Profile: What is a PWE Model?

  • Once you have generated a profile, you can score the whole database to show how well each record on the database fits the profile that you have generated.  This process is called building a model. The modelling technique included in FastStats is Predictive Weight of Evidence (PWE), from Apteco Ltd.

  • By scoring the database, you can find new records that fit the characteristics of the records that you have profiled.  If, for example, you profile your best customers and then score the database, the records that score highly are a good fit to the characteristics of your best customers, and are therefore good prospects to market to.

  • To make building a model easier to understand, think of a new design of a building – you would understand if the architect said he or she had modelled the design of the building on the features of another building.  In FastStats we are doing a similar process - we are modelling the design of our score on the features of the records we profiled.  Neither is an exact replica of the original, but both include identifiable characteristics.

  • By building a PWE model we are creating a score that can be applied to all the records on the database, and not just to the records that you can already select.  It is easy in FastStats to select your existing customers, it is less easy to select the best new prospects to market to.  The PWE Model you build will help you select those records that are worth marketing to from those that are not by helping find those who look similar to your existing customers.  In summary, the score is helping you segment the database by highlighting those records which are similar to the existing group and those that are not.

  • Unfortunately, although we might wish FastStats could just tell us who to market to, it does not have enough information to do this.  We have to help it by providing background knowledge and marketing expertise.  FastStats does not treat the segmentation (or separation) of the good and bad records as black and white.  Instead it produces a range of scores that we can use.  The highest scoring records are the best, but it is up to us to decide how far down the scoring scheme it is worth using the data.