Web Chapter 4A Procter & Gamble Supply Chain Restructuring: Procter & Gamble (P&G) Blends Models, Judgment, and GIS to Restructure the Supply Chain1 ------------------------------------- <<< INSERT FOOTNOTES: >>> -------------------------------1Adapted from J. D. Camm, T. E. Chorman, F. A. Dill, J. R. Evans, D. J. Sweeney, and G. W. Wegryn (1997, Jan./Feb.), “Blending OR/MS, Judgment, and GIS: Restructuring P&G’s Supply Chain,” Interfaces, Vol. 27, No. 1, pp. 128–142. ------------------------------------ <<< END FOOTNOTES >>> ---------------------------------------Introduction Procter & Gamble (P&G, Cincinnati, OH) produces, markets and distributes more than 300 brands of consumer goods worldwide in more than 140 countries. P&G has operating units (plants, divisions, facilities) in 58 locations around the globe. In 1995 worldwide sales were $33.5 billion with earnings of $2.64 billion. The company has grown continuously over the past 159 years. To maintain and accelerate the growth experienced continuously since the early 1800s, P&G performed a major restructuring called strengthening global effectiveness (SGE) to streamline work processes, drive out no valueadded costs, eliminate duplication, and rationalize manufacturing and distribution. As a consequence of this program, there were major impacts on P&G, which wrote off more than $1 billion in assets and transition costs. The program affected more than 6,000 people and saved $200 million annually before taxes. It involved hundreds of suppliers, more than 50 product lines, 60 plants, 10 distribution centers, and hundreds of customer zones. A major component of the initiative was to look carefully at the North American product supply chain, specifically to investigate plant consolidation. Before, there had been hundreds of suppliers, more than 50 product categories, more than 60 plants, 15 distribution centers (DCs), and more than 1,000 customers. As P&G became global in terms of brands, common formulas, and packages, there were economies of scale and fewer operations. Thus, plants needed to be closed to cut manufacturing expense and working capital, to improve speed to market, and to help avoid capital investment. P&G also wanted to deliver better consumer value by eliminating non valueadded costs; thus, they wanted to develop more efficient linkages with trade customers, reduce customer inventory, and eliminate the least productive sizes. The decision to restructure the supply chain seemed like the right approach. For more on supply chain management (SCM), see Chapter 8 in the text. P&G wanted to restructure the supply chain because • Deregulation of the trucking industry had lowered transportation costs. • A trend toward product compaction allowed more product to be shipped per truckload. • Recent focusing on total quality had led to higher levels of reliability and increased throughput at every plant. • Product life cycles had decreased to about 18–24 months instead of 3–5 years over a few decades. • Several corporate acquisitions had given P&G excess capacity. So executives focused on product sourcing: choosing the best site and operation level for manufacturing each product. The production scope at a given site is limited to products relying on similar technologies. Producing too many products at a site can be too complex. But large, singleproduct
plants can be risky (if demands shift). Since plant locations affect raw materials supply costs and the distributing of finished products, the distribution system had to be considered. The scope of the project was defined by these factors. DSS Project Organization Many teams were formed from multiple business functional areas to handle product sourcing options and the DC location options with customer assignments and transportation decisions (one team). Early in the project, P&G executives recognized that mathematical programming models would be needed to examine the potentially millions of possible alternatives and select a reasonable set of scenarios to investigate. More than 500 P&G employees were involved in this supply chain restructuring. It took months to collect and analyze data (plant closings had to include the “people factor”). A team from the University of Cincinnati joined the company’s operations research/management science (OR/MS) team to provide analytical support and objective input to the decision-making process. The combined OR/MS team was responsible for developing a DSS for the product strategy and distribution teams to use to obtain the best options for more detailed analysis. Modeling The structure of the teams followed the organizational structure of P&G in terms of a brand management philosophy. Each product-strategy team was familiar with its product category, and so all reasonable options (alternatives) for that brand would be investigated. This provided a natural decomposition in the modeling effort. The distribution team could work on DC choices based on its knowledge of the consolidations being considered by each of the product-strategy teams. Given the product-strategy teams’ possible consolidation plans, the distribution team could analyze changes for the DC network to support the complete consolidation plans. This could, however, lead to sub optimization because the entire system was not considered at once. However, similar projects at other firms had taken years to develop comprehensive supply chain and distribution models. This project had a tight time limit (1 year). This is a good example of satisficing because the ‘complete’ model would have been of no use due to the time constraint. A prototyping approach was used to develop the DSS. As teams required new features (as team members better understood the problem), they were added incrementally. As data became available, they were encoded and built into the software modules that were added to the system anticipating the data “arrival.” The OR/MS team’s objectives were to 1. Provide models and decision support for the product-strategy teams 2. Provide support to a team of experts in transportation and distribution for solving warehousing, distribution, and customer allocation problems 3. Obtain a best, complete supply chain solution across product-strategy and distribution teams. P&G had been using a successful comprehensive logistics optimization model to support its sourcing decisions for multiple product categories and multiple echelons for a few years. But it ran on a mainframe, took a long time for each run, and required a long cycle time to obtain an answer in a managerially meaningful report. There were a number of other weaknesses related to data display and access. Because of the many what-if cases the product-strategy teams had to run, an interactive tool with a quick turnaround time was critical for project success. Following the decomposition structure of the teams, the supply chain model was further decomposed into (1) a distribution location problem and (2) a set of product-sourcing problems (one for each product category). Distribution Center Location Model The team aggregated trade customer demand into 150 customer zones to determine the correct location of 5–12 facilities (plants and DCs) in the supply chain. The major considerations influencing the choice of DC locations were customer location, customer service, and sole sourcing. The DCs had to be close enough to customer zones to maintain customer service levels.
The model was an ordinary incapacitated facility location model for finding optimal DC locations and assigning customers to DCs, while minimizing the cost of all DC–customer zone assignments. There were 17 possible locations for DCs, both existing locations and reasonable alternatives based on earlier studies and analyses. Alternatives with 5–13 DCs for the 123 customer zones were considered. The models had about 2,000 variables (only 17 binary–one for each DC, meaning 0 [if closed] or 1 [if open] values) and 2,200 constraints. Families of optimal and near optimal solutions for each case were found and stored in a database for later access by the product-sourcing model. Near-optimal solutions give the decision makers more options from which to choose. This is important in the choice phase of decision making, especially if the decision maker has additional information that creates “implicit” constraints or preferences, unknown to the DSS development team. Furthermore, there may be other considerations not built into a model that can lead to the selection of a near-optimal solution. This also helped when working with the product-sourcing options in that the near-optimal solutions expanded the solution space, giving the analysts and decision makers a larger set from which to choose and a sense that the solution would be optimal (or at least closer to the global optimum value). The Product Sourcing Model The product-sourcing model (PSM) was a simple transportation model for each product category. The product-strategy teams specified plant location and capacity options to investigate, while each DC was a demand representing the total aggregated demand from its customer zones. Since manufacturing costs were the most important in the product-sourcing decision, good estimates were developed. Real data were used (when available) or a simple model generated the cost for shipping between two points. The generating algorithm was validated using known costs. The transportation models were solved with a readily available, fast computer code (an out-ofkilter algorithm). Thus, options could be evaluated in real time. Integration with a Geographic Information System The product-strategy teams wanted a powerful, flexible decision support system that could display and manipulate solutions for ease of interpretation. So, the product-sourcing model was integrated with the geographic information system (GIS) Mapinfo. Mapinfo had a readily customizable user interface through its Map basic programming language. The GIS provided the capability to make quick changes, run new scenarios, and compare the results. The whole system runs in seconds on a standard laptop PC. File manipulation is used to ship data from Mapinfo to the optimization software and back. The user interface provided by the GIS led directly to user acceptance of analytic techniques. The product-strategy team members interacted directly with the optimizers but created scenarios directly with the GIS and saw the results there as well. Often, insights given by the spatial visualization led to new and better options. The model acted as a laboratory in which product-strategy teams tested ideas and developed insights. The spatial visualization also readily identified database errors that could have gone undetected; that is, links between DCs and customers had to look reasonable. If a DC was linked to a customer in the middle of the ocean, the error could be seen. DSS Operation The solutions to the DC model were inputs to the product-sourcing model. The product-strategy teams chose the consistent potential plant locations, capacities, and manufacturing costs and solved the product-sourcing model. An optimal manufacturing-and-distribution plan was then found. Given the solution, the product-strategy teams created a new case, re-solved the model, and repeated this process until they were satisfied. This rapid process lead to the teams working in more than 1,000 sessions to evaluate alternatives. (Often DSS leads to extra analysis because the tools are so good.) It was felt that the objective value of any solution was within 10 percent of the real value because demands and costs were forecasted and some estimates were made. The model was validated to be within 2 percent of the optimum using existing data, even though it
was designed to represent the state of P&G in the year 2000. Once the teams had established reasonable alternatives, they focused on collecting data needed to select an optimal long-term solution based on risk-adjusted net present-value analysis. This was a simulation model of a net present-value analysis built in a spreadsheet. It was solved with @Risk, a popular spreadsheet add-in. While evaluating options, the OR/MS team discovered that the best product-sourcing options were indeed independent of DC locations, thus justifying the decomposition made initially. Manufacturing costs dominated transportation costs in alternatives of product-sourcing options. Integration of Human Judgment The distribution and product-sourcing models were integrated and solved. Using the models, each category team developed a number of potential sourcing options, varying from 2 to 12, rejecting some. During this process, they included many subjective considerations, including the minimum number of plants, which DCs to consider, and whether to use cross-border sourcing. They used the product-sourcing model to find the best options for each product category and then subjected these options to a more thorough financial and risk analysis. This analysis took into account costs including taxes, interest rates, labor rates, and utilities over a number of years into the future. The risk analysis considered the possibility of earthquakes, hurricanes, and so on. Political considerations were also considered in some of the site selections. Expert human judgment was considered a critical part of the DSS approach. Difficult problems like this need a hybrid approach which closely links expert human judgment and mathematical optimization. The GIS and consideration of qualitative factors lead to the system’s success. DSS Benefits Two years after completion of the DSS, P&G closed 12 sites and wrote off more than a billion dollars worth of assets and people-transition costs. The new North American manufacturing and distribution system is saving more than $250 million (pretaxes) annually. Most savings have resulted from lower manufacturing costs, because of fewer plants with less staff, and from a more efficient supply chain. Because there are fewer sites, delivery expenses have indeed increased. Regardless of the DSS, P&G would have closed plants and improved its operations. The OR/MS team feels that it was directly responsible for 10 percent of the total savings ($25 million). In addition, the solution implemented affected about 6,000 people because of changes in the product supply system. One-third retired early, another third were relocated, and the last third were retrained and hired by other companies. P&G treated these people fairly, with respect and dignity. Because of the project’s success, P&G now requires that all future sourcing decisions be based on analytic methods. P&G has also used the model to conduct competitive analyses. Tools similar to the model have gained widespread acceptance, with demand for their use in domestic, regional, and global sourcing studies. P&G has established a new internal Center for Expertise in Analytics for solving business problems. Questions 1. Why is the visual system with a graphical interface provided by a GIS an ideal workbench for analyzing distribution scenarios? 2. Why do you think so many people were involved in the P&G supply-chain restructuring effort? 3. Describe the different models and types that were developed for the DSS. How were they integrated? 4. Why was the decision to decompose the product-sourcing problem from the distribution center locations problem reasonable? 5. Why was it reasonable to aggregate the customers into a single customer zone? 6. Why was it important to take human judgment into consideration? 7. How well do you think P&G would have done without an analysis of this type? What could have happened? 8. Explain why this DSS project was a success.
Web Chapter 4B Scott Homes Selects a Mobile Home Supplier Scott Homes Constructs an Expert Choice Multicriteria Model-based DSS for Selecting a Mobile Home Supplier1 ------------------------------------- <<< INSERT FOOTNOTES: >>> -------------------------------1Contributed by the MASK Innovation DSS Class Team: Avryl Alvarez, Candace Kemp, Wakeelah Moore, and Veneetia Smith, Terry College of Business, The University of Georgia, Athens, GA. ------------------------------------ <<< END FOOTNOTES >>> ---------------------------------------Introduction Mr. Jesse Scott is the owner of Scott Housing Inc., a mobile home sales firm in Sylvania, GA. The decision facing Mr. Scott involved selecting potential manufacturers of mobile homes. The nature of the problem was to decide which suppliers could best ensure customer satisfaction and a profitable bottom line for Scott Housing through the quality of their products and other factors. This multiple-criteria (multiple goals, multiple objectives) decision-making problem is an ideal one to solve with the aid of Expert Choice, an implementation of the Analytic Hierarchy Process (AHP). The Analytic Hierarchy Process The Analytic Hierarchy Process, developed by Thomas Saaty (1995, 1996)(see also Forman and Selly, 2001), is an excellent modeling structure for representing multicriteria (multiple goals, multiple objectives) problems—with sets of criteria and alternatives (choices)—commonly found in business environments. The decision maker uses AHP to decompose a decision-making problem into relevant criteria and relevant alternatives. The AHP separates the analysis of the criteria from the alternatives, which helps a decision maker focus on small, manageable portions of the problem. The AHP manipulates quantitative and qualitative decision-making criteria in a fairly structured manner, allowing a decision maker to make trade-offs quickly and “expertly.” Expert Choice (Expert Choice Inc., Pittsburgh, PA, www.expertchoice.com; a functional demo version is available directly from their Web site) is an excellent commercial implementation of the AHP. A problem is represented as an inverted tree with a goal node at the top (Figure 1). All the weight of the decision is in the goal (1.000). Directly beneath and attached to the goal node are the criteria nodes. These are the factors that are important to the decision maker. The goal is decomposed into criteria, to which 100 percent of the weight of the decision from the goal is distributed. To distribute the weight (establish a utility function from the decision maker’s perspective), the decision maker pairwise compares the criteria: first criterion to second, first to third, . . ., first to last; then, second criterion to third, second to third, . . ., second to last; . . .; then, next to last criterion to the last one. In doing so, he or she establishes the importance of each criterion, that is, how much of the weight of the goal is distributed to each criterion or how important each criterion is. This objective method is performed by internally manipulating matrices mathematically. The manipulations are transparent to the user because the operational details of the method are not important to the decision maker. Finally, an inconsistency index indicates how consistent the comparisons were, thus identifying inconsistencies, errors in judgment, or simply errors. The AHP method is consistent with decision theory. 2 FIGURE 1 The Expert Choice Scott Mobile Home Model of the Decision Problem Before Pairwise Comparisons. At the top is the goal, the next layer contains criteria, and the bottom layer contains choices. By permission of Expert Choice Inc. The decision maker can make comparisons verbally (e.g., one criterion is moderately more important than another), graphically (with bar and pie charts), or numerically (with a matrix— comparisons are scaled from 1 to 9). Students and business professionals generally prefer to use
the graphical and verbal approaches to the matrix one (based on an informal sample). Beneath each criterion is the same set of choices (alternatives) in the simple case described here. Like the goal, the criteria decompose their weight into the choices, which capture 100 percent of the weight of each criterion. The decision maker pairwise compares the choices in terms of preference as they relate to the specific criterion under consideration. Each set of choices must be pairwise compared as they relate to each criterion. Again, all three modes of comparison are available, and an inconsistency index is derived for each set and reported. Finally, the results are synthesized and displayed on a bar graph. The choice with the most weight is the correct choice. However, the correct decision may not be the correct one under some conditions. For example, if there are two “identical” choices that split the weight (e.g., if you are selecting a car for purchase and you have two identical cars, they may split the weight and neither will have the most weight). Also, if the top few choices are very close, there may be a missing criterion that could be used to differentiate among these choices. This case is a simple decision-making situation with important ramifications for the decision maker. We did not use some of the more advanced features of Expert Choice, which is very flexible. Subcriteria can be incorporated into a model. There is a Ratings Model (see Case Application 2.3), useful for when there are more than seven choices or for recurring decisions (like hiring employees). Expert Choice has a sensitivity analysis module. And, there is a group version called TeamEC that synthesizes the results of a group of decision makers using the same model. Overall, the AHP as implemented in Expert Choice attempts to derive a decision maker’s preference (utility) structure in terms of the criteria and choices and help him or her make an expert choice. See Case Application 2.3 and expertchoice.com for additional details about the AHP and Expert Choice. The Scott Housing AHP Model During an interview Mr. Scott and the team determined that the following criteria were the most important in his decision process: price, customer service, quality of materials, supplier location, timing of delivery, special relationship with the supplier, and how long the supplier has been in the market. Currently Mr. Scott purchases mobile homes from two regular suppliers (Horton Homes and Jaguar) and is considering purchasing from two new suppliers (Schult and Batchlor). He is very much interested in how the model will look and how Schult and Batchlor will stack up against Horton Homes and Jaguar. The Criteria Quality decision making requires identification of the proper criteria, and we worked closely with Mr. Scott to determine these. The six important criteria are price, customer service, quality of materials, supplier location, length of time in business, and special relationship with the supplier. Supplier location is a major concern of our decision maker. A closer supplier means lower freight costs and faster deliveries. This criterion was weighted highest of all at 30.9 percent. Price is the second criterion. We originally thought that price would be the most important and recognized that location has an indirect relationship to the total cost of a mobile home. The price of the mobile home is of great importance when calculating profits. The price weight was 21.3 percent. Other criteria. Scott Housing Inc has had a reputation for providing excellent customer service and high-quality materials. It was reasonable that these criteria be included in the model. Great customer service is defined as being dependable, being accessible, and having a concern for customers. The pairwise comparison step yielded a weight of 15.5 percent. It was also important to Mr. Scott that length of time in business (older is better) and special relationships with the supplier be included in the model to impact on the decision. The Suppliers (Choices)
Horton Homes is located in Eatonton, GA, and has been in business since 1970. Their dramatic growth is directly linked to their commitment to offering the consumer an attractive, affordable home designed and built with modern technological advances. Horton Iron Works uses the latest in welding technology. I-beam frames are manufactured for extra strength and support, adding to the quality of the home. Horton manufactures accent moldings such as baseboards, door casings, and crown and chair rail moldings in various sizes, styles and colors. Horton owns more than 100 trucks to ensure the prompt and courteous delivery of homes to retail sales centers. Adhering to minimum standards higher than those recommended for site-built homes, Horton Homes builds to the federal safety and construction standards set by the U.S. Department of Housing and Urban Development (HUD). This guarantees that each home meets or exceeds strict federal regulations for such factors as design, construction, energy efficiency, wind and fire safety, ventilation, durability, and installation procedures. This is their way of ensuring a quality home. Jaguar Homes specializes in the design and building of luxurious one- and two-story homes. They have been in business for 35 years building single-family homes, apartments, and commercial properties in the Los Angeles, CA, area. Recently they have expanded their business to the east coast, particularly the Virginia area. Schult Homes was founded in 1934 and has prided itself on giving customers quality, craftsmanship, and service and having a solid relationship with retail dealers. Schult offers a wide variety of models, floor plans and prices to meet the customer’s needs. Production efficiency resulting from the Schult manufacturing process can save a homebuyer up to 30 percent per square foot over a site-built home while offering the same amenities and appearance. Batchlor Supply has been in the business of manufacturing homes for approximately 20 years. They are located in Raleigh, NC, and are a fairly new potential vendor for Scott Housing Inc. Batchlor Supply prides itself on quality customer service and economical pricing. The Model and the Results We constructed an Expert Choice model with the six criteria and four choices outlined above (Figure 1). After completing pairwise comparisons for the six criteria and for the four choices under each criterion (seven comparison sets in all), we synthesized the results (Figure 2). <<< INSERT FIGURE 2 HERE >>> <<< WAS FIGURE 5.9 >>> FIGURE 2 Synthesis of the Results of the Expert Choice Scott Mobile Home Decision Model By permission of Expert Choice Inc. Horton Homes was a clear winner with a total score of 0.360, Jaguar came in second with a score of 0.240, Batchlor received a score of 0.210, and Schult received a score of 0.190. Location overwhelmingly drove the model to this solution because it is the most important criterion by a wide margin. The inconsistency index was remarkably small—equal to 0.01. Mr. Scott agreed with these results and was impressed with the accuracy of Expert Choice. As he stated in a letter: It was my pleasure to work with The MASK Innovation on their project for their decision support class. I did not know that decision support software was so easily available (i.e., over the Internet). Amazingly, the group’s judgments matched mine and, in the future, I may consider using Expert Choice as a tool in my decision making. Development of the DSS helped Mr. Scott in making a strategic decision, and the effectiveness of his decision making was increased. Tools like ExpertChoice truly enhance the quality of the decision-making process. Questions 1. This is a multi criteria problem. What are the specific goals? 2. What are the criteria? Discuss why they are important. 3. How were the choices “guided” by the criteria? 4. How did Expert Choice enhance Mr. Scott’s ability to make a decision?
5. Download the demo version of Expert Choice and build a model similar to Mr. Scott’s for practice using the software, or use it to select an automobile, motorcycle, or vacation. Use your own judgments. How close are they, and the solution, to Mr. Scott’s? Why are they similar or different?