Market Segmentation

From PDMA Knowledge Wiki

Author(s)

Vince Farace, Satisfaction Management Systems (SMS) (www.satmansys.com)

Definition

Market segmentation is defined as a framework by which to sub-divide a larger heterogeneous market into smaller, more homogeneous parts. These segments can be defined in many different ways: demographic (men vs. women, young vs. old, or richer vs. poorer), behavioral (those who buy on the phone vs. the internet vs. retail, or those who pay with cash vs. credit cards), or attitudinal (those who believe that store brands are just as good as national brands vs. those who don't). There are many analytical techniques used to identify segments such as cluster analysis, factor analysis, or discriminate analysis. But the most common method is simply to hypothesize a potential segmentatio n definition and then to test whether any differences that are observed are statistically significant (See Chapter 13 of The PDMA HandBook 2nd Edition).

Description

The simplest and most effective way to manage an organization is to deliver one product or service that meets the needs of one type of customer. However, to the delight of many organizations over the years, broadening one’s product or service portfolio by appealing to new customers and/or expanding the portfolio’s success with current customers can lead to profitable business growth.

Consequently, most organizations are on the lookout for new products and services, or variations on current ones, that meet different needs or wants of current customers or which bring new customers into the fold. Market segmentation is a set of concepts and tools that can guide management thinking and lead to new profitable product or service offerings. Segmentation provides a rationale and starting point for New Product Development.

When a company moves beyond one product or service offered to one type of customer, it has started down the path of segmentation. If the product traditionally appeals to women, perhaps a variation will appeal specifically to men, thus segmenting the customer base and making it possible to increase the customer base and generate greater total revenue.

If a product appeals to young adults, perhaps a variation will appeal to ‘tweeners. There may be a product that appeals to ‘tweener males vs. ‘tweener females. Again, the result should be more customers and more revenue.

What began as segmentation based solely on gender, then, can be extended to segmentation based on many, many factors. The key is to find a clear-cut homogenous group of customers or potential customers whose demand for a uniquely configured product or service supports a sustainable market opportunity.

The rationale for segmenting is that customers within segments typically have more in common with each other than with customers in the remaining segments. This commonality helps focus new product development and marketing efforts toward that specific segment. Basically, customers who share similar characteristics on key defining attributes are presumed to have similar needs, and this similarity allows development of specialized products and/or messages that uniquely address those needs.

To become a strategic contributor to an organization, a segmentation initiative should accomplish some or all of the following outcomes – it should:

• Show which segments are likely to be most profitable

• Focus efforts to produce successful products and services for those segments

• Identify prospects to convert into customers of the new segment

• Point out current customers who need rescuing before they defect and who may be converted to the new segment

• Highlight up-sell and cross-sell opportunities with customers who might continue their current purchasing and branch into the new segment

• Show where current resources can be redirected

The Basics of Market Segmentation

To segment a market, two key questions must be answered: (a) what variables are most important to a successful segmentation, and (b) how many segments should we pursue?

Selecting Segmentation Variables

Segmentation need not be complicated – one or two variables can be adequate, like the ‘tweener males vs. females example above. In reality, though, in today’s global economy, simple segmentation may not be sufficient to capture the full market potential that exists, leaving “money on the table” for another organization to exploit. It’s often necessary to use multiple variables as the basis for segmentation.

The three sources of potential variables for segmentation are known as:

• Psychographic – “inside the head’ (but measurable) factors such as the individual’s opinions, wants, needs, aspirations, and thoughts.

• Demographic – observable characteristics of the individual, such as age, education, income, gender, political orientation and the like.

• Firmographics – characteristics of the household (number of members, number of children, presence of parents, etc.) or organization (employees, revenue, patents held, organizational structure, etc.)

This first step in segmentation – identifying which variables to include – is absolutely critical to a successful initiative. No amount of sophisticated statistical analysis will rescue a segmentation study based on irrelevant issues. Perhaps the least risky approach to deciding on segmentation variables is to triangulate a number of different sources of information:

• Management perception of business opportunities – use insight provided by the people who know the most about the business and its market opportunities

• Internal development initiatives – translate internal capabilities into new possibilities

• Existing/new research – integrate primary and secondary research, update it with new research if needed and add the results to the mix

• Industry experts – capitalize on their knowledge about industry as a whole

• Competitive intelligence – keep abreast of the competition

Is there one “best” approach? Unfortunately, no. Relying solely on management guidance is potentially problematic because it explicitly ignores what the market is saying, and as such can lead the analysis in an unproductive direction. On the other hand, blind reliance on statistical modeling procedures can be inherently conservative. There is a tendency to focus on past trends and to allow them to drive the analysis to a great degree. If the market is undergoing (or about to undergo) a revolutionary change, then the statistical modeling approach may miss opportunities that an insightful management may anticipate.

Regardless, for each of these approaches the 1st selection of variables need not necessarily be the last. In particular, subsequent analysis may well indicate — in fact, it usually does — that some of the measures selected for use in the segmentation model really add little value and simply confound and confuse the model.

Choosing Segments to Pursue

Before embarking on a segmentation initiative, it is often wise to contemplate how the organization would go about deciding how many new segments to pursue. To fulfill the promise of a new segment, there are product development, production, distribution, marketing, pricing, sales, service and other issues that have to be worked effectively to succeed with a new segment. So, the analysis may identify more segments than the organization can actually pursue. The analysis should include results describing which of the final segments are most worthwhile to the organization. Selecting one to three new segments is typically a practical outcome.

Analyzing Segmentation Variables

Once there is at least a preliminary list of segmentation variables defined, it is time to begin thinking about the analytical tools that will be brought to bear on the data.

There are two broad classes of statistical models that are used in segmentation models, and within each of these classes there are numerous specific algorithms that yield different results. The two primary statistical approaches to developing segmentation models are:

Hierarchical Methods

Hierarchical methods are used to develop segments from the ground up. A matrix is calculated that determines how similar each respondent in the data file is to every other respondent. The most similar respondents are grouped together, the similarity matrix is re-calculated, and the next-most-similar respondents are joined. This process continues until all respondents are joined together.

At each stage, diagnostic data are presented that help to determine if this stage constitutes a good stopping point in the segmentation process.

While this sounds quite simple in theory, there are over 20 different statistical algorithms — 11 of which are quite commonly available — that are used in the hierarchical segmentation models to achieve this general goal. (See Table 1 for a description of the different hierarchical clustering algorithms.)

Table 1: Hierarchical Clustering Methods

One of the key benefits of the hierarchical models is the fact that assorted diagnostic data are provided to help inform decisions regarding the proper number of segments to retain for the segmentation model, but even with these diagnostics there are no clear-cut unambiguous rules. A key limitation of all of the hierarchical segmentation models — more true of some of the methods than others — is that they can produce segments which in multidimensional space resemble elongated chains as opposed to the typically desired tight, spheroidal clusters.

Iterative Centroid Methods

Iterative centroid methods develop segments in a more deductive fashion. Given a set of variables and a specification of the number of segments to retain, a set of initial starting points or centroids equal to the number of desired segments is defined for each measure. Respondents are then added to the segment that they are most similar to, the centroids are re-calculated, and respondents are re-assigned to the new centroid that best characterizes their pattern of scores on the measures. This process keeps repeating (iterating) until no respondents change segments, and the centroids stop drifting.

There are 7 major variations in the iterative centroid methods – three of which control the vast majority of uses – available for the analysis. See Table 2 for a description of these methods.)

Table 2: Iterative Clustering Methods

The strength of this approach is that it tends to produce the tight, spheroidal segments that are typically preferred for subsequent analysis. The primary limitation is that it requires a priori determination of the number of segments, and provides few diagnostics that help to determine the number of segments that should be used in the analysis.

Market Segmentation in Action

Data Collection

With the preliminary list of segment variables in hand, it’s time to assemble customer and prospect data on each of these variables. These data typically come from some combination of three sources:

• Internal company databases – often a gold mine of sales, purchase, warranty, length of time a customer, etc., information

• Purchased secondary data – for example, if households are the focus, then there is abundant U.S. Census data. There are business-oriented databases across a wide set of topics.

• Primary data – new research. Data that aren’t available from the first two sources but which are critical the segmentation can usually be gathered through a survey process. In particular, psychographic information is often gathered through primary research.

Analysis Process

With the data set in hand, the next steps are:

• Finalize the key outcome criterion or criteria, and develop multivariate models that identify the key drivers that characterize the data.

• Review the key driver data with the client and use this information to develop measures for the segmentation.

• Review the types of data available on the proposed measures, and select 2-3 different hierarchical clustering algorithms that appear best suited for the project and data at hand.

• For each of the 2-3 hierarchical clustering algorithms, perform a Monte Carlo probability simulation study to determine the most likely number of segments that the analysis suggests. Typically, a small range in number of segments will be tested.

• If necessary, select 2-3 of the different iterative centroid methods and test alternative models representing each of the different solutions for the number of clusters, as determined by the previously completed Monte Carlo probability study. Note that at this stage it is not necessary to test alternative algorithms since the data will usually be quite clear as to the best approach to use.

• Review the findings of the iterative centroid models and, if necessary refine by dropping measures that are clearly random and do not contribute to the segmentation structure.

Conclusion

The end result of a useful segmentation was mentioned above but bears repeating – a segmentation initiative should:

• Show which segments are likely to be most profitable

• Focus efforts to produce successful products and services

• Identify prospects to convert into customers of the new segment(s)

• Point out current customers who need rescuing before they defect and who may be converted to the new segment(s)

• Highlight up-sell and cross-sell opportunities with customers who might continue their current purchasing and branch into the new segment(s)

• Show where current resources can be redirected

References

Anderberg, M.R. (1973), Cluster Analysis for Applications, New York: Academic Press, Inc.

Hartigan, J.A. (1975), Clustering Algorithms, New York: John Wiley & Sons, Inc. The Hartigan reference also discusses AID and CHAID. More recent work has focused on some of the ramifications of the different models.

Milligan, G.W. (1980), "An Examination of the Effect of Six Types of Error Perturbation on Fifteen Clustering Algorithms," Psychometrika, 45, 325 - 342.