next and review the attributes required for this model.
Select Save.
Select Next.
For the Data updates step, choose a frequency to retrain your model. This setting is important to update the accuracy of predictions as new data is ingested. Most businesses can retrain once per month and get a good accuracy for their prediction.
Select Next.
The Review and run step shows a summary of the configuration and provides a chance to make changes before you create the prediction.
Select Edit on any of the steps to review and make any changes.
If you're satisfied with your selections, select Save and run to start running the model. Select Done. The My predictions tab displays while the prediction is being created. The process can take several hours to complete depending on the amount of data used in the prediction.
Tip
There are statuses for tasks and processes. Most processes depend on other upstream processes, such as data sources and data profiling refreshes.
Select the status to open the Progress details pane and view the progress of the tasks. To cancel the job, select Cancel job at the bottom of the pane.
Under each task, you can select See details for more progress information, such as processing time, the last processing date, and any applicable errors and warnings associated with the task or process. Select the View system status at the bottom of the panel to see other processes in the system.
Go to Insights > Predictions.
In the My predictions tab, select the prediction you want to view.
There are three primary sections of data within the results page:
Training model performance: Grades A, B, or C indicate the performance of the prediction and can help you make the decision to use the results stored in the output table.
Grades are determined based on the following rules:
Likelihood to churn (number of customers): Groups of customers based on their predicted risk of churn. Optionally, create segments of customers with high churn risk. Such segments help to understand where your cutoff should be for segment membership.
Most influential factors: There are many factors that are taken into account when creating your prediction. Each of the factors has its importance calculated for the aggregated predictions a model creates. Use these factors to help validate your prediction results. Or use this information later to create segments that could help influence churn risk for customers.
Note
In the output table for this model, ChurnScore shows the predicted probability of churn and IsChurn is a binary label based on ChurnScore with 0.5 threshold. If this default threshold doesn't work for your scenario, create a new segment with your preferred threshold. Not all customers are necessarily active customers. Some of them may not have had any activity for a long time and are considered as churned already, based on you churn definition. Predicting the churn risk for customers who already churned isn't useful because they are not the audience of interest.
To view the churn score, go to Data > Tables and view the data tab for the output table you defined for this model.
Events
Mar 31, 11 PM - Apr 2, 11 PM
The ultimate Microsoft Fabric, Power BI, SQL, and AI community-led event. March 31 to April 2, 2025.
Register todayTraining
Module
Enrich data and predictions with Customer Insights - Data - Training
Enrich your data with other data sources and use AI to make predictions based on historical data.
Certification
Microsoft Certified: Azure Data Scientist Associate - Certifications
Manage data ingestion and preparation, model training and deployment, and machine learning solution monitoring with Python, Azure Machine Learning and MLflow.