The FastStats Modelling module provides you with tools to describe your current data and predict possible activity in the future. As a marketer, you often start by wanting an overview of existing customers and their existing behaviour.
The purpose of Cluster analysis is purely descriptive and it identifies sub-groups that share common characteristics - e.g. People who donate to animal charities are of 4 types - A, B, C and D. There are all sorts of different applications of clustering including both tactical and strategic - e.g. Classifying all customers as High Value, Average Value, Dormant etc.
Profile and Decision Tree can be very useful in describing a particular set of customers (e.g. people who transact in a particular way), although their main purpose is predictive - e.g. Find people who might purchase a particular product or go to a type of event.
The Modelling Environment can be used to further your investigations by creating dimensions that allow you to explore your customers' transactional patterns/behaviours - e.g. Is their average length of holiday increasing, staying the same, or decreasing?
Part of the point of marketing is to acquire new customers, but also to change the behaviour of existing customers. For this you need a predictive tool, such as Profile or Decision Tree. You may want to predict which customers within an 'average value' cluster are most likely to move up to the 'high value' and so are worth marketing to. You may simply want to identify likely responders when planning a campaign. Having previously created a Cluster model may be helpful in building a predictive model such as this, but it is not a pre-requisite.
In order for a model to be useful, the patterns and characteristics encapsulated in the model must apply more generally to data other than that used to build the model - for example, to people who are not yet customers. A Cluster model encapsulates natural patterns within one set of data, whereas a predictive model is based on the distinguishing characteristics between two sets of data. You can use the Model Report to test how well predictive models apply to new data - for example, predicting behaviour of non-customers.
Standard modelling encapsulates the essential features of the real world at the time the models are built. As time passes, models should be reviewed to ensure that they are still representative of current customer behaviour.
Behavioural modelling takes this further; it not only studies features at the time the model is built, but also has the benefit of allowing you to study features from a different point in time.
Follow the links below to find out more about the data modelling in FastStats.