When people talk about successful (service providers) and those that are not so successful, the customer determines at the end of the day who is successful and for what reason.
—Jerry Harvey
There are a number of key differences between the design, production, and delivery of a product and the design and delivery of a service. Some of these differences can have an influence on the way statistical process control (SPC) is employed. While the underlying theory is the same, deployment of SPC in the service sector often differs in certain respects from SPC deployment in the manufacturing sector.
Among the key differences between products and services1 that might affect SPC deployment are:
• Products are generally tangible, while services, even those with a tangible component, tend to be intangible in terms of customer focus.
• In many instances, services are created and delivered at the same time and by the same people. Products tend to be created in advance and different people do the manufacturing and delivery. For this reason, service defects are more often found by customers than in the case of manufacturing.
• Service processes tend to be more visible to customers than manufacturing processes.
• Key quality characteristics (KQC) of services tend to be less quantifiable and can be more subjective than KQCs for products.
In this chapter, we will discuss how these differences manifest in the use of SPC in services.
Defining Quality in Services
Because services tend to be intangible, there often is a human interaction involved between the service provider and the customer, and there is a greater tendency toward an attitude of “beauty being in the eye of the beholder” in services than in manufacturing. For these reasons, defining quality is often more difficult in the service sector. But if quality is ill defined, how then are we to judge whether the service is conforming or nonconforming?
A number of attempts have been made to define the dimensions of service quality. Two of these are presented in Table 8.1. The definition developed by Parasuraman et al.2 seems to be particularly good at highlighting the increased difficulty of defining quality in services. Their definition is that quality of a service is the difference between the customers’ expectation and their perception of the quality of the service rendered. Certainly there are aspects of this in product quality as well; however, with products there are usually more objective KQCs such as dimensional conformance that make defining quality somewhat more straightforward.
Table 8.1 Dimensions of service quality
SERVQUAL dimensions* |
Dimensions of service quality for hospitals** |
Tangibles |
Respect & caring |
Reliability |
Effectiveness & continuity |
Responsiveness |
Appropriateness |
Assurance |
Information |
Empathy |
Efficiency |
|
Meals |
|
First impression |
|
Staff diversity |
|
Efficacy |
Sources: *Parasuraman, Zeithaml, and Berry et al. (1988).
** Sower et al. (2001).
None appear to be universally applicable to all services. For this reason, these attempts should be viewed as starting points for the determination of the true dimensions of quality for the particular services and customers involved.
Rare Events
Some KQCs in service applications are rare events. Examples include surgical errors, lost time accidents, and erroneous tax returns prepared by a certified public accountant CPA. Because these events are so rare, control charts that track defectives (p-charts) and defects (u-charts) are not well suited for these applications. If the rational subgroup (see Chapter 4) is small (for example one work day), there will be many days where the control chart entry is zero. If the rational subgroup is large (for example one month), it will take more than 2.5 years to collect sufficient data to create control charts that fully characterize the process.
One solution is to not use control charts for rare events at all and to treat each rare event as a special cause. Because these events are, by definition, rare, there is little risk of wasting time looking for an assignable cause when only common cause variation is present. The investigation of rare events, however, should take a systems perspective rather than simply attempting to identify and punish the “guilty party.” According to Dr. Donna Cananiano, then surgeon-in-chief at Nationwide Children’s Hospital, “I would like to know right away if we have an (rare) event today…When you actually look at why the (healthcare professional) makes an error in the first place, it’s a systems problem.”3 So, in this approach, the same diligence and methodology should be brought to bear on the investigation of a rare event as with an out of control signal on a control chart.
An alternative solution is to track not the incidence of rare events but the elapsed time between rare events and plot that data on individual and moving range charts. The elapsed time between events can be transformed to a Weibull distribution which is sufficiently approximated by a normal distribution to allow the transformed data to be plotted on individual and moving range control charts or exponentially weighted moving average charts.4 While this is a better approach for control charting than plotting the rare event frequency, if the event is truly rare and serious, usually it is better to treat all rare events as if they were due to an assignable cause.
What Chart to Use?
Because services tend to be intangible, it is much more difficult to measure the KQCs for services than for products. While important to product quality, customer expectations and opinions tend to be more important determinants of service quality. Hospitals, for example, are very concerned with patients’ opinions about the overall hospital experience during their stays. Hospitals often go to great lengths using focus groups and other tools to determine how patients form their opinions about quality and what factors enter into these decisions. This is very important, because simply measuring factors that are easy to measure without regard for how those factors play into the customer’s opinion about quality provides a result that may not be meaningful or useful for driving quality improvement projects. Without this linkage to the customer, even very precise measurements of irrelevant factors will be of little value.
Once the important factors are determined, they are often measured using some form of survey instrument which must be assessed for validity and reliability. A valid and reliable instrument can produce meaningful data—but how do we employ SPC in the evaluation of that data? Often patients respond to survey questions by marking a scale of 1 to 5 or 1 to 7 with one end anchored with something like “Completely Agree” and the other end anchored with something like “Completely Disagree.” The responses are discrete data (only integer values are allowed), and are bounded (responses beyond the values in the scale are impossible).
One approach taken by some hospitals is to define a month as a rational subgroup. All of the responses for a particular month are analyzed as one sample. This turns the data set into continuous data bounded by the limits of the scale. The sample means are plotted on an x-bar chart (with variable sample size) and the sample standard deviations are plotted on an s-chart.
Another approach is to define a cutoff scale score that indicates a respondent is dissatisfied. One approach might be to assign the scale midpoint as the cutoff between a satisfied and a dissatisfied respondent as shown in Figure 8.1. Other organizations that aspire to delight customers might set the cutoff scale score higher. Each respondent can be classified as a satisfied or dissatisfied customer and that data can be used to construct a p-chart. Example 8.1 illustrates the use of a p-chart using survey data in this way.
Figure 8.1 Typical survey response scale
Examples of SPC Usage in Service Organizations
Example 8.1
Control Charts in Healthcare
A hospital launched a project to increase patient satisfaction with meals. The effort was spurred by two things:
• Research by Dr. Susan Schiffman at Duke University Hospital about the clinical importance of making meals more palatable so that patients want to eat, and
• The success of initiatives at hospitals such as M.D. Anderson Cancer Center and Medical Center Dallas to place the patient in the center of what a hospital does by offering more choice in meals.
Because the major commission that accredits healthcare organizations encourages the use of appropriate statistical tools in performance measurement, the hospital decided to incorporate SPC into the project. Based on focus groups conducted with recently discharged patients, the hospital developed a four-question survey they used to obtain feedback from patients as they were discharged about food service quality. The patients responded to the four statements using a 7-point scale, with 7 representing “strongly agree” and 1 representing “strongly disagree.” Any patient who responded to any one of the three items with a scale score below 4 was considered to be dissatisfied with the meal experience. The initial analysis revealed that more than 10 percent of patients were dissatisfied, and the most frequently cited reason for dissatisfaction was food taste.
Analysis of the focus group information revealed that patients judged the taste of the food by comparing it with their expectations. This led to two obvious ideas for decreasing the level of dissatisfaction:
• Patients on a liquid or bland diet who have steak and potatoes expectations were certain to be dissatisfied with hospital meals. The hospital decided to have a representative of the dietetics department meet with each patient to explain the diet specified by their physician and the necessity for that diet. This helped align the patients’ expectations with the restrictions of their prescribed diets.
• The patients were offered choices within the scope of their diets. No longer would a patient automatically receive green gelatin and chicken broth. They could choose from several flavors of gelatin and alternatives to chicken broth. This gave them some control over their dietary decisions.
As the control chart in Figure 8.2 shows, this project was very effective. The proportion of patients dissatisfied with hospital meals was reduced from more than 10 to 7 percent—a 30 percent improvement.
Figure 8.2 p-chart for survey data
Source: Created using Minitab 18.
Example 8.2
Control Charts in Retail
A locally owned hardware store was concerned about competition from the national chain building supply store that had recently located in their market area. Since the chain store had significantly greater buying power, price was not a feasible way to compete. Inspired by a news story about a building supply store that offered extraordinary guarantees, they decided that was one way they could beat the competition. They settled on the following extraordinary guarantee: “We guarantee that we will have what you want from our normal offerings in stock or we will provide it to you free within 24 hours!”
Since the cost of expedited shipping and free merchandise associated with a stockout under this guarantee can be very high, they felt they needed to assess how well their system was prepared to support it. They collected data each day on the number of customers served and the number of stockouts. They constructed a p-chart to analyze the data and to assess how well their process was prepared to support the extraordinary guarantee.
The p-chart shows that their process is in control; however, the average percent (
= 0.04606) of customers who asked to purchase an item that was out of stock was too high to be economically feasible to support the extraordinary guarantee. The owners determined that the process must be capable of achieving ( ≤ 0.01000) in order for the extraordinary guarantee to be feasible. Projects to improve their forecasting and inventory management policies were instituted with a goal process capability of ( ≤ 0.01000) within 6 months.
Figure 8.3 p-chart for stockout data
Source: Created using Minitab 18
Example 8.3
Control Charts in Finance
The accounts payable department of a large corporation processes more than 1,000 invoices per day. On a randomly selected day each week, 100 invoices are selected at random and reviewed for errors by the assistant manager of the department. In the past, the information was used to identify which accounting clerks were making errors. The offending clerks were warned to pay closer attention to their jobs to avoid further errors. Whenever the number of errors was above 4, all of the clerks were warned that they must do better.
A newly appointed assistant manager decided to use quality tools to assess the entire process. She constructed a p-chart to analyze the weekly sample of invoices. To her surprise, she found that the process was in control, however the overall proportion of invoices with errors in the samples was higher than she felt was necessary.
Further investigation revealed that the training process for newly hired clerks and the continuing education process for existing clerks were not as robust as she thought they should be. In addition, a number of invoices submitted for processing lacked critical information.
The assistant manager redesigned the training programs for the clerks and held seminars for the departments that submitted invoices for payment stressing the necessity for provision of complete and accurate information. She also halted the practice of department-wide meetings when errors exceeded an arbitrary limit. After a few months, the assistant manager observed a signal of 8 points in a row below the CL on the p-chart. This indicated that the changes she had made were effective, and she revised the control limits to reflect the state of the new process.
Chapter Take-Aways
• While there are key differences between the manufacture of products and the delivery of services, SPC is equally applicable in both sectors.
• Care must be taken when attempting to apply SPC to monitor rare events. Traditional control charts may not be the best option.
• Selection of the right control chart for a particular service application might not be as straight forward as in manufacturing applications.
• Specific examples were provided for the use of SPC in health-care, retail, and financial services applications.
Questions You Should be Asking About Your Work Environment
• How do you define quality for the services you provide? Were customers included in the discussion before the quality definition was established?
• How do you know whether you are satisfying or delighting your customers?
• Have you resisted using SPC because it is a “manufacturing tool” not applicable to services? How might you improve your operation if you implemented SPC?
• Do you have a continuous improvement (CI) program in place? How do you know how effective it is? Could you provide proof to an outsider who asked about your CI program’s effectiveness?