Analysis Capabilities

DECISION TREE ANALYSIS

Determining the specific cross tabulations that reveal differences among respondents can be very challenging. Thousands of combinations can be generated even with a small amount of data. A statistical technique called decision tree (CHAID) analysis is used to implement an orderly, systematic and statistically valid methodology to analyze cross tabulation data. All of the variables are considered simultaneously while categories are collapsed and combined, if necessary, to explain the most important relationships. The procedure pinpoints the critical associations among the thousands of possibilities that might be investigated.

Customer loyalty is critical because the expense of obtaining a new customer is usually much greater than the cost of keeping a current customer satisfied. The following example illustrates how decision tree (CHAID) analysis was used to develop a profile of customers most likely to switch banks. The research was based on telephone interviews with 250 randomly selected bank customers. Although the example concerns consumer research, the principles are equally applicable in the business-to-business environment. The variables used to segment customers include company size, industry, sales volume, length of time in business and number of employees.

The "dependent variable" consists of responses to the question: "How likely are you to switch banks in the next six months? Are you very likely, somewhat likely, not too likely or not at all likely?" The independent variables are comprised of three types of information: demographic data (age, gender, education, household income, marital status and race), satisfaction indicators (satisfaction with current bank, last time respondent switched banks) and the products used by the respondent (mortgage loan, checking account, savings account and IRA).

The decision tree associated with this application is illustrated on the following page. The decision tree contains five levels, each of which provides information regarding a customer's probability of switching banks. An explanation of the discoveries on each level follows.

Level 1

The top of the tree depicts the overall statistics for the bank-switching question (the dependent variable). For example, 3.6 percent of respondents (9 of the 250 respondents in the sample) are very likely to switch banks.

Level 2

The second level of the tree indicates age is a statistically significant indicator of likelihood to switch. Age categories that are not statistically different are combined. For example, the three age groups from 25 to 55 were consolidated into one category. Respondents younger 25 and older than 55 were combined to form a second category. Within the group composed of the older and younger respondents, 83.6 percent are not at all likely to switch. In contrast, only 56.7 percent of the group between the ages of 25 and 55 are not at all likely to switch.

Level 3

The third level indicates that among the age group more likely to switch (25 to 55), those who have switched banks within the past one to two years are more likely to switch again when compared with respondents who have not switched in the past one to two years (1.8 percent vs. 23.8 percent).

Level 4

The fourth level portrays a very important concept in analyzing customer satisfaction information. Respondents who are very satisfied with a company, product or service (in this case, the current bank) form a group distinct from all other satisfaction categories. Respondents who had switched banks within the past one to two years, but are now very satisfied with their current bank, are not very likely to switch again (0 percent). Among all the other respondents who had switched banks, half are very likely to switch again - and 30 percent are somewhat likely to switch. Consequently, four out of five customers are vulnerable to switching. These results are statistically significant at the 95 percent confidence level - even with sample sizes as small as 10 and 11 respondents. This suggests that loyalty is related to "very satisfied" customers - not just "somewhat satisfied". A customer loyalty measurement will be extremely overstated if the "very satisfied" and "somewhat satisfied" customers are consolidated to form a "top box".

Level 5

The final level of the tree indicates respondents in the 25 to 55 age group who have switched banks in the last one to two years and are less than very satisfied with the current bank can be further distinguished between those with and without a mortgage at the bank. The percentage of respondents who are very likely to switch ranges from 0 percent (customers with mortgages) to 62.5 percent (customers without mortgages).

Although not shown, summary statistics are computed to determine the statistical significance of each level of the tree. For example, age is statistically related to the likelihood of switching banks with a level of confidence of 99.4 percent. Even the final level of the tree structure, comparing sample sizes of only two and eight, provides a statistically significant relationship with a level of confidence of 98.1 percent.

In summary, the decision tree analysis has identified a profile of a typical customer who is very likely to switch to another bank: age 25 and 55, switched banks within the past one to two years, not completely satisfied with the current bank and without a mortgage at the current bank.

In this application, age was the variable most statistically related to likelihood of switching banks. Additional profiles can be explored by investigating other "branches" of the decision tree (including the demographic variables gender, education, household income, marital status and race, and product variables including checking, savings and IRA accounts). The statistical significance of each variable can be assessed relative to likelihood of switching banks. An important aspect of the complete analysis is identifying variables that are not significant factors in profiling customers who are likely to switch.