Clustering with knowledge instead of data (Marketing)

 In this post I would like to show how over time knowledge can supplement and replace data in data marketing.

This understanding, to my opinion is revolutionary because it goes against the current trend of accumulation of “data” which is prevalent in most companies. In fact, the more knowledge you have, the less data you need as I will demonstrate below.

Our first segmentation product in Japan was called Chomonicx and was launched in 2004. It was based on an earlier US model, Prizm from Claritas to be specific and segmented, as can be seen below, the Japanese population into 32 clusters at the Chome level. (A Chome being equivalent to a large block in the US. There are about 210,000 Chome in Japan.)

 

This product was quite successful as there was a need at the time for such segmentation. It has since been updated every 5 years and still exist today as Chomonicx 4.0

Our next product should have been at the Banchi level (A Chome being built around a few Banchi) but whereas a lot of data was available at the Chome level in Japan including of course the National Census, there was no data at the Banchi level and consequently nobody was using Banchi for marketing.

In-between Internet marketing exploded with the help of Google and Facebook and became the dominant channel for the new Multimedia marketing strategy for most large companies.  

Internet Marketing is of course at the individual level and consequently presents two challenges:   

The first one is privacy and the need to de-identify or anonymize data.

The second is getting access to the data. The breakthrough came in 2016 when we got access to the Zenrin building data from which we could build population proxy data at the household level. The result was point data as seen on the map below.

 

This revolution more than evolution was based on more data quantitatively, everything about every building but less data overall for a result 100 times more specific since the data although still an index could now be combined one to one with client data. The colors show the Chome cluster system (Chomonicx) and the points show the new data. There is simply no comparison as far as accuracy is concerned.

But amazingly, while doing this work,  we realized that almost all the information at the point level could be deduced from the address alone! This is where knowledge completely replaces data as you can actually build complex knowledge-based system with far less data provided you know how to do it.

We will describe the methodology in a future post but because of the structure of the address system in Japan, and more generally in many Asian countries, based on blocks, not on streets, seen as a gap between the blocks, it is actually easy to build a hierarchy clustering system constructed on the address which contains all the data needed and implies a natural segmentation easy to implement.