Data Science Nugget - 3

This is the third article in this series and will cover the application of Data Science to Customer Segmentation.

The need for customer segmentation draws from the premise that - All customers are not created equal. Customer segmentation is the practice of grouping enterprise's customers into groups that reflect similarity in behavior or need. The goal of segmentation is decide how to uniquely connect with each customer group in order to provide relevant services and maximize the revenue from that segment.

Customer segmentation can be done over a large number of parameters largely depending on the business need and context. However at a very high level, the major types of Customer segmentation are:-

• Demographic Segmentation - age, education, gender etc.
• Geographic Segmentation - Urban/Rural, State, District, City, Postal code etc.
• Behavioral Segmentation - Usage patterns like websites, mobile apps, past products bought, how frequently do they buy etc.
• Psychographic Segmentation - Lifestyle - prefers a high end phone, branded items, Luxury goods, interested in pets, opinions etc.

Customer segmentation is a dynamic activity as the customer behavior keeps changing with time and experience. Hence it is important to actively segment customers to track the dynamic changes based on the more recent data. A native approach is to segment customers in terms of their life time value (Customer Lifetime Value) and preferentially treat the more valued customers. Typically the priority treatment given to customers with a bank balance above a threshold.

The other common types are segmentation using

• Rule based segmentation - simple basic rules are configured and tracked. Like people who abandon items from their shopping cart are interested in them, so they can be targeted later. This is very common in the ecommerce world.
• Cluster analysis - uses a mathematical model to discover groups of similar customers. Cluster analysis is an example of Unsupervised Learning algorithm, a type of machine learning algorithm used to draw inferences from datasets without human intervention. A common cluster analysis method is a mathematical algorithm known as k-means cluster, where the algorithm groups data points into distinct non-overlapping homogeneous subgroups using a predefined number of "k" centroids. The homogeneous subgroups are known as “customer archetypes” or “personas”. Hierarchical cluster analysis (HCA) is another method of cluster analysis where one can build a hierarchy of clusters without having fixed number of clusters.
• RFM segmentation - is a powerful way to identify groups of customers that are special. RFM stands for Recency, Frequency and Monetary. It utilizes an objective and numerical scale that yields a concise and measurable depiction of customers. In some cases, Profitability is also added as an element, making this an RFMP model. Here the customers are valued on three scales of relevance to the Enterprise -

- Recency - How recently did they purchase ?

- Frequency - How frequently do they purchase ?

- Monetary - What is the monetary value of their purchase

The customers who score high on all the scales are the preferred customers. The scales can be rated from 1 to 5 or any as preferred.

In a business scenario, one can even combine one or more segmentation models in order to achieve their goals. Customer segmentation benefits both the Enterprise and their customers. It helps the helps the enterprise reward their loyal and profitable customers and promote the targeted behavior. At the same time the customers get the service quality that they deserve.