Here's an example of targeting variables in practice ... I am predicting the annual probability of a customer purchasing from weighted category 11.
It's a logistic regression equation:
- Logit = -1.851 + 0.018*(weighted category 10 / 1000) + 0.051*(weighted category 11 / 1000) + 0.006*(weighted category 13).
- Annual Response = EXP(Logit) / (1 + EXP(Logit)).
Ok, enough of the geeky stuff.
Let's say you wanted to populate your email campaign with the four most likely categories, sort of like what you see below:
You go into your predictions at a category level:
- Category 01 = 38%.
- Category 02 = 19%.
- Category 03 = 77%.
- Category 04 = 2%.
- Category 05 = 4%.
- Category 06 = 6%.
- Category 07 = 27%.
- Category 08 = 1%.
- Category 09 = 3%.
- Category 10 = 18%.
- Category 11 = 9%.
- Category 12 = 5%.
- Category 13 = 11%.
Clearly, this customer is likely to buy from categories 03 / 01 / 07 / 02, with category 10 a close fifth. Pick an item from these categories, populate your email campaign, and send customers to an appropriate landing page.
You can update your scores on a hourly / daily / weekly basis, whatever works for you. Work with your email service provider to populate based on category scores, and you've just improved the productivity of your email campaigns by 20%!
And if you are a catalog marketer, this is how you are going to paginate your personalized merchandise assortment in future catalogs.
Yes - use whatever modeling methodology you like - you don't have to copy my logistic regression structure using weighted variables - use whatever tactic you like.