Customer segmentation techniques

Customer segmentation techniques

Introduction

This article briefly describes the basics of customer segmentation, what it means and what its benefits are. Furthermore some of the most common techniques are also mentioned and described.

Customer segmentation in general

What is customer segmentation in general? It is sort of a practice, which divides customers in smaller groups based on multiple characteristics and gives the business a better understanding of its customers, which has a great value in B2C relationship. Some of the other benefits could be identification of least profitable customer group, improvement of customer service, data for making marketing decisions (which group to target) or avoiding unprofitable markets. The main goal of this whole process should be to maximize the value of each single customer. There could be other goals of customer segmentation. However, it rather depends on the particular business and its preferences. Examples of common segmentation objectives include development of new product, differentiated customer servicing or targeting prospects with the highest profit potential.
Of course the first expectation to make the customer segmentation possible is that the customer base is dividable. That is determined by the data, which are collected by the particular business, wanting to segment its customer base. The input keys are used to differentiate customers could be demographical (like age, gender, income), psychographic (lifestyle), geographic (geo location – where the customer lives) or behavioral (spending habits, product preferences).

Possible costumer segmentation techniques

There are many possible customer segmentation techniques. I will briefly explain some of the more common ones. The type of segmentation used will very on a lot of factors (goals, costs, etc.).

Figure 1 - graphical illustration on how usage of customer segmentation effects business revenue due to development of additional products, which is only one of many possible outcomes.(http://www.dobney.com/Research/segmentation.htm)

A priory segmentation

This is one of the simplest approaches of all, where the market is divided according to already existing segments (that is why a priory – “pre-existing”) such as gender or age. In some businesses this approach of segmentation could be sufficient, for example the technology sector has a strong relationship between age and usage or product preferences. In other sectors it might be more difficult to segment the customer base using this method. Without any doubt sing this method is better than pure mass marketing, however it is still quite crude.

Usage segmentation

Another rather simple approach of customer segmentation. There are two methods for usage segmentation: either the customers are divided based on their weight of use or by time and place of usage. With the first method it is obvious that customers who buy more are more important to the business that the other ones. In here for example a “Pareto analysis” could be used to identify the top 20 % of most valuable customers. This method is normally used in business-to-business markets. The second method divides customers based on time and place. At different times customers may want different products available, so this method takes that into consideration.

Needs based segmentation

Mostly, This approach uses co called “Conjoint Analysis”, which is and advanced research technique. (it also known as Discrete Choice Estimation) It gives customers choices and then analysis why they made them. The output of such analysis is a measurement of utility. Firstly, with conjoint analysis a product or service is divided into its constituent parts. Than the possible combinations of these parts are tested to find which combinations are preferred by the customers. Furthermore each part may have multiple attributes (for example memory has a size and frequency). These attributes are defined in levels a computer memory can for example have 4 GB or 8 GB and operate on 1600 or 2133 MHz. Attributes and levels are used to define products and one of the first steps of conjoint analysis is to define a set of product profiles, which represent choices for customers. The number of possible product profiles rapidly increases with each additional attribute therefore it is important to find balance between the number of attributes and complexity of customer choice, in order to get quality results. After this a range of statistical tools can be used to analyze which items customers choose or prefer from the product profiles. With need based segmentation the specific individual needs are identified and with the above approach they can be used to find other products, which meet the same customer requirements. Such products can then be offered to a particular customer by the seller.

Qualitative approach

An approach about why and for what they are (in this case customers) behaving the way they are. Usually the outcome of this method is not enough to base a statistic on, it rather focuses on getting several different surveying. The aim is to get the best picture about the market and the customers so that one could spot the gaps, the similarities and the differences and exploit them. Focus groups or individual depth interviews are very often used to clarify the behavior and attitude of people. Sometimes it might be a good idea to use another technique such as conflict groups, triads or paired interviews. Outcome of these techniques is, researcher (or a moderator) dependent that means he must be experienced in order to get the correct information out of the interview. It may happen that the participants may not be willing to speak their mind freely and will just repeat what they believe to be right. It is safe to say that a mind probe would make this approach a lot easier, unluckily this technology we do not possess. The discussion itself usually starts with a really broad term or subject and is later narrowed down by the moderator without putting too much pressure on the interviewed meanwhile getting an idea about how their beliefs and views of the subject changes throughout the session. The most valued facts about these participants are in the end their mood and their spontaneous reactions. Taking in count any information that is artificially created by the interviewees to cover up their true opinion is lowering the value of the session and of the finding itself. Though describing this technique defies its original purpose to be original every single time and it is contentious whether or not the results are valuable.

Clustering approach

This is one of the more advanced approaches, because it can be used to split customers based on the way they think or feel, rather than only who they are. This method start with individuals and forms segments with them based on similarities. At first it is necessary to choose one or more criteria, which will be used to assess the similarity between individuals (as already states it is common to use behavioral data with clustering). A good way to search for input data might be in surveys, which often focus on customer attitude with the business or product. If multiple criteria are in place, it might be necessary to standardize them so that they have comparable weight. Secondly it is important to choose clustering approach that will be used. There are two commonly applied approaches to clustering, partition clustering and hierarchical clustering.
With partition clustering, it is decided how many clusters will be (or better should be, because the result needn’t be possible) formed at the beginning. There are multiple methods of computation. One of the basic ones lies in specifying the “seed” observations for each one of the desired clusters. Than calculate the distance from each of the remaining observations to each seed and assign each observation to the nearest seed to form a set of clusters. This process can be repeated to reach greater homogeneity.
Hierarchical clustering is the second common approach and is based on assumption that closer objects are more related than the further ones. The algorithm begins by finding a most similar pair of observations in terms of previously defined criteria and joins them to form a group. The next part is to find another pair, which can be again joined together or joined to a previously formed group if the distance is shorter.
 

Figure 2 - graphical example of hierarchical clustering (http://people.revoledu.com/kardi/tutorial/Clustering/Numerical%20Example...)