Subscribe to DSC Newsletter

I have been doing some form of segmentation since the mid seventys. At the begining we just relied on the mathematics used in the segmentation strategy to "profile" the segments. This works fine with scientists and statisticians because they understood the math concepts.
For the last 15 years or so, as I became more involve with delivering segmentation results to a more general audience, I quicly discovered that the sementation profile this audience required need to be more qualitative then quantitative.
We actually employed the skills of people trained in Library Science and Biology to help create descriptive profiling. This special talent, required associating data not used to build the segment to members of a segment (i.e. demographics).
I have found myself and others spending 4 to 5 times the effort defining a profile then it took to do the segmentation itself.
Recently I was forced to do this "profiling" myself. I started to look at the techniques required and found that my quantitative approach was too esoteric for my audience.

Tags: Profiling, Segmentation

Views: 328

Reply to This

Replies to This Discussion

What I did was:
I asked the business subject matter experts to meet with the client and formulate a list of questions that could be answered with data I had available to me.
I formulated the data and built models for each segment that attempt to classify the data as being in a segment based on the questions of interest to the client. (nothing new here. this is a common technique used for years.)
I noticed that because I was using KXEN K2R/K2C to cluster that each of my independent variables where being remapped into groups and ranges. I then realised I might be able to rexamine these variables broken into irregular shapped bins against the frequency distribution among the segments.
I noticed patterns emerging that are quantitative in nature but supply qualitative descriptions. While the "elevator speech" for each segment still needs to be crafted I was able to focus the attention of the subject matter experts to areas that have some support in the math but are also related to the client's business needs.
I've had a similar experience, where factor analysis or clustering is a purely statistical methodology, but they can be reinterpreted using a more "qualitative" assessment that basically boils down to the key, understandable variables. e.g., factor 1 might be explained as customers with high income, low price sensitivity, and concern for quality. The actual factor 1 could include many more metrics, but the key ones fit within a qualitative heuristic.


Pell Research

Profiling is one the key and challenging aspect in the descriptive modeling analysis.
If more then 10 segments it is little difficult to explain the behavior of the clusters. What you need to do is give more graphical representations rather then quantitative numbers. Because business require how statistical conclusions to Business solutions.
I agree some clients are very graphical and some want to see the numbers.

One thing is for certain, there is lots of money spent on "branding" some marketing segmentation strategies.

In one customer loyalty segmentation strategy the analytical work that I did cost only about 10% of the total program.

In cases like this what the client is looking for is specific "common sense" differences between the segments that can be used by the "branding" team.
Infact most of the retail domains brands conducts their own segmentation rather then using the techniques for groups.

Because there are lot of money spending on banding for businesses to create the image in the market to target the specific profiles.

There are limited number of segmentation projects are implemented in the some fo the FMCG, CPG industries.

Here the logic is to understand the Loyalty analysis is to increase the customer retain and reduce the churn percentage then segmentation strategy.
Whereas Branding is to gives the companies to long-term profits using primary research surveys to understand the consumer needs and satisfaction surveys.
John: When you say that you built a descriptive profile by "associating data not used to build the segment to members of a segment". Why were these demographics not part of the original segmentation? Sounds like you were using some sort of cross-validation to confirm or refute the existence of an already existing model.

-Ralph Winters
The main data was store sales data. The client and the client support team asked us to refrain from adding very generalized demographic data. (The demographics covered the composite statistics of the population immediately local to the store.) However they were very interested in what segments ethnicity showed up strongly,

A colleague of mine refreshed my memory about a common pracitice of the financial community. They will also take and decile client data lets say by average worth to the financial institution and see how well these deciles fit per segment. If there are areas of under perfoming, they can determine marketing campaigns. If this variable was an element of the segmentation model you would get little actionable information.


On Data Science Central

© 2020   TechTarget, Inc.   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service