Monday 4 June, 2007

Direct marketing needn’t make sense…

…it only needs to make money. Didn’t Denny Hatch say something like that? Anyway, here’s something similar from a data mining book (Data Mining Techniques, Berry & Linoff): “If a direct mail firm obtains a model that can accurately predict which numbers of a prospect pool are most likely to respond to a certain solicitation, the firm may not care how or why the model works (italics mine).”

I’ve heard similar statements elsewhere.

And all database marketing books teach you basic statistics, e.g., confidence intervals, comparing percentages, and so on, so that you can take dispassionate decisions based on response data.

The first thing we are taught is that we must ‘test, test, and test.’ And we enviously read Western case studies of the wonders wrought by testing.

But something is missing.

When in college, we were taught that physics is, in a way, applied mathematics. It has to make sense. The calculations have to model the real world.

From what little I’ve read about business statistics, its purpose doesn’t seem to be very different.

Right now, I have two (naive?) questions:
1. Given that there is bound to be a time gap between the test and scale-up, and that much can change during that gap, how prudent is it to base the scale-up on test results? I suppose there cannot be any easy (rule of the thumb) answer to this. And some gurus have advised that tests should be repeated before one scales up. Nonetheless, I am yet to come across a case study on ignoring test results. (There are plenty on chimpanzees and ink dots choosing investment portfolios that outperform analysts’ selections [though they do sound too good to be true].)

2. Which of these will be most surprising?
a. Software meant for the express purpose of making segments makes some (peculiar) segments

b. It would have made segments even if the data wasn’t about neighbourhood demographic profiles and frequency of previous orders but about first alphabet of middle name and direction of office desk

c. These segments, no matter how they are created, can be projected into the future, e.g., if Segment A shows significantly higher response for products of Type X than Segment B today they will do so (in most cases) in the future

I’d be unsurprised by if statements a and b came out true, but somewhat surprised if c did, unless the differences between A and B are readily explicable (e.g., A are male, B are female, and X are feminine cosmetic products).

I am also very curious to know what loyalty marketers do with their double-digit segments.

There are quite a few enlightening and entertaining books against the senseless use of statistics to sensationalise or sell or both, particularly in the stock markets. Shouldn’t there be one examining its application to marketing, particualrly direct and loyalty marketing, too?