06-01-2014 11:17 AM
Hi recently i have started a new project which targets to identify propensity to purchase of any financial product for every active customer. However i am not sure on what methodlogy to follow...
Here is my plan
- First i will go to 6 months prior to today, and set my flag on ownership of specific product on T-6 mthns
- Then i will collect data as of T-6 mthns (or should i take today's data)
-I will take every specific product's data on my datamart expect the data related to my target variable..
-I will employ a decision tree to make a reduction in data size
-Then i will built my propensity model on outcomes of previous model
is that made sense?
LAstly i am not sure that
do i have to estimate model on all active customer segment or all customer segment. Hence if i calculated my scores on all customer segment many of the individuals will exhibit "0" value for many product ownership and that will deteriorate my models consistency..
am i right?
06-05-2014 01:09 PM
Sounds like a nice project. How did you come up with the 6-month observation window, just business knowledge? You can back up that decision with time series analysis or just testing for any seasonality. 6-months could make sense as long as it accounts for high sale seasons like Christmas, special promotions or holidays.
Without knowing too much about the data, you might need to assess what works best for your definitions of active/inactive and how often they go from one to another. It might make sense to have a model for actives, inactive for less than 2 quarters, inactive 2-4 quarters, and inactives more than a year. Establishing those segments will add a lot of value to your business and to your models.