I understand the need for an analyst to “prove” to themselves that certain structures and methods are in place. For example, the idea of “then when they show some significance statistically” above. This would be the proper, purely scientific way to approach a problem. Like Ron’s comment “but that’s a decision for the statisticians to make”.

As a database marketing person, I have to tell you that when you see an outcome repeated over and over and over in many different tests across many different industries, you tend to believe it is true without the need for elaborate science in place first to prove it is true.

This is the case with using behavioral data versus demographic data to drive profitability in marketing campaigns. If you build a predictive model using both elements – and the objective is behavior – you will find the behavioral variables always move to the top, get the largest weights.

Now, as I said in the post, having demographic data is better than having no data at all. Many times, especially in acquisition marketing, this is the case. So you use what you can get.

What drives me crazy (about marketers, not analysts) is they tend to ignore the behavioral and focus on the demographics, despite evidence the behavioral variables are controlling outcome.

Before this comment turns into a post, I think I will do a follow-up…

The only point where I dont think I agree with you in the implementation part. I would prefer to use all attributes and their information value first irrespective of their type and source. Then when they show some significance statistically, then find a good way to use it. The target answers can be “what”, “when”, “who” or anything. I am not much worried on one particular answer but more lean to the actionable insight that drives profit.

Bhupendra

I’ve taken the liberty to add to your post (and Avinash’s). In fact the topic is rich enough to merit a whole series of posts:
http://adelino.typepad.com/adelino_marketing/2007/08/web-analytics-a.html