Clustering Customers

Cluster Analysis Whereas AID starts with a population and splits it thin, cluster analysis usually starts with individuals and builds them into groups. Cluster analysis starts with details of more than 200 individuals. The measures can be demographic (as used in geodemographic segmentation), psyehographie (;is in life-

Error IS

10 -I 234 79586 Objects

"igurs 1 dendrogram showing cu;kter formation

A dendrogram shows the individual clusters and how easy it is to join them. In Figure 1 the cluster containing 1, 2, 3 and 10 forms early and so do the clusters containing 4, 7 and 9, and 5,6 and 8. These three clusters could become segments,

Marketing

Highlight 9.4

4. The process continues until measurements show that the individuals or clusters to be joined are not alike.

4. The process continues until measurements show that the individuals or clusters to be joined are not alike.

Cluster analysis is also different to AID in the way it examines al] the discriminating variables at once. Usually all the data gathered are on a set of uniform scales (say 1 to 7), used to represent demographic; attitude, or other dimensions.

1. Cluster analysis first looks at all the individuals and determines which two are most alike. Measures describe how alike the individuals are.

2. It then joins the most alike pair into a cluster that thus becomes a composite individual.

,1. Cluster analysis then looks for the next most alike pair and joins them. This could involve the eomposite cluster joining with one other individual.

look alike. If we try to force the three clusters to make two, there is a jump in 'error' as clusters 4, 7 and 9, and 5, 6 and 8 combine. The big jump suggests there are three natural segments. The individuals in the cluster are not exactly alike, but they may be close enough to be treated as segments.

Cluster analysis identified three benefit segments from 199 merit caters in the Netherlands. Table 1 gives some of the details and names the segments. Cluster analysis of the discriminating variables gave the segments. The descriptive variables were not used to find the clusters, but they show cluster differences and help target them.

The segments can help market meat products in the Netherlands. None of the segments like fatty meats, but the 'rural fat man' is not worried about fat, likes cheap cuts and is not looking for exclusivity. The 'urban quality seeker' is different. She wants quality, exclusiveness and no fat. She tends to live in northern towns and prefers steak to other cuts of meat.

Although cluster analysis is a simple process, a user has to answer several questions before using it. One question is: What is alike? A Porsche and a Trabant may be alike in size, but most people's attitude towards them is quite different. Another question is: What do you do when individuals join to make a new cluster? Do you take the average of them? These and other technical questions need answering by anyone using cluster analysis. Neglect these issues and G1GO (Garbage In, Garbage Out) rules. Cluster analysis is a powerful method that can produce convincing-looking" segments from random data. It also produces different results depending upon how the above questions are answered. The rules for its use are, therefore, test, test and re-test:

TABLE 1 THE USE OF CLUSTER ANALYSIS: THREE BENEFIT SEGMENTS IDENTIFIED AMONG MEAT EATERS

SEGMENT

RURAL FAT MAN URBAN QUALITY SEEKER ROUNDED MEAT EATER

TABLE 1 THE USE OF CLUSTER ANALYSIS: THREE BENEFIT SEGMENTS IDENTIFIED AMONG MEAT EATERS

SEGMENT

RURAL FAT MAN URBAN QUALITY SEEKER ROUNDED MEAT EATER

Cluster size (%)

35

41

24

Discriminating variables

Fatness

1.88 0.03

2.20 -0.88

1.42 -0.63

Bxelusiveness

-0.92

0.53

0.23

Convenience

0.55

0.59

0.45

Descriptive variables

Preferences

Sirloin steak

2.70

6.94

7.52

Pork belly steaks

7.81

4.17

5.56

Brisket beef steaks

6.16

4.93

Region

Residence

East Rural

North Urban

West

Gender

Male

Female

Conjoint analysis is a powerful tool that can measure the weight individuals put on the elements of a product or service. It often helps form customer segments. For example, Novotel could examine how much extra customers are willing to pay for a larger room, more expensive furnishings, a better TV in the room and so on. Sometimes it is called trade-off analysis because customers trade off one desirable feature against another; a king-sized bed versus a Teletext TV perhaps. Conjoint analysis examines the desires of individual customers. Often researchers use cluster analysis to combine these into segments.

SOURCES: John Saunders, 'Cluster analysis'. Journal of Marketing Management, 10, 1-3 (1994), pp. 1,1-28; Michel Wedel and Cor Kistemaker, 'Consumer benefit segmentation usin^ cUisterwisu linear regression*, International Journal nf Research in Marketing, i,, 1 (1989), pp. 45-59: Dick R. Witt ink, Marco Vriens and Wira Burhcime. 'Commercial use of conjoint analysis in Europe: results and critical reflection', International Jountai of Research in Marketing, 11, 1 (1994), pp. 73-S4.

1. Use well-proven methods.

2. See if the clusters are 'natural' by recreating them using different measures of alikcness.

3. Use some other data to see if the same clusters emerge from them.

4. Test the clusters practically to see if they do behave differently. This can sometimes be done using old data. Recently a bank was able to validate its segments by showing how they had responded differently to past sales promotions.

Factor Analysis

Faetor analysis is often used in conjunction with cluster analysis. It identifies correlated variables and can reduce their combined effect. Researchers often collect considerable psychographic and other data in segmentation studies and much of it is usually intercorrelated. For example, age, income, family size, size of house and debt are all interrelated for middle-class people. Factor analysis could combine them into a single factor called 'maturity'. This reduces the computational effort in clustering and prevents the results being biased towards groups of correlated variables.

77iis Smirnoff vodka ad leaves its target market in little doubt.

Insiders Guide For Making Money with Garage Sales

Insiders Guide For Making Money with Garage Sales

Someones Junk Is Someone Elses Treasure. Attending A Swap Meet, Garage Sales or Setting Up A Booth At A Flea Market Can Be Fun Exciting!

Get My Free Ebook


Post a comment