The general APL model described in the last section must first be customized to the firm in question before the full benefits of the approach are realizable.
Indeed, the process of customizing the model and exploring the hypothesized linkages using the firm’s data can be as valuable, if not more so, than using the final model to make predictions of profit impact, because it helps focus managers’ attention on relationships between key performance metrics.
Customizing the general APL model is a multi-phase process that for maximum benefit should involve a wide cross-section of the firm’s senior management team. We next describe the phases in the APL model customization process.
Phase 1: Hypothesized Linkages
The first phase of developing customized APL models involves the specifica- tion of an initial set of hypothesized causal linkages based on the general APL model described above, but modified by management experience and intuition.
Also helpful to consider are the observations of employees, as well as those of customers, each of whom might help establish expected linkages for the vari- ables that concern their own behavior. The customized model that is specified should include linkages between specific variables rather than between the groups of variables shown on Fig. 1.
Phase 2: Measurement and Data Collection
In the second phase of model customization, data are collected to test the validity of the initial hypothesized model and make revisions as the empirical evidence indicates. Prior to data collection, however, metrics must be defined for each 1
2 3 4 5 6 7 8 9 1011 11 12 13 14 15 16 17 18 19 2011 21 22 23 24 25 26 27 28 29 3011 31 32 33 34 35 36 37 38 39 40
60 MARC J. EPSTEIN, PIYUSH KUMAR AND ROBERT A. WESTBROOK
12345678910111112131415161718192011212223242526272829301131323334353637383940 The Drivers of Customer and Corporate Profitability61
61
Table 1. Measurement of APL Model Variables
Domain Variable Illustrative Measures Data Source
Firm Actions Operations Use of process design alternative A,B,C, etc. All obtained from company
Human Resources No. and type of training programs records
Marketing & Sales Local vs. centralized customer service Finance & Accounting Debt to equity ratio
External No. of lobbyists; public relations budget
Delivered Product Product/Service Characteristics Failure rate; performance specifications Company testing records or Service Employee Actions Response time to fulfill customer orders Mystery shopper study
Price Schedule Ratio of price to average of competitors Market survey Customer Communications No. of ad exposures per month Reader/Viewer survey Customer Actions Perceptions (All based on 7- or 10-point rating scales) All obtained from custom
Product/Service Service dependability or syndicated survey
Price Reasonableness of price; frequency of discount research on a representative Buying Process Ease of ordering; on-time delivery cross-sectional sample Relationship Able to rely on sales rep. expertise of customers and Brand/Co. Image Industry leader; innovativeness prospective customers Attitudes
New Customer Attraction Purchase intention rating scale (5- or 11-point) All obtained from custom or Customer Satisfaction Overall satisfaction rating scale (7- or 10- point) syndicated customer surveys Overt Customer Behavior
Initial Purchase % Market who have purchased at least once All obtained from custom or Account Share % All customer purchases of product/svc. per period syndicated tracking survey
Repurchase % Customers making repurchase of all customers in market
New Bus. Referrals % Customers making a referral; avg. no. referrals each Price Acceptance Average of prices paid by customer
Economic Impact Cost of Action
Customer Revenues $ expense of action or investment cost General ledger
$ sales to specific customers per period Historical account records
of the variables to comprise the customized model. Table 1 provides a guide to appropriate measures and possible data collection methods for each of the groups of variables in the model.
Delivered Product or Service. As shown in Table 1, several measures and data collection methods are useful for quantifying variables within the Delivered Product/Service. Objective measures of service quality, such as failure rates or response times, can be obtained from both company records and scientific obser- vation. Mystery shopping can also be used to objectively measure specific aspects of employees’ behavior, such as courtesy and adherence to service stan- dards, using components of service quality scales (Parasuraman, Zeithaml &
Berry, 1988). Measures of price schedules can be developed based on the firm’s own pricing data and the data on competitors’ pricing. Finally, firms can quan- tify their customer communications based on their advertising expenditures and scores from their advertising tests.
Customer Attitudes and Perceptions. Table 1 provides an illustrative listing of customer attitudes and perceptions, their measures, and relevant data collection methodologies. As the information in the table illustrates, while attitudes and perceptions drive overt behavior, they cannot be measured based on behavior.
Most of these variables, such as customers’ post-purchase satisfaction, will be measured using standard multi-item scales administered to samples of customers in surveys. If the firm does not have suitable measures on hand in its research library, it can either undertake its own research to obtain them or purchase them from syndicated data sources that have already gathered suitable information.
Customer Actions. Table 1 also contains a partial list of customer actions, their measures and possible data collection methods. Firms with a narrow customer base will typically have internal data pertaining to the purchase records of each of their customers. Inferences regarding purchase frequency, the mix of prod- ucts purchased, and the prices paid, can be drawn from these data. For firms with a broad customer base, these data can either be drawn from standardized sources of syndicated data or be collected by the firm itself. For such firms even data collected from a small representative sample of customers might be sufficient to understand the action-profit links. These data will typically be divided according to customer segments.
Costs. Finally, for purposes of calibrating APL models it is important to iden- tify not just marketing costs but all costs related to firm actions. Activity-based costing can then be used to assign these costs to each customer segment.
1 2 3 4 5 6 7 8 9 1011 11 12 13 14 15 16 17 18 19 2011 21 22 23 24 25 26 27 28 29 3011 31 32 33 34 35 36 37 38 39 40
62 MARC J. EPSTEIN, PIYUSH KUMAR AND ROBERT A. WESTBROOK
Unit of Analysis. An important issue in the data-gathering phase is establishing the appropriate unit of analysis for the variables in the model. For example, in the case of a bank, should customer survey data be gathered for analysis at the level of individual customers, office branches, metropolitan areas, states, or the company as a whole over time? For some linkages, e.g. between customer perceptions and satisfaction, the proper unit of analysis will be the individual customer, and the large customer base permits the use of sampling of customers for both measures, i.e. a typical survey. For others, such as the linkage between Delivered Product/Service metrics and Customer Perceptions, the data on the former may not be available for individual customers. Instead, the unit of analysis may have to be the bank office branch, because that is the most plen- tiful unit at which both types of data may be available. If so, the customer survey data measuring customer perceptions will have to be aggregated to a branch average, and then related to the appropriate measure of the Delivered Product/Service for the branch (possibly also by aggregating employee charac- teristics among other measures). In some instances, the unit of analysis may have to be the company as a whole, such as for relating Customer Satisfaction to Customer Retention, in which case both measures would be gathered for a number of succeeding time periods, e.g. months, quarters, or even years. Annual measures at the level of the company as whole are least desirable, since they require either extensive past data, or delay while the needed information is gath- ered annually. In general, APL models employ multiple units of analysis to make the best use of available information.
Phase 3: Assessment of Relationships
In the third phase of model customization, an analysis is made of the data gath- ered during the preceding phase for each of the model variables. The analytic methods of choice are simple and multiple regression analysis, since they provide quantitative estimates of the strength of the linkages in the model, depending on the number of predictor variables (one for simple regression, two or more for multiple regression). Regression results are obtained for each set of variables that have been hypothesized to be related. For example, each of the variables in the variable group from which an arrow emanates in the model can serve as an independent variable in the regression analysis, while each vari- able receiving an arrow is the dependent variable.
The assessment of goodness of fit of the proposed model to the data is an important step. If the goodness of fit is found to be poor, the model development effort must return to Phase 1, in order to identify key variables missing from the model. The regression analysis will directly indicate which of the predictors is 1
2 3 4 5 6 7 8 9 1011 11 12 13 14 15 16 17 18 19 2011 21 22 23 24 25 26 27 28 29 3011 31 32 33 34 35 36 37 38 39 40
The Drivers of Customer and Corporate Profitability 63
most strongly related to each dependent variable in the model. In general, the variables with the strongest relationships are retained for interpretation and appli- cation. As the initial model is evaluated empirically in this fashion, some link- ages will be dropped for lack of evidence of strong relationship, while others may need to be added, depending on the success of the regression analyses.
Ultimately a final model will emerge, and consideration can turn to application of the model to supporting management decision-making.
Phase 4: Monitoring
The fourth and final phase of model customization consists of monitoring the APL model developed in the previous phase for any changes in the relation- ships between variables that indicate new or altered linkages. External factors, such as increased customer expectations or the evolution of the competitive context, as well as internal factors, such as changes in the morale of the firm’s labor force, may require revisions in the APL model that has been developed.
The changed circumstances may indicate the need to: (1) add or drop variables to the model, (2) add or drop linkages between variables, or (3) modify the estimated strength of existing linkages, all of which have the potential to alter the profit impact of a particular firm action.
For example, consider a firm whose customized APL model includes the following linkages: increasing the incentive compensation offered to the cleaning staff at a hotel (Firm Action) might improve the care with which employees clean guest rooms (Delivered Product/Service). In turn, this improvement increases the perceived cleanliness of guestrooms by customers, and drives in higher customer satisfaction, retention of customers, and customer revenues. Since the incremen- tal revenues exceed the added compensation costs, corporate profitability is increased as well. Over time, however, additional linkages emerge: higher levels of customer satisfaction (Customer Actions) lead to improved interactions between customers and customer-contact employees (another facet of the Delivered Product/Service), which increase the number of compliments about the staff. Employee job satisfaction is increased by the expressions of customer support, and the turnover of employees is reduced, lowering personnel hiring and training costs. These added linkages that emerge over time have thus amplified the profitability of the original action of increasing incentive compensation.