July 16, 2013

1991: Statistical Models

Songs that were popular in 1991:

  • "Everything I Do" - Bryan Adams
  • "Black or White" - Michael Jackson
  • "Joyride" - Roxette (we're seeing a lot of Roxette, don't you think?  I bet that ends, soon.)
  • "Wind of Change" - Scorpions
  • "Losing My Religion" - REM
One of the great illusions of "statistical modeling", or as it is popularly known today as "Big Data", is that most techniques, regardless how different they are, do the exact same thing.

If you work at a large company, like Lands' End, statistical modeling means everything.  I created a model in 1991 that increased the productivity of catalogs by about 1%.  This increase resulted in a million dollars of incremental profit - generated by me, a lowly statistical analyst known for producing 3-D contour plots.

You get noticed as a 25 year old analyst when you produce a million dollars of profit all on your own.

But the attention isn't deserved, honestly.

Here's the thing.  If a model is built responsibly, then 97% of the names of responsibly built model #1 will be the same as the names of responsibly built model #2.

So when your vendor has a "Big Data" solution for you - please keep in mind that 97% of the names are going to be identical to the names selected by your current methodology.

Identical.

The argument, then, is about the effectiveness of the 3% of names that are different.

At Lands' End, in 1991, I picked names, 3% different than the names previously selected, names that were fundamentally better.  This was a crucial step in earning credibility among Management.

If you work for a large company, this opportunity exists for each and every one of you - those working for catalog marketers, those working for email marketers, those working in the retargeting industry.  Take advantage of the opportunity presented to you!

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.

Have You Ever Looked At Performance This Way?

I bring this up annually. But the analysis is so darn much fun! Take every customer who purchased between Black Friday and Cyber Monday last...