Forging links

Editor’s note: Kyle Lundby is director, consumer and technology research, and Carrie Christianson DeMay is research consultant, at Data Recognition Corporation, Maple Grove, Minn.

In today’s increasingly competitive business climate, organizations are searching for low-cost but effective means to improve customer loyalty and profitability. However, determining the key drivers of loyalty and profitability remains a significant challenge for many organizations. In other words, if the goal is to improve customer loyalty and profitability, how do you determine where to focus time and energy in order to have the greatest impact? One solution is linkage research - a method that uses statistical modeling to identify relationships (i.e., identify the links) among different organizational metrics. For example, a linkage model for a retail store with locations across the United States might show how employee attitudes relate to customer attitudes (overall satisfaction, willingness to recommend, etc.) or other more tangible outcomes (return visits, share of wallet, etc.).

Linkage research can also tie employee and customer measures to bottom-line financial outcomes such as sales growth or other profitability measures. For example, the CEO of an automobile company may have a hunch that customers who interact with highly service-oriented, well-dressed and knowledgeable sales associates are more likely to purchase a vehicle than customers who interact with sales associates who have fewer of these characteristics. In linking terms, the CEO believes that the attitudes (service orientation) skills (knowledge of product) and appearance of associates at individual dealerships impacts the experience of customers (as indicated by their level of satisfaction) and their purchase decision.

Linkage researchers test these hunches by building statistical models that represent the different measures (e.g., employee attitudes and behaviors, customer attitudes and behaviors, and financials) and the strength of their interrelationships. In the car company example, linkage analyses may reveal that service orientation and associate knowledge is more important than appearance in determining customer satisfaction and their decision to purchase or not purchase a new vehicle. For organizations engaged in linkage research, the real value comes from the relationships that are identified in these models. In essence, they are roadmaps of cause and effect and suggest specific areas to target for improvement. In the car company example, emphasizing service quality and associate training should have a greater impact on customer satisfaction and sales growth than worrying about the appearance of sales associates. Thus, linkage research helps organizations determine where to focus their efforts to achieve desired outcomes. Rather than relying solely on hunches or intuition, which may be right only part of the time, decisions can be based on data and relationships between the measures that really count.

An overview of linkage research

Although there is really no one best way to conduct a linkage study, there are some fairly common steps. However, before the first step is taken, it is important to consider the following six linking-readiness factors:

1. What are the objectives of the study and what is the level of commitment within the organization? Is the objective to decrease employee turnover, improve customer satisfaction and loyalty, increase company profits, all of the above, or something completely different? As with any project, if the objectives are not fully understood, linkage research can be a time-consuming effort. Organizations also need to understand that linking research represents a different way of doing business and they must be committed to this path. For example, in many companies, employee opinion data, customer opinion data, and financials have always been collected, but not integrated. Linkage research brings these and other appropriate data together in one overall model. It also takes time and cooperation among groups who may not be accustomed to working together (e.g., finance, human resources, marketing). If a company is not prepared to integrate these different sources of information and committed to the long haul, it may not be worth expending the effort. However, if the objectives are clear and the commitment is there, the rewards of linkage research can be significant.

2. Do you have sufficient internal expertise to conduct a linkage study? Although linkage research makes sense, the statistical analyses it uses are somewhat complex and require a good understanding of statistical principles, research methods and the ability to use different analytical tools (e.g., correlation, regression, structural equations modeling). It also requires knowledge of some less common techniques such as aggregating data, combining multiple files and evaluating within-group agreement.

3. Are you familiar with the history and current best practices in linkage research? Although you don’t need to be aware of every linkage study that has been published or presented, some familiarity with this line of research will be helpful. If for no other reason, it will save time and effort to know what has worked, and not worked, in the past and under what conditions.

4. How complex is your linking model likely to be? One of the first steps in any linkage study is to create a conceptual model representing your best informed guess of what causes what. The conceptual model should depict the important issues (e.g., customer attitudes, employee attitudes) and how you believe they are related to one another. Linkage models can be simple (depicting only two variables) or complex (involving many variables measuring different aspects of customer and employee attitudes over multiple points in time). Of course, as the complexity of the model increases, so does the complexity of the statistical analyses.

5. Do data exist for all of the variables in your model? If the goal is to determine which employee attitudes impact customer satisfaction, it will be necessary to have both employee and customer opinion data. In reality, organizations rarely have measures for all of the variables they would like to include. If the missing variables are “nice to haves” rather than “need to haves,” they can be left out. However, if they are central features of the model, new measures will need to be created and data will need to be collected.

6. What will be your unit of analysis? In addition to developing a conceptual model of what causes what, it is important to think about the unit of analysis that will be depicted in the model. Although most linkage studies have used locations as the unit of analysis, it is possible to use other groupings, such as functional areas, job types, geographic region, or some other meaningful unit. Considering the unit of analysis is important for several reasons:

  • The unit of analysis is where actions are likely to be taken. For example, if the service orientation of personnel in an automobile dealership contributes significantly to the service and purchase experience of customers, the next step would be to identify dealerships where employee service orientation falls below some threshold and take appropriate actions within those dealerships.
  • The unit is the level at which data are analyzed. A requirement of linkage research is that the data be analyzed at a common level. In the car company example, employee and customer opinions were collected from individuals but they will need to be rolled up or aggregated to the dealership level for analysis. Once chosen, this common denominator (dealership in this case) is used to tie the different sources of data together in one large data file.
  • There must be a sufficient number of units to conduct the analyses. Although the lower limit depends on a number of factors (e.g., quality of data, whether datapoints exist over multiple points in time), more locations are generally preferable.

Moving forward

After considering these readiness factors and making the decision to move forward, the following seven steps are commonly carried out in a given linkage study.

1. Form a linking team - In contrast to traditional survey research programs, linkage research requires collaboration among organizational groups who may not be accustomed to working together. For example, in many organizations, employee opinion surveys are conducted by human resources while customer surveys are conducted by marketing research. Because linking uses employee and customer data, these two groups must work together. Finance is another group that may not be accustomed to working with HR or marketing from a survey research standpoint. Forming a team representing these different areas facilitates the process of building and then carrying out the research. This cross-functional team approach can be helpful in several ways. It ensures that all groups are represented and often leads to a better-integrated model. It can also help build buy-in and commitment to the project, and can greatly facilitate the data gathering process.

2. Develop a conceptual model - Building a conceptual model is one of the most important pieces of any linkage study. The conceptual model is essentially the team’s view of what happens in their organization. The model is typically based on each team member’s practical experience and intuition, as well as any pre-existing research. This model represents the team’s initial prediction of what causes what and will be modified accordingly after the data are analyzed.

Forming a conceptual model before analyses are conducted brings with it three key benefits. It forces the team to look at the organization as a system and consider what really is important for the organization. It ensures that the team takes stock of which data are available and which are not. It prevents the team from charging forward and analyzing all data just because they can, or only those pieces of information that are easy to obtain. Just because certain pieces of information are not currently available does not mean they should be excluded. Alternatively, just because the data are available, does not mean they should be included.

3. Gather data and aggregate to appropriate unit of analysis - Once the conceptual model has been developed, the linking team must obtain data for each variable in the model. If the model depicts a link between customer opinions of auto dealer service and actual number of cars sold, data will need to be obtained for each of these variables. However, given that these data represent different levels of analysis (i.e., opinions are recorded from individuals; cars sold is recorded as the number for each dealership per month), they will need to be rolled up or aggregated to the same level. In this case, it means that the customer opinions will need to be averaged together to create an overall customer opinion score for each dealership. Once aggregated to the dealership level, the customer opinion data can be merged with sales data to form one file. Combining data for all variables into one file is necessary for conducting the statistical analyses.

Table 1

4. Conduct the analyses - With all of the data at the same level and in one file, the next step is to conduct the analyses. Several different procedures have been used by linkage researchers - correlation, multiple regression, structural equations modeling, relative importance. The choice of procedure depends on several factors, including the quality of the data, the quantity of the data (i.e., the number of units), as well as the complexity of the model and the linking team’s comfort with the various statistical procedures. Table 1 provides a brief summary of some of the strengths and weaknesses of different approaches. For example, correlation works well with only two variables but it is less appropriate to use to develop more complex models. While structural equations modeling can deal with many variables, it is more difficult to explain and to run. Multiple regression is easier to explain, but it does not work well when variables are highly correlated with one another. While each method has advantages and disadvantages, the ultimate objective remains the same - to determine the strength of association (i.e., the links) between the variables in the model.

5. Interpret the results - With the analyses complete, the next step is to display the values for each link. Figure 1 displays an example of this for a chain of retail stores. At the left are four dimensions from an employee opinion survey (service orientation, engagement, satisfaction and communication). Customer perceptions of service quality are in the middle and outcomes (sales growth and customer retention) are to the right. Solid lines indicate a strong relationship while dashed lines indicate a weak or non-existent one. Of the employee attitude variables, it can be seen that service orientation and engagement have the greatest impact on customer perceptions of service quality. Customer perceptions of service quality, in turn, have a significant effect on both outcomes. Taken together, this model suggests that improving employee opinions regarding service orientation and engagement should have a direct effect on customer attitudes, which will ultimately improve sales growth and return visits.

Figure 1

6. Take action - After the analyses are conducted and the results displayed in the model, the next step is to decide how and where to focus attention. According to Figure 1, it would appear that customer satisfaction, sales growth and retention can be improved by increasing employee engagement and service orientation. The next step is to determine if engagement and service orientation are problem areas for the entire organization or for particular locations. If all store locations received low scores on both variables, then the organization must take action system-wide. However, if some locations had high scores on these variables and others low scores, action is really only needed in those locations where scores are low. If, on the other hand, engagement and service orientation scores are high for all stores, the next step would be to take a closer look at scores on the other variables in the model (satisfaction and communication) for each store location. While satisfaction and communication had a weaker effect on customer satisfaction, they still exerted some influence.

Whether a model is simple or complex, the team should follow this same general process of first looking for the critical links, then examining current scores, then determining how widespread the issue is. If the first pass yields no needed action (all store locations are doing well on the most important variables - service orientation and engagement), the team should then drill down to the next most important variables and continue this until they determine where to focus attention.

7. Revise/re-analyze model over time - Once the root causes are identified and appropriate actions taken, time will need to pass to allow the interventions to have an effect on other variables in the model. The amount of time to wait will vary as a function of how quickly the intervention can take place and also the type of issue. Communication problems, for example, can generally be improved faster than structural or process issues. In either case, after the appropriate interval has passed, the model can be re-analyzed to determine if the linkages still hold and if low-scoring issues have improved. Depending on what is found at that point, new priorities can be set and additional actions taken.

Growing interest

Although the practice of linking employee, customer and financial data has its beginnings in the early 1980s, interest in this technique continues to grow. This interest is fueled by several different organizational needs:

  • The need to economize and make the most of scarce resources. Linking research can help by pointing organizations in directions that will give them the greatest return for their investment. Rather than focusing on too many things or only on those that receive low scores (but may not be important) linkage research helps organizations focus on the issues that are really important to the organization’s success, whether that is measured in terms of customer attitudes or bottom-line outcome measures.
  • The need to measure and validate management models/approaches (e.g., HR metrics, balanced scorecard). An assumption underlying most organizational metrics and the balanced scorecard is that employee, customer, process and financial issues are all important to organizational success and are also interrelated. However, critics of the scorecard note that while the idea is sound, these relationships are rarely tested using actual metrics. The modeling techniques of linkage research, however, provide the tools to quantify those associations.
  • The need to work in more integrated manner that aligns separate measurements with common organizational goals. Traditionally, survey research programs have been conducted by different groups (e.g., marketing, HR) at different times and for different reasons. Linkage research, by design, pulls these disparate groups and their data together. In addition to providing the organization with a big-picture view of what causes what, it tends to change the way that survey research is viewed. Rather than being annual or semi-annual events, surveys become an important part of the organization’s long-term strategy for success.
  • The need for proven tools for improving organizational performance. Research has clearly shown that employee, customer and financial issues are significantly related and that by knowing what causes what, organizations can take targeted actions to improve specific outcomes. Rather than focusing on everything or only on low-scoring issues (that are not necessarily important), linkage research helps organizations choose their actions more effectively.

Art and science

In this article, we have reviewed many of the key aspects that can make your linkage efforts successful. However, it is important to note that linkage research is as much an art as it is a science and that there is no one best way to conduct a proper linkage study. Also, most linkage teams develop their skills over time. While linkage research can be challenging, it is well worth the effort.