Multi-dimensional Customer Satisfaction - part 4

In the last three blogs customer satisfaction has been examined as a multi-dimensional measurement. Customers in the business world really have many contacts when communicating with another business. Since a business customer has many contacts, customer satisfaction cannot accurately be described with a measurement from a single individual within the customer company. In the previous blogs measurements from several functional departments within the customer organization were analyzed to indicate how differences can occur and how they can be detected. While there is more to explore within this fertile area of differences (such as looking for the key satisfiers in each area), the question must also be asked whether the individual measurements can be combined to provide an overall measurement for the customer. This is the topic for this blog; examining ways to combine individual scores when multiple measurements have been taken within a single customer organization.

The “I don’t know what else to do” Approach

I believe it necessary to address the approach that the uninitiated will take so that there is a base position from which to compare other methodologies. There are two key theorems in statistics that seem to be the backbone to the “I don’t know what else to do.” The first theorem is the law of large numbers. A paraphrase of the law says as the number of samples (repetitions of an experiment) increases, the difference between the observed results and the theoretical results gets smaller and smaller. This doesn’t apply to this situation since there are not a large number of samples for a given company and they are not similar since they represent different organizations within the customer company. Throw out the law of large numbers argument for use in combining satisfaction scores.

The second theorem is the central limit theorem. I usually refer to this theorem as the “Big Cohuna” of statistics since it is the base from which much of inferential statistics is founded. It provides the very strong theoretical base for the use of sampling to predict the characteristics of a population. Again, a paraphrase of the central limit theorem says that the means of samples are themselves normally distributed no matter what the shape of the population. For example, if fifty samples are taken from a population and each sample has 30 observations, when the average of each sample is calculated and then plotted, the distribution of the sample averages will follow the bell- shaped curve, known in statistics as the normal distribution. This will be true whether the samples were taken from a bi-modal distribution (one with two peaks) or a flat distribution (one with no peaks) or any other shape. Once again this very strong theorem does not apply since each observation must have similar characteristics. So throw out the central limit theorem argument for use in combining scores.

In spite of the preceding discussion there will be those who say “take the individual scores and combine them by taking the average.” While this average will give a measurement, it certainly is not clear what meaning the measurement has. Although not quite analogous, this measure has about as much meaning as measuring the body temperature behind the ear (between the ear and the scalp); while there will be a temperature measured, there will be no clear understanding how it relates to the health of the body. A similar conclusion can be drawn from the combination of the multiple measurements by simply taking the arithmetic average. The obvious flaw is that each measure is given equal weight. Since it is highly unlikely (virtually impossible) for each measure to have an identical impact on overall satisfaction, the arithmetic average may be dangerous since it will give the same weight to the least and most important measures. A high satisfaction score from the least important will counter a low satisfaction score from the most important. Thus, the conclusion might be that all is well with a specific customer even though the most important measurement indicates very low satisfaction.

Adding the “seat-of-the-pants” Correction Factor

As soon as it is obvious that the arithmetic average gives equal weight to each measurement, the simple answer is to change the weights. If there are three measurements for a specific customer, each score is weighted by 1/3 so that the total score is 1/3 of the first score plus 1/3 of the second score plus 1/3 of the third score. The key is that the sum of the weights must equal one. Therefore, a seat-of-the-pants solution is to change the weights on each measure from equal values to other values that are more representative of the business relationship. The weights given may reflect the perception of influence of each person surveyed on the business relationship. For example if the management organization has the most influence and the other two organizations (purchasing and operations are about equal) one could weight the measurement from management at one half and the other two at one quarter each.

As long as the sum of the weights equals one, any combination of weights is possible. The only question is what is the correct combination of weights? Clearly there are an infinite number of combinations and there is most likely a “best” combination of weights that best describes the overall satisfaction of the customer company. Without additional knowledge the weights given to each measurement might best be selected to reflect the company strategy. For example, if the company is a high technology company that sells to the operations organization of the customer, the weight placed on the measurement from operations should be greatest. If, on the other hand, the company sells a commodity product and differentiates itself through its service organization, either the purchasing or operations organization of the customer might receive the greatest weight.

In each of the examples noted in the previous paragraph, the weights used are somewhat arbitrarily chosen. While the weights might be “reasonable,” the likelihood they are the best combination of weights is remote. When the weight of one half is given to management, it could be that a weight of 4/10 or 6/10 might better reflect the customer. Since the weights are chosen based on strategy or perception of the customer, they represent management’s best assessment of the customer based on knowledge of the customer and the current business strategy. The fact is that many times the experience of the company executives is an excellent source for assessing the weights. While the weights they give may appear arbitrary, they come from years of experience in the industry and knowledge of the customers and hence, should not be taken lightly. Thus, when the “seat-of-the-pants” approach uses the experience of the company executives, it is probably the best estimate available.

Getting Weights from a Customer Contact Model

Most companies that I have worked with look at customer contact in a qualitative way. The only quantitative measure of customer contact is time with the customer. In order to put more accurate weights on the different measures of customer satisfaction from a customer company, an assessment of the customer contact in each of the measures provides a more accurate assessment of the value of the measurement. Recent research indicates customer contact has at least three dimensions; namely time with the customer, the richness of information transferred, and the intimacy of the contact. I will examine the current research on customer contact in the next blog in order to provide sufficient detail to understand how to apply the contact information into the weights for the multi-dimensional measurement of customer satisfaction.