January 11, 2016

Targeting Variables

Here's how I approach a targeting variable.

Essentially, I weight historical demand by coefficients. 

In this model, I measured 12-month future spend on the basis of demand spent in the following increments ... 0-1 month ago ... 1-2 months ago ... 2-3 months ago ... 4-6 months ago ... 7-9 months ago ... 10-12 months ago ... 13-18 months ago ... 19-24 months ago ... 25-36 months ago ... and 37-48 months ago. Look at the coefficients.
  • One Month Demand = 0.668.
  • Two Month Demand = 0.584.
  • Three Month Demand = 0.521.
  • 4-6 Month Demand = 0.443.
  • 7-9 Month Demand = 0.344.
  • 10-12 Month Demand = 0.280.
  • 13-18 Month Demand = 0.201.
  • 19-24 Month Demand = 0.146.
  • 25-36 Month Demand = 0.093.
  • 37-48 Month Demand = 0.061.
Next, I divide each coefficient by the one month demand coefficient of 0.668.
  • One Month Demand = 100.0%
  • Two Month Demand = 87.4%
  • Three Month Demand = 78.0%
  • 4-6 Month Demand = 66.3%
  • 7-9 Month Demand = 51.5%
  • 10-12 Month Demand = 41.9%
  • 13-18 Month Demand = 30.1%
  • 19-24 Month Demand = 21.9%
  • 25-36 Month Demand = 13.9%
  • 37-48 Month Demand = 9.1%
So, if I am scoring the file today, and a customer purchased from a product category three months ago, I weight each dollar by 78%.

I do this for each targeting variable.

I know, I know, statistical gurus are going to tell you the 1,493 reasons why this isn't an appropriate methodology. Too bad.

Tomorrow, I will show you the targeting models, and then, I will describe how we use the targeting models in email marketing.

Memorial Day Weekend

There's the obvious reason why we "celebrate" Memorial Day. And then there's the secondary outcome ... an unofficial ...