In simple terms, a model is a mathematical tool that represents objects and their relationships. Within the AI Workbench, a model is essentially an algorithm that processes data, identifies patterns, and generates predictions. For instance, you can build a model to estimate the likelihood of a VIP customer leaving your service.
When you save the settings of a model the system creates a new Predictive Indicator with the Settings you chose.
You can filter the customer data that’s part of the model run by Segments, Processing Purposes, and how much customer data in the past will be considered (Observation period). You can also configure some of the model basic and advanced parameters.
The condition field (in the screen above represented by the Churn condition) establishes a custom boolean criterion that will be used by the model to help defining the value of the indicator for each customer. In the image we can read the Churn condition as: count all orders grouped by month in the last 2 months that are smaller than 2. Meaning that customers with zero or only one order will have a great impact on the churn indicator.
You can pick other different Churn Condition Templates to leverage existing common scenarios or starting point of your custom criteria building.
Use the Visual Editor or the Editable Text to modify the churn condition to fit into your business needs. The Visual Editor allows you to insert Operators, Variables, Functions, Customer Schema attributes, and Literals. The Editable Text allows you to freely type-in the Condition instead of visually building it.
You can also leverage pre-built Templates to see examples, or pick one Condition based on common scenarios or as a starting point for your custom criteria.
Please see some Churn Condition Templates available out of box in the table below:
| Description | Condition |
|---|
| No Orders in the Last 90 Days | count(Activities.Orders.Id , inLast(90 day)) = 0 |
| No Orders in the Last 3 Months | count(Activities.Orders.Id , inLast(3 month)) = 0 |
| No Orders in the Last Year | count(Activities.Orders.Id , inLast(1 year)) = 0 |
| Repeated Low Number of Monthly Orders in Last 3 Months | count(Activities.Orders.Id , groupBy(month) , inLast(3 month)) < 2 |
| Repeated Low Average Spend in the Last 3 Months | avg(Activities.Orders.Amount , groupBy(month) , inLast(3 month)) < 5 |
As you can tell, functions such avg, count, max, min, sum are supported. Other functions such as groupBy will group the aggregating functions per a certain amount of time, and inLast filters the data by a certain amount of time in the past.
If you are configuring the Customer Lifetime Value CLV Formula instead which projects a resulting value to the model. The CLV Formula screen is very similar to the Churn Condition, sharing most of its elements.
Some of the Templates offered by the CLV Formula field:
| Description | Condition |
|---|
| Sum Amount for Last 90 Days | sum(Activities.Orders.Amount , inLast(90 day)) |
| Sum Amount Without Tax for Last 90 Days | sum(Activities.Orders.Amount , inLast(90 day)) - sum(Activities.Orders.Tax , inLast(90 day)) |
| Average Amount for Last 90 Days | avg(Activities.Orders.Amount , inLast(90 day)) |
After successfully running your model configuration (or Settings), the system will catalog that in a Run ID, and you can choose which RUN ID will get published into the corresponding Predictive Indicator in the main CDP Console.
The system also allows you the ad-hoc generation of different Run IDs by choosing Run Prediction, Stop a currently executing Run, check the execution Log, Copy a Run ID with Status Completed to create another Predictive Indicator with copied over Settings, Edit the Scheduled Prediction to keep your Indicator up-to-date, or Delete a Run.
Go back to the main CDP Console, select Activate on your new Predictive Indicator so you can use it to generate various Insights such as Segments, Explorations, Audiences, CX Journeys and Flows.
Going back to the AI Workbench Console, you can also Schedule Prediction executions of an existing Model Run to keep the corresponding Predictive Indicator updated.
