Use Retail channel churn model
Create a 360-degree view of your customers in a clear, intuitive way, focused at the customer level. Use Dynamics 365 Customer Insights to select and combine the most relevant, important data across multiple business lines into a comprehensive, cohesive customer view. Microsoft Cloud for Retail includes Retail channel churn model solution that lets you predict and understand customer churn. This solution lets you create a retail channel churn predictive model, and then use the model's predictions to prevent potential loss or churn.
Summary tab
The Summary tab shows at-a-glance information about customers' personal details, life moments, financial holdings, and credit and debit cards. This tab is your starting point to provide personalized experiences, reveal important opportunities, and improve customer satisfaction.
Retail channel churn predictive model
The AI-based churn predictive model, designed for omnichannel retail and built atop Customer Insights helps gain cross-channel insights into the chance of retail customer churn. Run your company data through this model, training it to improve its predictions and identify the factors that contribute to churn, at the customer level.
You can use three data entities to create and train your model: a customer entity, a session entity, and a transaction entity. The model uses inputs that you map to fields from these entities. When your model runs, it stores its churn predictions in an output entity, and provides explainability elements—factors with the most influence on churn risk predictions. It displays these factors along with their level of influence.
Prerequisites
Retail components are available within Microsoft Cloud for Retail in Microsoft Cloud Solution Center. For more information about deploying these components, go to Deploy Retail channel churn model in Microsoft Cloud for Retail.
At least Contributor permissions in Microsoft Dynamics 365 Customer Insights. For more information, go to User permissions.
An understanding of what churn means for your organization. A customer is considered to have churned if the customer's purchase value or volume drop below thresholds that you define.
Entities with fields that map to inputs for your retail churn prediction model:
Important
The deployment of the Retail channel churn model solution requires you to create these entities in Customer Insights and they should exist in your Customer Insights B2C environment. This step is required since these entities aren't part of Microsoft Cloud for Retail, which doesn't have a defined common data model because of the huge variability of retail enterprises. For more information, go to Deploy Retail channel churn model in Microsoft Cloud for Retail.
Customer data
A customer entity has fields—also called attributes—that's data about customers, but not their visits or their purchases. A churn prediction model has one required input and 10 optional inputs that you map to customer entity fields when you create your churn model. To prepare, make sure your customer entity has fields you can map to the model's inputs.
If your customer entity doesn't have all the attributes to be included in your churn model, you can map, match, and merge the missing attributes so that they're available as churn model inputs. For more information, go to Data unification overview.
If you can't find suitable attributes in your source data, you may be able to add data sources, and then map, match, and merge them.
Customer data inputs
Locate your customer entity and note which field you'll map to each input listed in the following table. You can't create a model without a field for each required input. Your model's predictions will be more accurate if you map fields to optional inputs, the more of them the better.
Tip
If you're new to this, consider copying the following table of inputs and then noting which field you'll map to each input.
Input to map | Required or Optional |
---|---|
Customer ID | Required |
Agreement to marketing activity | Optional |
Loyalty membership | Optional |
Occupation status | Optional |
Birth date | Optional |
Gender code | Optional |
Annual income | Optional |
Relationship duration | Optional |
Distance to nearest store | Optional |
Gender | Optional |
Customer relationship duration | Optional |
Session data
A session entity has fields that have data about customer visits, but not their purchases. The churn model has four required and five optional inputs for session data. Providing optional inputs will improve the accuracy of predictions.
If your session entity doesn't have all the attributes that you want to include in your churn model, you may be able map, match, and merge the missing attributes so that they're available as churn model inputs—if your source dataset includes suitable attributes to use. For more information, go to Data unification overview.
If you can't find suitable attributes in your source data, you may be able to add data sources, and then map, match, and merge them.
Session data inputs
Locate your session entity and note which field you'll map to each input listed in the following table. You can't create a model without a field for each required input. Your model's predictions will be more accurate if you map fields to optional inputs, the more of them the better.
Tip
If you're new to this, consider copying the following table of inputs and then noting which field you'll map to each input.
Input to map | Required or Optional |
---|---|
Session ID | Required |
Customer ID | Required |
Session timestamp | Required |
Session channel | Required |
Visit type | Optional |
Session purpose type | Optional |
Session duration | Optional |
Customer satisfaction with session | Optional |
Profile login | Optional |
Transaction data
Important
Because they directly reflect customer purchasing behavior, transactions are key for predicting customer churn—a customer whose transaction volume or value falls below a certain threshold has churned. You define what churn means for your business by setting those thresholds when you create your churn model.
A transaction entity has fields that have data about customer purchases and sessions. A retail churn predictive model has seven required inputs and six optional inputs that you map to transaction entity fields when you create your model. To prepare, make sure your transaction entity has fields you can map to the model's inputs. Each optional input that you map will improve the accuracy of your model's predictions.
If your transaction entity doesn't have all the attributes that you want to include in your churn model, you may be able map, match, and merge the missing attributes so that they're available as churn model inputs—if your source dataset includes suitable attributes to use. For more information, go to Data unification overview.
If you can't find suitable attributes in your source data, you may be able to add data sources, and then map, match, and merge them.
Transaction data inputs
Locate your transaction entity and note which field you'll map to each input listed in the following table. You can't create a model without a field for each required input. Your model's predictions will be more accurate if you map fields to optional inputs, the more of them the better.
Tip
If you're new to this, consider copying the following table of inputs and then noting which field you'll map to each input.
Input to map | Required or Optional |
---|---|
Transaction ID | Required |
Customer ID | Required |
Transaction timestamp | Required |
Transaction session ID | Required |
Transaction type | Required |
Transaction channel type | Required |
Transaction amount | Required |
Transaction location ID | Optional |
Product catalog ID | Optional |
Discount applied amount | Optional |
On time delivery | Optional |
Payment type | Optional |
Customer review score | Optional |
Create a retail channel churn predictive model
In the Dynamics 365 Customer Insights portal, select Intelligence > Predictions
Select the Retail channel churn tile, then select Use model.
Important
If the prerequisite entities aren't present, you won't see the Retail channel churn tile.
The Model name screen opens.
Model name
- Select Name and enter an easy-to-read name for your churn model.
- Select Output entity name and enter a name for your model's output entity, using only letters and numbers (no spaces). Your model's predictions will be stored in this entity.
- At the bottom of the screen, select Next.
Model preferences
Set model preferences to help the model create predictions that fit your business: how many days' data to evaluate, and what input value thresholds indicate churn.
- On the Preferences screen, select Prediction period and set the numbers of days the model will use to evaluate the chance of churn.
- Select Transaction volume decline threshold and set the percentage of transaction frequency that indicates churn (for example, if you set it to 0.2, the model will interpret an 80% drop in transaction frequency as churn).
- Select Transaction value decline threshold and set the percentage of transaction value that indicates churn (for example, if you set it to 0.1, the model will interpret a 90% drop in transaction value as churn).
- At the bottom of the screen, select Next.
Required data
- Locate the customer, session, and transaction entities that you identified as prerequisites.
- For each entity, select Add data and choose the source entity you identified.
- For each field on the form, select the corresponding customer, session or transaction input that you identified as a prerequisite.
- When all the fields are populated, select Save.
- At the bottom of the screen, select Next.
Data updates
Here, you set a frequency for retraining your model. Retraining improves the accuracy of predictions.
- Select Weekly or Monthly. Most businesses can retrain once per month and get a good accuracy for their prediction. Select Show example for a bit more help deciding.
- At the bottom, select Next.
Review and run
Review your model details. Select Edit beside any value to change it, or select a prior process step.
If everything looks good, skip this step. If you're not ready to run your churn model yet, select Save draft and then Close at the bottom-right corner. When you're ready to resume working on it, select Intelligence > Predictions in the Customer Insights navigation pane, and then on the My predictions tab select the edit icon next to the draft model's Prediction name.
Select Save and run to begin the prediction process. On the My predictions tab, you can see the status of your predictions. The process may take several hours to complete depending on the amount of data used in the prediction.
Review prediction status and results
To see the status and results for a prediction:
Go to Intelligence > Predictions and select the My predictions tab.
Select the three dots next to the name of prediction you want to review results for, and then select View.
There are three primary sections of data within the results page:
Training model performance: A, B, or C are possible scores. This score indicates the performance of the prediction and can help you make the decision to use the results stored in the output entity. Scores are determined based on the following rules:
- A when the model accurately predicted at least 50% of the total predictions, and when the percentage of inaccurate predictions for customers who became dormant is lower than 10%.
- B when the model accurately predicted at least 50% of the total predictions, and when the percentage of inaccurate predictions for customers who became dormant is greater than 10%.
- C when the model accurately predicted less than 50% of the total predictions.
Likelihood to churn: Groups of customers based on their predicted risk of churn. This data can help you later if you want to create a segment of customers with high churn risk. Such segments help you to understand, for example, where your cutoff should be for customer retention segments.
Most influential factors: There are many factors that are considered when creating your prediction. Each of the factors has its importance calculated for the aggregated predictions that a model creates. You can use these factors to help validate your prediction results. You can also use this information to create segments that could help influence churn risk for customers.
- Record level explainability: The output entity contains an explainability table listing the important factors that influenced each retail churn score. You can export these tables for various purposes.
Fix a failed prediction
If a prediction failed, you'll see an error message that explains what went wrong. For example, when the model ran, it didn't find any customers who were "churning" and the model failed training—this result might mean that you set the transaction thresholds too low.
To attempt to fix a prediction by reviewing error logs:
Go to Intelligence > Predictions, and then select the My predictions tab.
Select a prediction to review, and then select Logs.
Review all the errors. Each error's description of what condition caused it can help you decide how to fix the problem. For example, an error that says there's not enough data to accurately predict churn can be fixed by loading more data. Or, an error that says your model is producing a prediction of zero churning customers might mean you need to revise your model preferences with higher thresholds that indicate churn.
Manually refresh a prediction
Note
Predictions automatically refresh when your data refreshes, as configured in settings.
- Go to Intelligence > Predictions, and then select the My predictions tab.
- Select the vertical ellipses next to the prediction you want to refresh.
- Select Refresh.
Delete a prediction
Note
Deleting a prediction removes its output entity.
- Go to Intelligence > Predictions, and then select the My predictions tab.
- Select the vertical ellipsis next to the prediction that you want to delete.
- Select Delete.
Integration
The main output of the model is an entity with churn scores across your customer base and at the customer level. First- and third-party platforms and services can use this entity output via API for reporting and planning.
Compliance
Note
The churn prediction feature uses automated means to evaluate data and make predictions based on that data, and therefore has the capability to be used as a method of profiling, as that term is defined by various privacy laws and regulations. Retailers' use of this feature to process data may be subject to those laws or regulations. You are responsible for ensuring that your use of Dynamics 365 Customer Insights, including the predictive churn feature, complies with all applicable laws and regulations, including laws related to privacy, personal data, biometric data, data protection, and confidentiality of communications.
See also
Compliance for Microsoft Cloud for Retail Support for Microsoft Cloud for Retail What is Microsoft Cloud for Retail?