The Prediction Models module allows you to create and manage Prediction Models to optimize your marketing campaigns. For example, Prediction Models can predict which customers are more likely to make a purchase after a Campaign, allowing you to spend more resources on more promising customers.
On the Prediction Models page, you can see a button to create a new model, and a list of your existing models. You can click on any Model to view its configuration.
You can also see Status Information of the Models here. For models where calculation is finished successfully, click on the model to see its curation status. You can click on the Question Mark icon (?) next to Curation Overview to learn more.
Create new Prediction Model
Start with creating a new Prediction model by clicking the button: Create new Prediction Model.
Choose a Model Template.
-
Customer Lifetime Value: Customer lifetime value (CLV) is the customer value for a specified period. It is a measurement of how valuable a customer is to your company, not just on a purchase-by-purchase basis but across the whole relationship.
It is important to understand the Value of your Customers, as it helps you make better business decisions, such as which customers to invest costly Print Campaigns for. With Customer Acquisition Costs rising across the board, predicting the Value a Customer will return in their lifetime can maximise profits while cutting costs.
- Interval-Based Reaction: This type of Model takes into account that different Customers have their individual purchase patterns. Hence, while the Inactivity-Based Reactivation gives a fixed definition of Inactive Customer (e.g those who haven't made a purchase for 4 or 6 weeks), This Model identifies purchase frequency of all customers, and targets Customers those who have been inactive recently, compared to their own usual purchase frequency.
-
Inactivity-Based Reactivation: A Interval-Based Reaction Model targets customer who have made purchases in the past, but have been inactive since. In a "Classic" Re-activation model, Inactive Customers are identified as those who have not made a purchase in a fixed amount of time. This time period will largely depend on the nature of your Business.
In a Inactivity-Based Reactivation model, we follow the same procedure as a CLV model, with the exception of a Custom Filter added by Spotler Predict to identify Inactive Customers.
- Second Purchase Prediction: With a Second Purchase Prediction Model, the target group is restricted to one-time buyers in addition to the information from the CLV model. This can be valuable if you have a large number of One-Time Buyers, whom you wish to turn into Repeat Customers.
In this example we explain more about the settings of a Custom Prediction. You will navigate through 3 steps:
Step 1 - Info
Give your Model a meaningful name. You can optionally add a label and a description.
Backtest
Choose if you want to create a model for a back test. A backtest offers the possibility to train a Model on past data and get a forecast for a past point in time (as opposed to a live Model). As the forecast is based on a point in time in the past, the Model performance can be evaluated immediately.
Click on Next.
Step 2 - Target Variable
Target variable describes the event that should be predicted and consists of a temporal and a contextual component. The temporal component describes the time period in the future for which a forecast should be created. The contextual component describes the event that should be forecast.
- Provide the time-offset for your data. The time offset (or time lag) is the time between between the last selection and the day the marketing activity reaches the customer (in days). The time offset ensures that process times that are necessary (e.g. address printing) are taken into account in the model. If there is no time lag, select 1 day(s).
- Provide the Prediction Period: the time period for which a prediction should be created. The Prediction Period defines the future period for which a forecast of the purchase probability and the expected turnover should be created.
-
Define your transaction filter. Here you can choose how events are taken into account.
Define the influence the order events have on the target variable. Events that are marked positive are predicted by the Model.
Events that are marked negative are subtracted from positive events. Only when a positive amount is preserved during the netting, the sum of the events will count as positive.
Neutral events are used to calculate variables, but do not influence the Model target. Events marked with ignore are removed from the raw data at the beginning of data preparation.
-
Define a target filter.
Examples:- If you want to predict the purchasing probability and expected revenue within a product category e.g. "T-shirts" you can specify the product category using the provided filter input field.
- If you want to predict the purchasing probability and expected revenue for a combination of a brand and product category like "T-shirts from Nike" you can specify the brand and the product category using the provided filter input field.
Step 3 - Target Group
The target group defines which customers should be scored by the Model.
Examples:
- Number of purchases a customer made before today.
- Time passed since the last purchase.
- Country a customer lives in.
Where no conditions are chosen, the Model will by default be trained on all available customers.
Click on Save & Close.
Once your Model is created, it should show the status initial. You can now run the Model by clicking on the Commit button. Note that once a model is processed, it can no longer be edited.