Segmentation is to split the market into different groups – each group with similar characteristics. Why do we do this? Because, STP or Segmentation, Targeting, Positioning will help the company select the right segment to target with right positioning. You could not sell your product or services to all people on earth. For example, not all people will buy fruit juices, only those who concerned about health will be interested.
Generally, there are many methods to perform segmentation: Geographic, Demographic, Psychographic, and Behavior. 
To obtain the data, there are several methods from marketing research (focus groups, survey, etc.) to customer database (purchase data, customer profile, etc.).
I’d discuss about few algorithm or concepts that are popular among data guys for segmentation below.
1) RFM Segmentation
RFM is to split customer based on purchase behavior
- ‘Recency’ – how recent is the last purchase
- ‘Frequency’ – how often the customer purchases
- ‘Monetary Value’ – how much money the customer spend
You can see more on how to perform RFM here.
It is defined in a way that R is the most predictive of future purchases, followed by F and M. In other words, those who recently purchased the product are more likely to purchase in the future – especially if they’ve purchased many times.
Key benefits of RFM is that it only requires minimal data and no need for advance statistical modelling knowledge to run. In fact, it could be done in just Microsoft Excel.
Many retailers use RFM to understand their customer purchase pattern and how can they priority which customer to target.
2) K-means Clustering
K-means clustering asked analysis for number for clusters (or segments) and which variables to use. Then, it groups the data to be segments for us.
You can see more how the algorithm works here.
Unlike RFM segmentation, K-means clustering can take any variables as input e.g. Gender, Age, Purchase frequency, etc. Thus, it makes K-means clustering more flexible to implement.
However, doing so requires analysts to understand the business well to understand which actually is significant variable and which type of segmentation we are doing i.e. Are we using behavior segmentation or geographic segmentation or both? Moreover, it is quite sensitive to different scales. This means that the analyst must knows how to prepare the data before running the algorithm.
For me, one of the most useful case for K-means clustering for segmentation is actually to run it among purchase category. For example, a retailer sells product in facial care, hair care, body care, cosmetics and miscellaneous. We can calculate % of of purchase is contributed from each specific category. (Hint: This will be on the same scale as we’re doing it in term of %, not sales) Then, we run K-means algorithm on the data. This will allow the retailer to understand their segments and plan appropriate marketing strategy for each segment.
Note: Some of you may have heard of hierarchical clustering as well. The usage is similar to K-means but the way the algorithm works requires more computational resources and not suitable for big data set. So, I think it’s not a good choice for segmentation.
Even though RFM may seem different, I often found that combining RFM and K-means together could give a very interesting result. We can know that the people who come often buy products in which category and which category has lower loyalty. 
At Davoy, we’re expert in consumer analytics and we can help you with segmentation. Just fill in the contact form and we’ll get back to you for free first consultation!
