About predictive models

Models can be used to predict what is likely to happen in the future, based on patterns in past data. For example, you might use a model to predict which customers are least likely to churn, or most likely to respond to a particular offer, based on characteristics such as income, age, and the organizations and memberships they subscribe to.

Models can be used in the same way as rules, but while rules may be based on corporate policies, business logic, or other assumptions, models are built on actual observations of past results, and can discover patterns that may not otherwise be apparent. While rules bring common business logic to applications, models lend insight and predictive power.

Data for building and scoring models

Two different types of data are used in the modeling process:

To build the model, you need information about the thing you want to predict. For example, if you want to predict churn, you need information about customers who have churned in the past. This is often referred to as historical or analytical data, and must contain some or all of the fields in the project data model, plus an additional field that records the outcome or result you want to predict. This extra field is used as the target for modeling.
To use the model in predicting future results, you need data about the group or population you are interested in, such as potential customers or incoming claims, for example. This is often referred to as operational data, or scoring data. The project data model is typically based on this data.
If the target field is included in the data model, the Operational column should not be selected for this field, because it is not available when scoring the model. Having built the model using historical data, the goal in scoring is to apply the model to new data where the outcome is not already known.