Creating custom predictions
In this article, you will get know how to set up your first custom prediction. The instructions include the procedures that need to be performed in every step in the Predictions wizard:
- Select the model type
- Select customers to be analyzed
- Choose the prediction target
- Select model inputs
- Schedule recalculation and result settings
- Save the prediction
Prerequisites
- Enable the Custom predictions. It usually takes several hours to initialize the Predictions module and perform necessary calculations.
- You should know how to create segmentations or have a segment of customers ready for use as the prediction target.
- Create an expression based on which the prediction will be made. For example, if you use an expression that calculates Email Open Rate in last 30 days, you will get a prediction for Email Open Rate for the next 30 days.
Configure a prediction
Select the model type
- Go to
> New prediction.
- On the pop-up select one of the following options:
-
Use predefined prediction - This option allows you to use the ready-made prediction scenarios and adjust its settings to your preferences in a user-friendly, step-by-step configuration form by following hints shown on the interface. The following business objectives and variants are covered:
Business objective Variant Reduce churn - Find customers who are likely to leave;
- Find customers who are likely to unsubscribeIdentify new potential customers Find best matching customers for an annual/seasonal campaign Predict buying behavior - Find customers who will buy a specific item, brand, or category;
- Find customers who will buy any itemPredict lifetime value Find customers who are likely to generate the most value Note: As a result of performing the steps in the Use predefined prediction form, a set of analyses will be created and available on the list of analyses (such as segmentations, expressions, and so on). -
Create from scratch - This option allows you to create your own configuration of a custom prediction to achieve your custom business objective. If you use this option, select one of the following models:
- Regression is suited for creating predictions that return numerical results. It’s best used in cases such as:
- Predicting the amount of money spent by particular group of customers in a defined time range.
- Predicting the amount of items purchased in the defined time range.
- Classification is suited for performing the analyses that return the true/false (or 1/0) values. It’s best used when you want to find answers to the following questions:
- Will a customer belong to a particular group of customers?
- Will a customer leave in the next 30 days?
- Regression is suited for creating predictions that return numerical results. It’s best used in cases such as:
-
Select customers to be analyzed
Select the audience for whom you want to prepare a prediction.

- In the Audience section, click Define.
- Click Choose segmentation and perform one of the following actions:
- From the list, select an existing segmentation.
You can use the search field. - To define a new group segmentation, at the bottom of the dropdown list, click the Create new button.
- In the Segmentation name field, enter the name of the group of customers and click Next step.
- Build the segmentation.
See this article for instructions. - Click Create segmentation.
- From the list, select an existing segmentation.
- Confirm by clicking Apply.
Choose the prediction target
Select the expression based on which the prediction will be made. For example, if you select an expression that calculates Email Open Rate in last 30 days, you will get prediction for Email Open Rate for the next 30 days.

- From the dropdown list, select the expression you prepared previously.
- Confirm by clicking Apply.
Select model inputs
Select the features that the system will use to support the prediction process. A feature is a variable or an event that can be used by AI engine to make a prediction. The list already contains predefined features.

- Click Add feature.
- From the dropdown list, select one of two options:
- Manually - You can pick features on your own.
- On the list, select the checkboxes next to the features you want to include.
- Confirm your selection by clicking Add.
- Automatically - Features are added to the list by the AI engine (recommended).
- Manually - You can pick features on your own.
- After making a selection, click Apply.
Feature analysis
The unit of the feature depends on the type of the feature. For the features that are events, the unit is an occurrence of an event.

Assuming that a feature is an event, then:
- Count - The number of the event occurrence
- Min - The minimum number of the event occurrences
- Max - The maximum number of the event occurrences
- Missing - The number of customers for whom the feature is not counted
- Mean - The mean occurrence of the event
Schedule recalculation and result settings
In this section, define the frequency of recalculating the prediction and settings of the event that is generated for customers for whom the prediction is made.

-
In the Model configuration section, select the number of days after which the prediction is recalculated.
-
From the How far in advance do you want to make a prediction? dropdown, select the number of days for which you want to make a prediction (calculated from the current date).
-
To switch on repeatable calculation of the prediction, select the Set up recurring prediction calculation checkbox.
By default, the calculation is performed once. -
In the How would you like to display results? section, select the scale of prediction results - two- or five-point scale.
The scale is represented as ascore_label
parameter of thesnr.prediction.score
event. -
Use the slider to define the number of features displayed in the parameters of the event generated for a customer for whom the prediction was made.
Note: One of the features is aprediction
parameter, it is included in thesnr.prediction.score
event and it can be used while selecting customers for your campaigns. -
In the Name field, enter the value of the
scoreName
parameter of the event generated (snr.prediction.score
) when a prediction is made.Important:You can use the following characters:
- a-z - only lower case
- 0-9
- special characters:
.
and_
-
Click Apply.
Save the prediction
To save the prediction:
- as a draft, click Save.
- and calculate, click Save & Calculate.
Tip: You can check the logs of the activated prediction. Learn more.
Result: The output of the prediction is an event generated on the profile card.
Understanding custom prediction events
The outcomes of all custom predictions are saved as snr.prediction.score
events in the Synerise platform. Thanks to that, the predictions events are available in the Analytics, Communication or Automation modules. The schema of a prediction event is standardized. The attributes are explained in the table below the figure.

Attribute name | Description |
---|---|
action | An event type for the prediction - snr.prediction.score |
added by | An entity that generated this event. In the case of prediction events, it is always Synerise. |
modelId | A unique identifier of the prediction |
modelName | The name of the prediction |
score_label | Prediction output: two- or five-point scale label. You can select the scale in the settings of the prediction. |
scoreName (deprecated) | The name of the prediction assigned while creating a prediction |
score | The result of the prediction (based on the expression which was selected as a target while preparing a prediction). |
percentile | Prediction percentile |
classes | Available classes: true , false (only for the Classification type of model) |
prediction | Predicted class (only for Classification type of model) |
feature_influence | Local feature importance |
clientId | Customer’s identifier |
time | Time when a prediction was generated, as a Unix timestamp |
configurationVersion | Allows to verify whether two consecutive calculations ran with the settings (only for debug) |
modelVersion | Model version allows to verify whether two consecutive calculations ran on the same model parameters (only for debug) |