Market Segmentation, Cluster Analysis & Market Segmentation

The Maxwell Statistics Corporation provides full analytic services for market segmentation:

A caffeinated drink producer commissioned a market segmentation to address slowing sales. The segmentation revealed the client’s drink was succeeding in two market niches: Suburban Moms and Rave Teens.

Marketing felt that the two niches diminished success in each other’s niche. The Rave Teens would not want to be seen as drinking a Suburban Mom drink, and the Suburban Moms would be equally disapproving of the Rave Teens.

Market share was greater among the Suburban Moms. To address the conflicting niches, the client created a new packaging to establish a new brand aimed at the Rave Teens.

Promotions of the old brand focused on its role in providing an invigorating break for harried mothers driving between children’s’ activities.

Promotions of the new brand emphasized the drink keeping thin girls awake as they danced through the night.

Segmentation is the first step in a targeted promotional campaign. Segmentation reveals the market niches in the audience. Once the segments are known, promotions can be tailored to each segment’s interests and attitudes.

Segmentation finds groups that are as homogenous and distinct from one another as possible.

A market segmentation uses data to divide people into groups so that each group is as homogenous as possible and as distinct from all other groups as possible. The people in the resulting segments are very similar to others in their own segments, and different from people in the other segments.

Market segmentation considers data on multiple aspects of each person and finds groups with similar patterns. For example, research in the 1980s found a cluster of people who were females, parents, upper middle class, suburban, minivan drivers, who spent their weekends watching their kids play soccer. Those were the Soccer Moms.

To see how segmentation works, consider the households in the following hypothetical data.

This hypothetical data includes income, the number of recent elections voted in, and ratings of attitudes towards unions. In the raw data, it is hard to see the two groups.

Household	Income	Votes in Last 10 Elections	Union Attitude (1=Negative, 7=Positive)
A	$90,000	9	7
B	$100,000	8	5
C	$40,000	3	7
D	$50,000	2	5
E	$40,000	1	3
F	$100,000	9	1
G	$110,000	9	3
H	$30,000	1	1

Household	Income	Votes	Union Attitude
Low Income & Low Participation Segment
E	$40,000	1	3
H	$30,000	1	1
D	$50,000	2	5
C	$40,000	3	7
High Income & High Participation Segment
B	$100,000	8	5
A	$90,000	9	7
F	$100,000	9	1
G	$110,000	9	3

In this data, there are two segments: wealthier households that vote more often, and less wealthy households that vote less often. (In this hypothetical data, attitudes towards unions do not help distinguish the two groups.)

To get a sense of how the segmentation works, consider a chart of the data that was segmented above:

A look at the chart reveals the clusters that the segmentation pulled out of the data. The wealthier high-voters are in the upper right corner. The other segment is in the lower left corner.

This hypothetical data is simple enough that one could segment it without advanced analysis: there are three measurements and eight people. In a segmentation for a typical market research study, there are typically 10 to 50 measurements and 400 to 4,000 respondents. This data is provided only to clarify what the segmentation does.

What else should I know when I’m talking with clients about market segmentation?

Segmentation reveals which measurements distinguish between the groups and which measurements are not related to group membership. For example, in the hypothetical data segmented above, each segment is spread evenly across the range of attitudes towards unions. A pattern like this reveals that attitudes towards unions do not distinguish between the groups.

The following chart shows the same eight households along with four new households. The new households fall between the previously found segments. This does not change the fact that there is a group of targetable high-election-participation/high-income families. Or that there is a group of targetable low-participation/low income families.

When many respondent features are included in the segmentation analysis, segments tend to look less homogenous on any given feature, even though the segments are just as homogenous overall. For example, the Soccer Mom segment in a study might have only 80% driving minivans, and 75% with soccer-playing kids. To see why this happens, consider a middle-aged woman living in the suburbs who drives her kids to soccer games in a Volvo station wagon. Is this a soccer mom? You bet. How about her neighbor who drives a minivan and who is driving her kids to Youth symphony rather than soccer? The neighbor has just as many features in common with the Soccer Mom segment, and gets categorized there as well.

Segmentation can be done on pre-existing data or data collected for the segmentation

Segmentation can be done on data collected specifically for the segmentation or on pre-existing data. A common approach for segmentation is to ask respondents for their “AIOs” (Attitudes, Interests, and Opinions). Another fruitful approach is to ask about media outlets (e.g., what television shows they watch, and what printed press they read). Asking about media outlets reveals a lot about the nature of the segments, and how to reach them.

If a client asks for a segmentation, the key questions are what are the dimensions that will produce segments that are most useful. For example, a PR firm might be more comfortable with single-gender segments, feeling they can see more clearly how to reach the target if they know the target’s gender. In that case, gender is very important and its importance can be reflected in the segmentation.

As with any research, the segmentation cannot serve well any purpose that requires information that is not collected. For example, if the client expects the segments to reveal where to advertise, it is best to ask about media outlets in the survey and enter the outlets in the segmentation.

If a client has not asked for a segmentation, you can suggest that a segmentation analysis be added to a study. Getting a basic segmentation of 10 to 30 questions from single survey usually costs roughly $1,000 to $1,500. Adding such a segment would serve mostly to inform the client about new possibilities, and would not be as reliably actionable as a study that was created from the outset for a market segmentation.

To allow new clients to learn about market segmentation, the Maxwell Statistics Corporation offers each new client a first segmentation analysis at a 50% discount.

The Maxwell Statistics Corporation works with a four-day turnaround for most analyses.