Development of the city of quality: A hypertext-based group decision support system for quality...

20
Decision Support Systems 11 (1994) 299-318 299 North-Holland Development of the city of quality: A hypertext-based group decision support system for quality function deployment 1 Michael Wolfe West Virginia University, Morgantown, WV,, USA The goal of the City of Quality system is to provide a group support system (GSS) for strategic planning of large- scale system development projects-a GSS which can signifi- cantly enhance these projects by providing tools for detailed requirements management, modeling, and analysis. It is an adaptation and implementation, using hypertext, of an ensem- ble of Japanese management techniques. The name of these techniques translates, roughly, as Quality Function Deploy- ment (QFD), which is part of Total Quality Management (TQM). These manual techniques have been developed over a number of years as a response to the problems of developing large, complex systems that must satisfy conflicting require- ments. They can be an important strategic planning tool. When successfully applied, QFD can save 50 percent of the system development costs; however, many organizations find it difficult to apply QFD successfully.Decision support system (DSS) and GSS technology can help address this problem. Hypertext tools allow the development of a low-cost DSS implementation to support QFD. This DSS can be extended to a GSS with network management software. Keywords: Quality function deployment; QFD; Total quality management; TQM; Group support systems; Deci- sion support systems; Group decision support sys- tems; Strategic planning; Taguchi method. Correspondence to: Michael Wolfe, Department of Manage- ment, West Virginia University, Morgantown, WV 26506- 6025 2, USA. 1 Research sponsored by the Air Force Office of Scientific Research/Air Force Systems Command, United States Air Force, under Contract F49620-88-C-0053.The United States Government is authorized to reproduce and distribute reprints for governmental purposes notwithstanding any copyright notation hereon. A version of this paper was presented at the First ISDSS conference, Austin, TX, September 1990. This is a substantial revision of that paper. 2 Michael Wolfe is on a research assignment, and is tem- porarily away from West Virginia University. He may be reached at P.O. Box 31648, Dayton, OH 45431 until Jan- uary, 1993. 1. Introduction The acquisition of any large, complex system normally involves many decisions executed by a large number of people over a period of several years. The decisions are usually negotiated by user-sponsors, designers, and (sometimes) end- users. Individual end-users often have conflicting requirements for systems, such as high perfor- mance and low cost; hence, theory and practice lead to expectations of significant difficulties in reaching a satisfactory decision whenever multi- ple users and system developers are involved. This is true regardless of the type of system being developed. An unstructured, manual approach to this pro- cess involves meetings between various designers and decision-makers and meetings to resolve con- flicts until a system is completed many years after its inception. Such methods for communicating and resolving these conflicts are unreliable and awkward [24]. The resulting systems often fail to address important end-user requirements and frequently require substantial redesign expendi- tures. A possible solution is to provide a structure within which the group may work to develop the Michael D. Wolfe is an Assistant Pro- fessor of Management at West Vir- ginia University. He received his Ph.D. from the Department of Man- agement Science and Information Systems at the University of Texas at Austin in 1988. Currently, he is work- ing with the Air Force Logistics com- munity, doing basic research on group support systems, and research on ap- plications of hypertext-based group support systems for system acquisi- tions. Before joining West Virginia, he worked for the Center for Cybernetic Studies at the University of Texas at Austin, where he did research on decision support systems for logistics. He has also worked for Rockwell International researching applications of artificial intelligence. 0167-9236/94/$07.00 © 1994 - Elsevier Science B.V. All rights reserved SSDI 0167-9236(92)00014-6

Transcript of Development of the city of quality: A hypertext-based group decision support system for quality...

Page 1: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

Decision Support Systems 11 (1994) 299-318 299 North-Holland

Development of the city of quality: A hypertext-based group decision support system for quality function deployment 1

M i c h a e l W o l f e

West Virginia University, Morgantown, WV,, USA

The goal of the City of Quality system is to provide a group support system (GSS) for strategic planning of large- scale system development projects-a GSS which can signifi- cantly enhance these projects by providing tools for detailed requirements management, modeling, and analysis. It is an adaptation and implementation, using hypertext, of an ensem- ble of Japanese management techniques. The name of these techniques translates, roughly, as Quality Function Deploy- ment (QFD), which is part of Total Quality Management (TQM). These manual techniques have been developed over a number of years as a response to the problems of developing large, complex systems that must satisfy conflicting require- ments. They can be an important strategic planning tool. When successfully applied, QFD can save 50 percent of the system development costs; however, many organizations find it difficult to apply QFD successfully. Decision support system (DSS) and GSS technology can help address this problem. Hypertext tools allow the development of a low-cost DSS implementation to support QFD. This DSS can be extended to a GSS with network management software.

Keywords: Quality function deployment; QFD; Total quality management; TQM; Group support systems; Deci- sion support systems; Group decision support sys- tems; Strategic planning; Taguchi method.

Correspondence to: Michael Wolfe, Department of Manage- ment, West Virginia University, Morgantown, WV 26506- 6025 2, USA. 1 Research sponsored by the Air Force Office of Scientific

Research/Air Force Systems Command, United States Air Force, under Contract F49620-88-C-0053. The United States Government is authorized to reproduce and distribute reprints for governmental purposes notwithstanding any copyright notation hereon. A version of this paper was presented at the First ISDSS conference, Austin, TX, September 1990. This is a substantial revision of that paper.

2 Michael Wolfe is on a research assignment, and is tem- porarily away from West Virginia University. He may be reached at P.O. Box 31648, Dayton, OH 45431 until Jan- uary, 1993.

1. I n t r o d u c t i o n

The acquisi t ion of any large, complex system

normal ly involves many decisions executed by a large n u m b e r of people over a per iod of several years. The decisions are usually negot ia ted by user-sponsors, designers, and (sometimes) end- users. Individual end-users of ten have conflicting

requ i rements for systems, such as high perfor- mance and low cost; hence, theory and practice lead to expectat ions of significant difficulties in

reaching a satisfactory decision whenever mult i-

ple users and system developers are involved. This is t rue regardless of the type of system being

developed. A n uns t ruc tured , manua l approach to this pro-

cess involves meet ings be tween various designers and decis ion-makers and meet ings to resolve con- flicts unt i l a system is completed many years after its incept ion. Such methods for communica t ing and resolving these conflicts are unre l iab le and

awkward [24]. The resul t ing systems often fail to address impor tan t end-user r equ i rements and

f requent ly require substant ia l redesign expendi- tures.

A possible solut ion is to provide a s tructure

within which the group may work to develop the

Michael D. Wolfe is an Assistant Pro- fessor of Management at West Vir- ginia University. He received his Ph.D. from the Department of Man- agement Science and Information Systems at the University of Texas at Austin in 1988. Currently, he is work- ing with the Air Force Logistics com- munity, doing basic research on group support systems, and research on ap- plications of hypertext-based group support systems for system acquisi- tions. Before joining West Virginia,

he worked for the Center for Cybernetic Studies at the University of Texas at Austin, where he did research on decision support systems for logistics. He has also worked for Rockwell International researching applications of artificial intelligence.

0167-9236/94/$07.00 © 1994 - Elsevier Science B.V. All rights reserved SSDI 0167-9236(92)00014-6

Page 2: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

300 Michael Wolfe / Deuelopment of the city of quality

system. Structure, with or without computer sup- port, tends to enhance the quality of group pro- cesses [33]. A Japanese method for structuring the process is, roughly translated, Quality Func- tion Deployment (QFD). QFD has existed since 1967 but has only recently received a great deal of attention as a managerial tool, following an article in the Harvard Business Review [12].

QFD is an integrated set of tools for recording user requirements, technical specifications that satisfy these requirements, any trade-offs that might be necessary among these technical specifi- cations, and several other items that might help in the system development process. QFD has recently been recognized as an important strate- gic planning tool [16]; when successfully applied, it can result in a 50-percent savings in system development cost. However, although it has been investigated by more than 200 United States cor- porations, QFD has been successfully used at only a handful of these organizations [41]. In addition, a recent survey of 90 organizations adopting the technology showed that 83-percent of those organizations modified the details of the procedure to fit their particular needs [19]. While the structure provided by QFD can be of signifi- cant benefit, the tool can be very difficult to use. A decision support system (DSS) for QFD par- tially addresses this problem. One possible tool

for developing such a DSS is hypertext, which also allows organizations to customize the tool easily.

Hypertext dates back many years (to the Memex system described by Bush in 1945) and has been used to implement other system devel- opment tools [7]. The contribution of the City of Quality (Coq) system is neither QFD nor hyper- text but a melding of the two. Implementing QFD using hypertext allows the many persons involved in the system acquisition process to at- tach rationales, models, histories, computer-aided design (CAD) and computer-aided manufacturing (CAM) drawings, and voices wherever useful to the QFD documents; to retrieve relevant sections by author, keyword, or phrase; and to manage the overwhelming quantity of information associated with the development of a large, complex system that spans several years and involves a number of developers and end-users.

The purpose of this effort was to develop a complete set of QFD tools in hypertext, to re- search the potential of these tools to improve the life cycle of some large systems, and to investigate enhancements to manual QFD made possible by hypertext. Section 2 describes the QFD method- ology in more detail, along with its problems and limitations. Section 3 is a brief description of the working view of hypertext used for Coq imple-

o m F-

Z" , v,~ ,

DUSTOMER REQUIREMENTS ,z, o~ i

" Goes Fast

O ~ Large Payload

D. • Etc.

Reliable

i~ Easy to Fix n

Etc.

Easy to Find Spare Parts

PERFORMANCE

~, 6 o ~ ~;

, f

4

SUPPORTABILITY

o

F- "r C9 &U

Z &U :E

0

20

40

w #" lO

J 5

4 s

Fig. 1. Most commonly used QF D matrix. This matrix takes qualitative, end-user functional requirements and shows their relationship with engineering control parameters. The symbol ~ indicates that a given control parameter addresses the respective functional requirement. Requi rements and specifications shown are as entered by simulated users from the logistics community (MTBF is mean time between failure; M T T R is mean time to repair).

Page 3: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

Michael Wolfe / Development of the city of quality 301

mentation. Section 4 describes the actual imple- mentation and compares it with other hypertext system development tools such as glBIS. Section 5 discusses how this system may be used as a framework and interface in support of Taguchi methods, the analytic hierarchy process, or the Japanese management method of nemawashi / ringi. Finally, Section 6 discusses possible exten- sions and improvements.

2. Quality function deployment

2.1. Background

The history of QFD is somewhat obscure. The basic, generic ideas behind it are more than 40 years old, although the specific QFD structures have only recently been introduced in the U.S. An early English-language paper by Japanese au- thors Kogure and Akao [15] provides one brief history. This paper shows the first QFD reports appearing in Japanese in 1967, although with fewer than ten reports per year before the late 1970s. The article is somewhat confusing because it also states that QFD originated in 1972, when it was first put into use at the Kobe shipyards. In a later article, Akao claims to have developed QFD in 1966 [1]. His definition of QFD in that article is a system in which "responsibilities for producing a quality item must be assigned to all parts of a corporation."

He uses a chart, similar to that shown in Figure 1, to assign these responsibilities.

According to a second brief history by Claus- ing and Pugh [6], the generic ideas that developed into QFD have their roots in value-analysis/value engineering. Value-analysis/value engineering was popular in the 1950s; however, the specific QFD methodology has become popular in the U.S. only since 1986. This is not inconsistent with Akao's history.

A third history, according to Warfield [31] is that the fundamental, underlying QFD ideas were based on the work of Hall and Love, as enhanced by Hill and Warfield under the name Unified Program Planning (first published in 1972) [13]. However, this cannot be the case if QFD origi- nated in Japan in 1967, although the introduction of Unified Program Planning does coincide with the introduction of QFD at Kobe.

QFD, at least by that name, clearly reached the U.S. via Japan. However, its Japanese origins are uncertain. Although Akao claims to have introduced QFD in 1966, Schubert credits Mizuno with its development [22]. Regardless, it is cur- rently in use at several Japanese organizations and is being studied by many more.

According to Sullivan, QFD was developed as one set of tools to address part of the Japanese Industrial Standard Z8101-1981 called Com- pany-Wide Quality Control (CWQC). The Z8101- 1981 standard defines CWQC as "a system of means to economically produce goods or services which satisfy customers' requirements." Thus, Japanese CWQC is a superset of American Total Quality Management (TQM) [27]. QFD helps TQM by reducing the possibility that an essential aspect of quality will be overlooked in the strate- gic planning process. This relates to the work of Garvin [10] who notes that managers often over- look one or more crucial dimensions of quality in system designs.

While QFD cannot guarantee that no essential aspect of total quality will be overlooked, it can be a very important and useful tool for enhancing the system development process. It is well-docu- mented that using QFD methods for strategic planning can reduce development time by 50 percent for a wide range of products, services, and software systems while improving the quality of the final system [6,14,18,22].

It is not clear precisely what constitutes QFD. King defines QFD as, "a system for designing a product or service based on customer demands and involving all members of the ... organization

Narrowly defined, QFD refers to the organiza- tion that makes the design improvement possible. Broadly defined, QFD also includes the charts that document the design process. [14] Sullivan's definition is, "an overall concept that provides a means of translating customer require- ments into the appropriate technical require- ments for each stage of product development and production." [26]

For the purposes of this work, QFD is most easily understood as a formal structure for im- proving the system development process using charts similar to that shown in Figure 1. Notably,

Page 4: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

302 Michael Wolfe / Development of the city of quality

more than 80 matrices and charts have been identified as Q F D in one or more of the specific QFD methodologies. However, all the QFD methodologies have the matrix shown in Figure 1 in common as one of their charts or subcharts. Correctly employed, a chart such as Figure 1 can ensure that each customer requirement is ad- dressed by some design parameter and that no design parameter becomes part of the final speci- fication if it is not relevant to some customer requirement.

As simple as this chart appears, successfully using it can be a formidable task. At least four distinct subtasks are required to complete it. The first subtask is to determine and organize a "good" list of cu s t om er / end -us e r requirements, and to group these requirements into appropriate categor ies-as is done in the first two columns on

the left-hand side. The second subtask is to prior- itize these requirements using a column such as the one on the right-hand side. The third subtask is to develop a list of control parameters or specifications that determine the system or prod- uct design and to organize t h e m - a s is done in the top rows. The fourth and last subtask is to deter- mine which parameters influence which require- ments and to ensure each requirement has been addressed.

Following Kogure and Akao 's 1983 English- language article on QFD, a second article ap- peared by Sullivan in 1986 [271 which introduced another version of QFD based on four major documents. The first of these documents is an extension of Figure 1 with additional columns and rows. The other documents extend the struc- ture from a single phase of the system life cycle

I Support-

Performance ability

CUSTOMER ATTRIBUTES

Goes Fast

Etc.

Easy to Fix < Reliable

Etc.

Measurement Units

Our Product

Competitor A's Product

Competitor B's Product

Technical Difficulty Imputed Importance

Eetlmated Cost

Targets

Cuetomer Perceptions

1 2 3 4 5

Us )( )(

A • ...... •

B ~'-- - " --+

Fig. 2. The House of Quality. A printed template is shown with items entered by a simulated user from the logistics community. Items entered by the user are shown as typed (i n t y p e d font ).

Page 5: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

Michael Wolfe / Development of the city of quality 303

to other phases. A further, more detailed version of QFD was introduced by King, who calls his extension a "generic QFD." King's version uses approximately 30 matrices and charts. All three versions were primarily used by engineering man- agement. In 1988, Hauser and Clausing published a description of QFD in the Harvard Business Review [12] which introduced the methodology to general management. This methodology is also known as the Makabe Method [11]. King refers to the Makabe methodology as the "focused ap- proach" which is,

good for . . . parts and components, but ... awk- ward for computers, automobiles and other com- plex systems. It is good for minor improvements . . . but is not well suited for cost effective inno- vation [14].

In spite of King's criticism, the Makabe or Hauser and Clausing version of QFD is the one most familiar to general managers. It provides an overview of the entire project without overwhelm- ing the general manager with excessive detail. Consequently, the Makabe methodology was the starting point for the Coq.

2.2. Specific implementation

The Hauser and Clausing paper described a creative combination of QFD charts attributed to the Kobe shipyards in Japan. The combination is formed into a house-like structure, the "House of Quality," which they describe as a "conceptual map that provides the means for interfunctional planning and communications for quality function deployment" [12].

The House is shown in Figure 2. The labeled parts of the house are as follows.

• The rows in Area 1 include the customer at- tributes (functional requirements) organized into appropriate classifications. Also in this area is the column called "Relative Importance" for prioritizing these attributes. Capturing this "voice of the customer" is one of the most important contributions QFD makes to the de- velopment of successful products and systems.

• The columns in Area 2 are for technical speci- fications or engineering characteristics. The

first row of Area 5 contains the units used for these technical specifications. (This layout is taken from the original article [12].) Going from user requirements to technical specifications involves translating from the qualitative to the quantitative-a difficult problem referred to as "Feigenbaum's bottleneck" [8,17].

• Area 3, the "roof" of the house, captures the trade-offs between the various engineering pa- rameters. In the example shown, it is indicated that increasing the maximum velocity will ad- versely impact both the carrying capacity and the reliability of the final system. Conflict is indicated by X (the only situation shown). Weaker conflicts would be indicated by ×. Strong synergy would be indicated by v ~, and weaker synergy by va.

• Area 4 indicates the extent to which each end- user concern has been addressed by a design control parameter. The intersection of each technical specification column and customer requirement row forms a field in the middle of the house. These fields contain the correlations between the pairs. The same symbols are used as in the roof. Parameters that directly address a requirement are indicated by v~; parameters that weakly or indirectly address a requirement are indicated by v,'. Strong conflicts are indi- cated by × and weak conflicts are indicated by ×. In Figure 2, the user requirement "Relia- ble" is quantified as "mean time between fail- ures" (MTBF). A very high MTBF conflicts (weakly) with the user requirement of very high maximum velocity. Conversely, the maximum velocity specification directly addresses the cus- tomer's concern for a fast vehicle.

• Area 5, labeled "Objective measures," was originally designed to allow producers to evalu- ate their existing product lines against the com- petitors' using the technical measures identi- fied in Area 2. Area 5 must be adapted to the specific system being developed; Figure 2 shows how this area might be used to evaluate a large system for which there are only three produc- ers. The producer using the House is compar- ing his current system with those produced by his two competitors. However, the House may also be used when there is a much larger num- ber or much more nebulous set of alternatives. When the House is used for such situations, Area 5 must be redesigned.

Page 6: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

304 Michael Wolfe / Development of the city of quality

• Area 6 continues the idea of Area 5 by graph- ing, on a scale from one to five, how each alternative identified in Area 5 is perceived to address the functional requirements. The ex- ample shows how the producer 's current system ("Us") is rated a five on the criterion of "Goes Fast" but is much worse on other criteria (indi- cated by the line marked with ×). Likewise, for the "Goes Fast" criterion, Competitor B is worst with a rating of one, but excels in other areas (indicated by the line marked with +). When used in situations where Area 5 must be changed, Area 6 must be redesigned as well.

• Area 7 is used for the new design. Targets are set for all control parameters that determine the new design along with cost, technical diffi- culty (risk), and relative importance of achiev- ing each target. This provides management with a valuable means to direct resources by show- ing the experts' best estimates of the costs and benefits of each improvement in the product or current system design.

This methodology for QFD has been success- fully implemented using paper forms. These suc- cessful implementations have resulted in 50-per- cent cost and time savings in the system develop- ment process, as well as products and systems which better satisfy customer requirements. How- ever, the sources cited above also warn about failures in attempting to apply QFD.

2.3. L imi ta t ions

.

.

.

this was cited as the primary cause of failure by 67 percent of those who reported failure. Figure 2 is adapted from the original article by Hauser and Clausing, including the " E t c . "

(Only the specific requirements and engineer- ing characteristics were changed for our pro- ject-acquiring an Air Force weapons system). While the seven areas shown in the figure represent important considerations in structur- ing the system development process, it is not clear how to implement a real development project using a paper form based on this fig- ure. A paper form like Figure 2 has no provision for the justification of the entries; thus, there is no corporate record of the decisions that led to the final design. The paper forms offer no inherent provision for the fact that a group of users and designers are involved in the project and that the project will span many years. That is, there is no way to automate the search for specific items for further analysis or a review to ensure each user requirement has been adequately ad- dressed. A printed set of paper forms is relatively in- flexible; yet, vendors of such forms warn that " Q F D charts should not be copied . . . the charts need to be customized to an individual operation" [14]. When QFD is introduced by ordering a set of paper forms, the discovery that the forms must be modified for the orga- nization can be a major problem.

QFD was not originally implemented as a computerized group decision support system (GDSS), but as a set of charts and procedures. When these charts and procedures are imple- mented as a set of paper forms to be filled out, the QFD system can have certain serious limita- tions. Some of these limitations are:

1. Paper forms are inadequate for large numbers of requirements and specifications; this is the most common reason for failure. Paper forms cannot accommodate the 100 + requirements and specifications characteristic of real sys- tems. In addition, users cannot handle the cognitive overload of a 10,000 + cell matrix;

Partially because of these limitations, fewer than ten of 200 American companies that at- tempted to use QFD were able to realize its full potential [4]. A possible solution to these prob- lems is a DSS or group support system (GSS). These systems provide tools which help mitigate the problems listed above. Ideally, such a tool would be extremely flexible so that the interface could be adapted to the organization and the specific project. A DSS would be a single-user system that assists a decision-maker in imple- menting QFD by automating some of the work associated with QFD. A GSS would be a multi- user concurrent or asynchronous system that would allow all participants to access the system throughout the system life cycle. A possible vehi-

Page 7: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

Michael Wolfe / Development of the city of quality

Fig. 3. Linear writing.

305

cle for implementing such a DSS or GSS is hyper- text.

3. Hypertext

The concept of hypertext is credited to Van- nevar Bush, but the word was coined by Ted Nelson [23,30]. Pure hypertext is described as "an approach to information management in which data is (sic) stored in a network of nodes con- nected by links" [23].

Thus, hypertext is a nonlinear mode of writing. Ordinary written documents may be visualized as one-dimensional or linear graphs (Figure 3). The topology is given by the arrangement of pages from front to back in the document. Hypertext, on the other hand, allows the author to create an arbitrary topology on the 'pages of the document as in Figure 4. This topology may also be consid- ered a network in which each node represents a page (or screen) of text.

However, the value of hypertext is not this simple, two-dimensional structure which corre- sponds to text and footnotes; rather, the value of hypertext is its ability to develop documents that capture multidimensional ideas. In fact, if the document involves orthogonal concepts Ci, i = 1

. . . n, it follows that the logical topological struc- ture of the document is the n-dimensional space,

le-I c,. (a) i = l

Such a document cannot be captured canonically with ordinary text; canonical capture requires hy- pertext. The following example illustrates that certain topological constructs based on one-di- mensional paths cannot be easily mapped into one or two dimensions.

Example Consider three houses and three deliv- ery persons. The delivery persons may not cross each others' paths (delivery persons may cross their own paths), but each must visit all three houses. Draw the three houses and the paths the delivery-persons must take to do this.

Two dimensions are inadequate to contain cer- tain types of paths (e.g., the topological con- structs in question); therefore, it is impossible for all three delivery persons to deliver to all three houses without crossing paths. The puzzle can be solved if additional dimensions are allowed (i.e., if one of the delivery persons is allowed to fly).

Root Node Breadth

- - - 0

0 C~- O-~-O-~O-~,-<D O-J-O-~-O-~-C~,~D 0-~-(:~ 0,,~0

Fig. 4. Typical hypertext network.

Page 8: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

306 Michael Wolfe / Development of the city of quality

A well-designed hypertext document will con- form naturally to the underlying structure of the concepts being conveyed. Thus, it is more straightforward for readers to navigate and allows them to readily proceed along any dimension of the document. In addition, hypertext allows the users to search for arbitrary string patterns. This provides virtual links between pages that are re- lated in ways relevant to the readers, although this link may not have been foreseen by the author.

In his Memex, Bush envisioned a sophisticated version of the microfilm or microfiche reader in which any item in the world's library could be easily retrieved; however, the storage of informa- tion on electronic media encourages much more creative methods for communicating information. At a minimum, graphics and digitized sound are added if the hardware will support them; the result is called hypermedia.

Hypermedia is not limited to static text, graph- ics, and sound. In its most general form, each page may include one or more complete pro- grams. These programs are called message-han- dlers because they respond to messages from ei- ther the user or other message-handlers. Such a hypertext document is characterized by, in addi- tion to its text and graphics, a set M of messages and M' of message-handlers. [M' I > I M] since there may be, at various points in the document,

different message-handlers that respond to the same message.

For example, each numbered area of the House in Figure 2 is a message-handler that responds to some user action, typically a mouseclick or text entry. The message-handlers, which will be de- scribed later, allow the users to work on the subtask associated with that part of the House, or to see the results of that work. In general, hyper- media is a network connecting available elec- tronic resources in whatever manner the author finds most suitable for the presentation of infor- mation. Each page of this hypertext document can access the CAD/CAM programs, anima- tions, explanations, etc., necessary to convey de- sired information.

4. Hypertext implementation of the city of quality

4.1. Overview

The House of Quality, as an implementation of QFD, is part of an overall system for managing requirements over the system life cycle-from problem definition to implementation. The House helps ensure that the organization considers end-user requirements in the top-level design of systems and products. A complete system for QFD requires a number of Houses to provide for

The City of Quality

First Test

Horse el QuMity

For Help House Here

P~rts Frooess Production Deployment PILqning Plenning

Selec t a House f o r A n a l y s i s

User: [l'ltchael Wolfe I Home

Fig. 5. Top level of the Goq hypertext document. Each House accesses the subsystem of that name (e.g. clicking the mouse on the

Parts Deployment House accesses the Parts Deployment Subsystem). Help I?1 provides the user with assistance. Home exits the

Coq application. The user may also change the document title.

Page 9: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

Michael Wolfe / Development of the city of quality 307

intertemporal communication and to support multiple users of the system. The basic idea be- hind the City of Quality is to use hypertext to extend the House of Quality to an electronic collection of multiple, related Houses that span the entire system development life cycle.

With no electronic support, requirements management can be difficult. The systems devel- opment cycle of a complex project spans several years. By final production, some initial require- ments have been modified or deleted and new ones have been added. In addition, a few require- ments may have been overlooked. When the re- quirements are maintained manually, formal modifications are appended as written codicils to the original document, while implicit, informal modifications evolve throughout the development cycle without being recorded. Since the original list of requirements is usually quite long (several hundred or more requirements are common), the list with codicils appended is very difficult to follow. Consequently, it is difficult to determine if the final product meets the modified list.

The Coq allows managers to track each re- quirement. Thus, maintaining the set of require- ments becomes more explicit and manageable. A manager may electronically verify if any user re- quirement from the House of Quality (Figure 2) has not been addressed at each stage in the system life cycle. Conversely, the manager may select a difficult, risky, or costly implementation

issue in production, check which requirements are addressed, and determine the value of the issue to overall customer satisfaction.

4.2. Top level of the hypertext document

The top level of the Coq document is shown in Figure 5. This node in the document allows the users to identify themselves, set a title for the project, or access the subsystems. Clicking the mouse on the box labeled "User:" allows the user to change the name of the user. Clicking the mouse on the document title (initially set to "First Test") prompts for a new document title. Clicking the mouse on either of the four Houses accesses that subsystem.

The left-most House of Quality is somewhat different from the other three because it must support the seven separate tasks depicted in Fig- ure 2. In particular, this is where the "voice of the customer" is captured as qualitative, func- tional requirements and translated into quantita- tive engineering characteristics. The remaining three Houses do not obtain their left-hand-side data from the users, but rather from a message- handler that activates automatically when the House is viewed. The message-handler obtains the requisite data from the House to the left of the House being viewed. The remaining Houses are more likely to be accessed by engineering management, although the data entered are elec-

Home ~ Home

I ~ 1 , 1 ~ I ! I,I I i ,i I I i I I I I l "

I I

11 " ""

I I I +.l l l l l I ' , ' , I , I II II I •

I Jl II I

I m~,~ ~ ' - ~ , , I ' + " ' ~ h ' ~ i l l Ii Ii Iil I I ~ + ~ I I I I I I I I •

I " - I l l l l I

For H°~ ~ou+, Here H o u s e o f Q u a l i t y P~e+s for coq me.~

Fig. 6. The House of Quality as seen by the user. Text within the House is not necessarily legible, because this node is intended to be an icon which supports access to the seven QFD subsystems described in Section 2.

Page 10: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

308 Michael Wolfe / Deuelopment of the city of quality

Home ~ Home

CUSTOMER ATTRI BUTES Primar U Attr ibutes

¢9 For Help Mouse Here House of Qual i ty Press to return

Fig. 7. First screen for acquiring Customer Requirements of a new system. No requirements have been entered.

tronically related back to the House of Quality to support queries relevant to general management.

When the House of Quality is accessed, the user is transferred to the node shown in Figure 6, which is similar to Figure 2 (from which it was derived). Clicking on Home (which appears at every node) allows the user to exit the system. This particular implementation of hypertext (Hy- percard) automatically saves all data entered in the system. This prevents the user from acciden- tally losing data by failing to save. The arrow at

the lower right of the screen returns to the previ- ous node (in this case, the Coq main menu, shown in Figure 5). The House is actually divided into seven parts, each of which accesses one of the separate subsystems described following Fig- ure 2. For example, clicking the mouse anywhere in the section beneath the "Customer Require- ments" accesses a subsystem to capture the "voice of the customer."

Clicking the mouse anywhere to the right of "Engineering Characteristics" accesses the sub-

Home

CUSTOMER ATTRIBUTES Primarg Attr ibutes

Performance

~ Features | Reliability

-~ Conformance Durability Serviceability .,4 ,~ AEsthetics

Iv Perceived quality

10

5

20

10 15 15

5 10

Home

For Help Mouse Here House of Qual i ty Press to return

Fig. 8. Sample completed initial screen for acquiring Customer Requirements. Figure 7 has been filled in with Garvin's eight dimensions of quality. When used to support Garvin's approach to TQM, this is filled in for the users before initial use.

Page 11: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

Michael Wolfe / Development of the city of quality 309

system to capture the engineering experts' quan- titative characteristics for specifying the project under development. Other areas of the House access the appropriate subsystem as well.

4.3. Subsystems

Subsystem 1: Capturing the "Voice of the Cus- tomer"

Subsystem 1 in Figure 2 captures the "Voice of the Customers". This subsystem is designed to implement the first stage of multi-attribute, mul- tiple-person decision-making by acquiring the matrix of required attributes and relative weights for each decision-maker involved.

When the user elects to capture the customer requirements subsystem (by clicking the mouse anywhere near the "1" in Figure 2), a simple link is invoked to that part of the hypertext document. The display is shown in Figure 7.

This node has certain features common to most nodes in the document. The arrow in the lower right-hand corner returns the user to the node viewed immediately prior to the current one. The little houses return the user "Home"; of course, "Help" is available by clicking on the question mark.

Notably, all attributes must be organized in groups of eight, for reasons that will be explained below. The first eight attributes are the Primary Attributes. These are normally classes of require- ments rather than specific ones. When used for

TQM, this node is filled with Garvin's eight di- mensions of quality: Performance, Features, Reli- ability, Conformance, Durability, Serviceability, Aesthetics, and Perceived Quality [10] as shown in Figure 8.

A feature of this implementation is that the user is allowed no more than eight classes of requirements; each class may have no more than eight subclasses; etc. This prevents the growth of massive matrices that defy analysis. This hierar- chical approach is, by the 7 _ 2 rule, conceptually better for the user than a flat approach with many attributes or subattributes at the same level. The only limit on the total number of require- ments the document can accommodate is that imposed by secondary storage limitations.

As explained at the help page, the user is to select (with the mouse) one of eight primary slots. A blank slot is actually a message-handler (i.e., clicking the mouse activates a program). When the program is activated, a new node is added to the document and the user is transferred to it (Figure 9).

When the new node is created, a new arc is created between the node shown in Figure 7 and the new card. Thus, the entire program that created the card is replaced by a single arc to the new card. In this way, the hypertext document is self-modifying.

This new node, created as needed, illustrates two features of the Coq which enable use by multiple persons representing different customer

Home

Name of customer requirement

Weight of customer requirement

Person suggesting requirement

Person modifying requirement's weight

Requirement Level

(Display Comments) CRttoch comment) (History of Changes)

For 14e~ ivlouse I.bre

Fig. 9. Node to acquire a Specific Requirement.

Press to return

Page 12: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

310 Michael Wolfe / Development of the city of quality

interests. A non-disruptive annotation feature is provided by the "Attach Comment" message- handler. Each user is allowed to attach one or more comments. Features for editing comments, viewing other user comments, etc., are available. The only restriction on these comments is that imposed by secondary storage. This is listed as an important feature in hypertext systems [2], but has previously been available only in much more expensive systems than the Coq.

A second feature is the History. A record is kept of every modification and the person re- sponsible for it. The system tracks the user, al- though this implementation uses an "honor" sys- tem since password checking is not employed. Instead, the user is asked to provide a "real" name for tracking purposes. A password system may be added if the Coq is used in a more hostile environment.

The downward-pointing arrow at the bottom of the screen is a very important feature of the Coq. It is an integral part of the scheme that allows for the addition of an unlimited number of user requirements by using recursion. If this card is used for a class of requirements rather than a specific requirement, the downward-pointing ar- row will indicate the subclasses or its members. When first accessed, the arrow creates a new node similar to the one shown in Figure 7; then the program that created the node is deleted and

replaced with an arc to the new node. However, note that Figure 9 creates a node like Figure 7, which, in turn, is capable of creating eight new nodes like Figure 9. The result is a recursively defined, self-modifying document that allows the hierarchy of classes of requirements to be easily extended to an arbitrary depth.

Note that the user is not allowed to enter data directly into the fields of the nodes like Figure 9. Instead, the Coq has message-handlers which, when activated by a mouseclick, start a program to prompt for requirement and weight. The pro- gram then automatically updates the fields for originator, modifier, and history, using the user's name (which is obtained before the system may be accessed). While this program currently passes user inputs directly to text fields, this processing allows future enhancements to check inputs, per- form calculations, and pass messages to a local area network (LAN).

Subsystem 2: Capturing the Engineering Specifi- cations

The second subdocument of the Coq, corre- sponding to Areas 2 and 7 of Figure 2, acquires the technical specifications for the system under development. It is similar to the subdocument for acquiring customer requirements, although some-

m

~ m _zE

IZ W(b

,~_.~ ,Ibl~ O I

TIohnio - I tilfioulty

Imputid i I I p o I t l O l

Estimlted Cost

TIrgets

Logistics Characteristics

~ "~ ~ ~ ..

" ~ ~ o

Hour hours dags ~ kgals galsl 9rade ~; $ /hou hr ¢

7 18 10 5 10 10 4 10

17 3 10 15 6 5 50 I0

~ 5 25 10 10 9 10 10

10 90 75 5 3 5

Home

2

95

House of O u a l i t y <~ For Help Mouse Here Press for House

Fig. 10. Screen for acquiring Technical Specifications (based on Area 2 of Figure 2). This screen acquires specification parameters and information useful for managerial design decisions. Engineering characteristics and values are shown as they might be typed by an actual user (LRU is line replaceable unit).

Page 13: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

Michael Wolfe / Development of the city of quality 311

what different information is captured. Again, this is a self-modifying document which uses re- cursion to capture a list of technical specifications limited only by the availability of secondary stor- age. The node for acquiring these specifications is shown in Figure 10.

When this screen is implemented as a paper document, Estimated Cost, Difficulty, etc., are entered as numbers. These numbers may be based on some calculations performed separately. Im- plemented in hypertext, cost and difficulty may be modeled as functions of target value, and the model embedded directly in the document. The model would then be "deep knowledge," while the system displays the numerical value as "surface knowledge" to support managerial deci- sion-making. This storage of deep knowledge in the DSS and display of surface knowledge is another important enhancement over paper im- plementations [5]. Likewise, overall cost may be automatically computed and sensitivity analysis provided.

Another drawback when using a paper imple- mentation of QFD is that a manager must "eyeball" the numbers to select the best design. When implemented as an electronic document, the manager has immediate access to a variety of decision support tools which may be used to help select compromise designs.

Subsystem 3: The Roof The third area of Figure 2 records the trade-

offs between the various engineering characteris- tics. This is captured via a roof (Figure 11).

The user selects a square on the roof if the characteristics interact either synergistically or antagonistically. The basic QFD methodology only records these relationships on a scale of -2 to 2; the former is strong antagonism and the latter is strong synergy. Initially, when the user selects the roof of the house, top-level classes of characteristics are selected. Clicking the mouse on a class of characteristics brings up the specific roof based on the individual members of the class. For example, Figure 11 is reached by click- ing on a column for Logistics characteristics, which is part of a roof of general specifications such as Logistics, Performance, Cost, etc. This process can be repeated until the lowest-level roof is reached with trade-offs between specific characteristics. Likewise, one can move to more general roofs by clicking on the name of the class, listed above the eight members.

The roof is one of the great strengths and weaknesses of QFD. In the paper implementa- tions, it grows quadratically in the number of characteristics. A complex, difficult decision is thus reduced to a large number of small, manage- able decisions; however, the total number of small

Home

k-

_1=

IE.¢

I B Z UJ~

For Help House Here Press for House House of Quality

Home

Fig. 11. "Roof' of the House of Quality showing trade-offs between various Engineering Characteristics. Characteristics are as typed by a simulated user.

Page 14: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

312 Michael Wolfe / Development of the city of quality

Home

CUSTOMER ATTRIBUTES

Performance

Features

Reliability

Conformance

Durability

Serviceability

/Esthet ics

Perceived qual i ty

Logistics Characteristics

Home

~H

~= ~

X X X X X X

Press for House For HelpMouseH~ro

Fig. 12. Measure of which Engineering Characteristics Address which Customer Requirement (using the notation indicated in Figure 2).

decisions may well become unmanageable. In the Coq implementation, the screen can display a maximum of eight characteristics; this restricts the user to 24 trade-offs at one time. Users are not encouraged to compare each engineering characteristic against every other; rather, they should perform the more relevant comparisons. Nevertheless, the number of decisions can easily become formidable in a large system. However, while no single system or tool can completely resolve all the problems inherent in unmanage- able complexity, the Coq provides one approach to this problem and can be a solution in many cases.

Subsystem 4: The Central Matrix The fourth area of the House records the

interactions between requirements and specifica- tions, as illustrated in Figure 12. These values (again from - 2 to 2) are displayed as described following Figure 2. Because this is a hypertext document, each non-zero value may be explained on a card similar to the one used to acquire customer requirements. A hypertext implementa- tion can also accept the values for these fields either as a numerical value, a simple formula, or a complex computer program. In addition, com- ments such as those used in the requirements definition phase may be pasted on at will. The manager need only see the final result, but deep

knowledge potentially lies behind every surface number.

As with the roofs when the central matrix is first selected, the top-level classes of attributes and characteristics are displayed. However, by selecting a class with the mouse, the user may select the interaction matrix for the members of that class. As with the other subsystems, this enables the entire system to manage a quantity of data limited only by secondary storage capabili- ties.

Again, as with the roofs the total number of cells to be filled grows quadratically with the overall size of the problem (i.e., number of rows multiplied by number of columns). Thus, a single, monolithic, unmanageably complex decision is broken down into a number of small, manageable decisions. But, as with the roofs the number of these decisions can become unmanageable. The Coq attempts to avoid this explosion of small decisions by limiting the user to an 8 × 8 matrix; however, this is not always successful. Large, complex systems usually remain large and com- plex. If the design guidelines embodied in Suh's Axioms are followed, design parameters are cho- sen with minimal interactions so that this central matrix remains upper or lower triangular with few off-diagonal elements [25]. Thus, the process of filling in the Coq encourages a design that respects Suh's axioms which, in turn, leads to a

Page 15: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

Michael Wolfe / Development of the city of quality 313

design that is easily matched to the Coq capabili- ties.

Remaining subsystems

Of the seven subsystems of the first House discussed in Section 2, two subsystems, corre- sponding to Areas 5 and 6 of Figure 2, must be adapted to each organization using the House. The remaining five subsystems have been dis- cussed above (the subsystem corresponding to Area 7 was discussed with the subsystem corre- sponding to Area 2). As illustrated in Figure 2, the Areas 5 and 6 of the House allow users to compare existing designs and to display these comparisons graphically. These subsystems have not yet been implemented in the current version of the Coq. When implemented, these two parts will help decision-makers evaluate existing sys- tems and use these evaluations to guide the sys- tem development process.

4.4. Comparison with other hypertext tools

Another class of hypertext tools to support system development has been developed based on the methodology called Issue-Based Informa- tion System. One tool, glBIS, uses a graphical approach to capture the type of system issues addressed by QFD [7]. A newer version called rlBIS is a real-time implementation [20]. The older glBIS is intended to support generic system design. The implementation is a more expensive, less portable version than the Coq because it requires specialized hardware and software and has seen little use outside of the organization where it was developed. The Coq, conversely, runs on the low-cost Macintosh Classic TM. In principle, glBIS is equivalent to the Coq, al- though the flavor and emphasis are very differ- ent.

While an empirical comparison of the two systems might be of some limited interest, both are basically different approaches to a common problem, glBIS is based primarily on graphs; the Coq is based primarily on matrices. The equiva- lence is easily seen.

Theorem Graph-based representations are infor- mationally equivalent to matrix-based representa- tions.

Proof Any graph may be captured by a node-in- cidence matrix. Thus, the matrix approach is at least as strong as the graph-based approach. Con- versely, while not all matrices are node-incidence matrices (and thus not so easily captured by di- rect transformation to a graph), one may imagine a graph as a collection of labeled arcs (one for each cell in the matrix) with the contents of the cell as the label, the start of the arc as the row identifier, and the end of the arc as the column identifier. Hence, graph-based information repre- sentations are at least as strong as matrix-based representations. Thus, the two representations are informationally equivalent. QED.

Thus, both the glBIS graphical and the Coq matrix representations are capable of capturing the same data. The preferred system is a matter of decision-maker style. Thus, it would not be particularly informative if, for example, 45 per- cent of decision-makers preferred the Coq and 55 percent preferred glBIS or vice versa. There is no compelling reason to deprive graphically oriented decision-makers of a graphical tool, or to deprive matrix-oriented decision-makers of a matrix tool just because one class of decision-makers is in the majority. The issues of cost and availability are, however, more likely to make the Coq more de- sirable in many situations.

The more recent implementation of IBIS, rlBIS, is not a real competitor to the Coq. rlBIS is described as a system that specifically supports software system design [20], while the Coq is intended to support generic system design.

4.5. The Coq as a group decision support system

As a GDSS, the version of Coq described above is a low-cost system which may be dis- tributed to operational and middle managers as well as to the engineering staff responsible for entering numbers and models in the various parts of the city. To achieve this low cost, the Coq was implemented as an asynchronous system with communication via "sneakernet" (i.e., by some- one carrying around the floppy disk). Where re- sources such as a LAN are available, a number of enhancements are possible. These are described in Section 6 on possible extensions to the Coq, especially Section 6.2.

Page 16: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

314 Michael Wolfe / Development of the city of quality

5. Potential uses of the city of quality

5.1. A template to support Taguchi methods

The Coq is well-suited to support a simplified variant on the modified Taguchi approach for selecting the best design. The original Taguchi approach was to minimize the cost to the system end-users. End-users are assumed to have target values which, if achieved, allow them to use the system without incurring any additional cost. Given technological limits, this ideal system can- not be achieved. For example, an ideal vehicle would require no fuel whatsoever, provide zero risk of accident, etc. Thus, the ideal system is a vector ~- of target parameter values, while an actual system is a vector of parameter values t ~ T, where T is the set of all systems possible within the technological constraints. These pa- rameters are not truly comparable-e.g., it is not really possible to quantify the relative value of fuel efficiency versus safety-however, the Taguchi approach is to use a quadratic approximation. The cost of a unit deviation from each target T i is given by Cui. The total unit cost to the end-user is then estimated as

C u ( t ) = E C u i ( T i - t i) 2. (2) i

The Taguchi approach is to produce a system t that minimizes Equation (2).

A more realistic approach includes both the end-user's cost and the developer's cost Cd(t) of producing the system [29]. This modified Taguchi approach is based on minimizing total cost,

Cu(t) + ca(t) , (3) rather than just end-user's cost.

The Coq may be used to capture the data and perform the necessary computations to support the modified Taguchi approach. The row Esti- mated Cost may be used to record components Cdi of C a. The producer's cost is then,

C a = ~_,Cai(ti). (4) i

A row must be added (or the row Imputed Importance used) to record the consumer's unit costs Cui.

The Coq methodology indicates that Equation (4) is not adequate for a quadratic approximation

because it omits the cross terms (i.e., the syner- gies and antagonisms captured by the roof of the house). The correct quadratic approximation is,

Cd(t) = ~-,Cd(ti) + E ECiy(ti, tj), (5) i i j

where the coefficients Ciy represent the cross terms (synergies and antagonisms) between char- acteristics i and j. Synergies are captured as negative cii and antagonisms as positive cij.

The basic Coq template must be further modi- fied. Instead of recording synergies and antago- nisms on a Likert scale of - 2 to 2, more accurate cost estimates must be recorded or a model must be developed to provide an estimate of these costs from the Likert values. In addition, the cost components Cu(t i) must be recorded rather than just Imputed Importance.

Finally, a module must be added to compute C u + C d and the target ~'* that minimizes Cu + Ca, but this is a "straightforward" quadratic pro- gramming problem.

5.2. A Template to support Nemawashi / Ringi

Japanese group decision-making often relies on nemawashi/ringi, a process in which a coordi- nator (ritsuan-sha) brings the group to a consen- sus. This process may be assisted by the Coq. A complete discussion of nemawashi/ringi is be- yond the scope of this paper but is detailed by Watabe, Holsapple, and Whinston in their paper [32].

Briefly, the final stage of the nemawashi/ringi process (ringi) consists of circulating a formal document that describes the decision made among all group members. At this point, unanimous consensus is almost certain because of the pre- ceding informal negotiations spearheaded by the ritsuan-sha. This informal negotiation process-the nemawashi-normally consists of five steps:

1. Information collection (joho shyushyu), 2. Data analysis and alternative generation

( ritsuan ), 3. Tentative plan selection by the coordinator

( sentaku ), 4. Negotiation and persuasion (nemawashi), and 5. Formal document circulation (r/ng/).

Page 17: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

Michael Wolfe / Development of the city of quality 315

These five steps have traditionally been per- formed by Japanese decision-makers without us- ing technological support; however, the process can be enhanced by such support. Tools that would help include ad hoc models for computing the group rankings of the alternatives chosen. These ad hoc models will be referred to as the group preference aggregation models, or group models. By Arrow's theorem [3] no group model will work under all circumstances; however, if a consensus is attainable, there must exist some model that will yield that consensus as a function of the individual preferences.

To apply the group models, the coordinator must obtain a matrix of decision-maker require- ments and weights, a matrix evaluating each al- ternative against each requirement, a vector of decision-maker relative influence, and a vector of decision-maker difficulty of persuasion. Then, us- ing Multi-Attribute Utility Theory (MAUT) or some other model of individual preferences, a matrix M giving the cardinal rating of each alter- native for each participant is obtained. Thus, a group model is a mapping q~ from M to a vector g giving the group preference,

c b : M ~ g . (6)

A DSS can help a facilitator apply each group model to compute candidate group preference functions for consensus.

With the assistance of a DSS, the ritsuan-sha selects the group model most appropriate for the group and decision under consideration. The original nemawashi/ringi paper suggests six broad classes of group decision models along with several variants of each class. A detailed descrip- tion of each model is beyond the scope of this paper; however, all these models may be imple- mented in hypertext, and the system allows the coordinators to select the models they deem most appropriate.

In the simplest version of the Coq described above, all members of the group attach their contribution to a single hypertext document exist- ing on a single floppy disk which is circulated via "sneakernet." The ritsuan-sha can then use this document, with its detailed history of contribu- tions from preliminary nemawashi, to obtain a candidate for ringi.

Using the Coq for nemawashi/ringi requires the addition of special nodes for the ritsuan-sha.

For example, the ritsuan-sha must have a special card with links to each group model which may be used to obtain a proposed consensus. Basi- cally, the existing Houses of Quality allow the ritsuan-sha to obtain matrices M i for each expert i. Rows of Mi correspond to technical specifica- tions, s, while the columns of M i correspond to proposed and existing designs, d. Thus Mi(s, d) is a measure of the extent to which design d addresses specification s. A second matrix, N~, has user requirements as rows, r, and technical specifications as columns, s. N/(r, s) indicates the extent to which specification s addresses require- ment r.

The ritsuan-sha also has a vector R~ for each end-user or decision-maker j, indicating the weight j put on each requirement; vectors I and P indicate the influence and pursuadability of each end-user or decision-maker; and the set D indicates alternative designs.

When the appropriate group model is selected, it accesses a mapping ~h(M, N, R, I, P) : D ~ which quantifies the overall utility of each design.

The message-handlers for the coordinator have not been implemented at this time but are a reasonable extension of the Coq.

5.3. The City of quality as a template for the analytical hierarchy process

In combining individual preferences to form a group preference, cardinal preferences offer an alternative to ordinal preferences. This is theoret- ically more appealing-specifically, Arrow's theo- rem shows that there is no logically consistent group preference aggregation function which combines individual ordinal preferences into a group preference, However, there are a variety of logically consistent methods for combining cardi- nal utilities to produce a group utility (e.g. weighted averages).

Typically, however, cardinal preferences ob- tained from decision-makers are inconsistent. The analytic hierarchy process (AHP) is one method for obtaining an estimate of cardinal preferences from a set of "noisy" preferences provided by the decision-makers [21]. The AHP as specified is logically inconsistent and, therefore, seriously flawed as a tool for computing these cardinal preferences [9,28]; nevertheless, it has a place in the Coq.

Page 18: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

316 Michael Wolfe / Development of the city of quality

Given Arrow's contention that cardinal utili- ties cannot be obtained and his theorem that ordinal utilities cannot be combined, nothing is a panacea for resolving group conflicts. However, any tool that provides a starting point for reach- ing consensus is potentially useful; the AHP can be used for this purpose. The AHP provides a "strawperson" which the group facilitator can propose as a basis for further discussion. The AHP is attractive, when used in this manner, and the Coq can be readily modified for use with it. This is evidenced by the basic hierarchical char- acter of the storage and organization of the user requirements in the Coq. This hierarchical struc- ture can provide the hierarchy necessary to AHP. The second requirement for AHP is a system to rapidly make the comparisons between each at- tribute. This, in turn, may be done by modifying the existing "roof" structure to include a module that captures the relative utilities of the user requirements needed by the AHP. This is planned for a future extension of the Coq.

6. City of quality limitations and directions for further research

6.1. Limitations

The Coq has several limitations: it is currently limited to one user at a time, the main House of Quality does not fully support the QFD method- ology, and the other Houses in the City are not fully functional.

The Coq is currently implemented as an asyn- chronous, multi-user system for QFD; thus, users must wait to see each other's ideas. Ideally, users will be able to access the Coq from their desktop computers at any time. This would include fully concurrent access.

A second problem is that, as implemented, the House of Quality does not completely provide for the QFD methodology. In particular, all customer requirements should be acquired before they are ranked, and the complete list of specification parameters should be obtained before any target values are assigned to the parameters. The cur- rent system combines acquisition and quantifica- tion for simplicity. Planned enhancements will support at least four separate brainstorming exer- cises: (1) acquiring the list of end-user require- ments; (2) prioritizing those requirements; (3) ac-

quiring the list of design parameters; and (4) setting targets for those parameters.

Finally, as planned, the other Houses in the City that follow the House of Quality should turn system specifications into subsystem requirements and carry them through process and production planning in an automated fashion.

The completed Coq should:

1. automatically flag any user requirement that has not been addressed through final produc- tion;

2. automatically flag any technical specification for the system, subsystem, process or produc- tion that is not needed to address a specific user requirement; and

3. provide modules for sensitivity analysis that allow decision-makers to see immediately the impact of budget restrictions on requirements, the impact of relaxing target values on cost, etc.

The result would be a strategic planning tool that would help prevent systems in which crucial requirements are overlooked, and unnecessary, excessive costs arise from meeting irrelevant spec- ifications.

As originally envisioned, the current imple- mentation of the Coq would ultimately provide such a system, running concurrently on dis- tributed central processing units (CPUs). How- ever, the current version is limited by awkward data structures. Much of the data are redundant. Data displayed at various nodes of the document are kept at all nodes where displayed, and are copied between nodes by conventional imperative programming. These data redundancy and man- agement problems caused a revision in the origi- nal plan. The current system has become a proto- type and test-bed that demonstrates the feasibil- ity of a sophisticated hypertext GSS for QFD, provides potential users with a better understand- ing of what the final system might look like, and illustrates the need for extensions and further research.

6.2. What You See Is What I See extension to the Coq

An exciting possibility in computer-supported cooperative work (CSCW) is the situation in which various group members can see exactly what is on

Page 19: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

Michael Wolfe / Development of the city of quality 317

each others' computer screens. This concept is called WYSIWlS (What You See Is What I See) [24]. If resources are available in the form of a networked set of computers all running Hyper- card, this concept can be extended to the Coq. The planned implementation is a relatively low- cost version. The Coq can be extended with two message-handles named ATP and NBP that han- dle the transport protocol and name binding on the LAN. When these utilities are installed, all users must start identical copies of the Coq and must enable the network communication feature. This allows each copy of the Coq to post to and receive from the network. Note that ATP only posts and receives messages; if the user inputs text or mouseclicks (for graphics) into the Coq without using a message-handler, communication will not ensue.

To extend the Coq to WYSIWlS, each Coq message-handler must be modified by the addi- tion of the line, A T P s e n d " ' m e s s a g e " "

which sends a copy of "message" to the network. In addition, the idle message-handler must be modified to include the lines, A T V r e c e i v e " ' m e s s a g e " "

s e n d " ' m e s s a g e " "

which check the network for messages and pass them to the Coq. As each message is entered, it will traverse the network updating every machine, and the system will have WYSIWlS.

6.2. Relational database revision of the Coq A relational database management system

(RDBMS) is required to achieve many Coq objec- tives. As implemented, the Coq data structures are not easily managed and do not readily sup- port a comprehensive system to manage a com- plex project over its entire life cycle. By storing parameters on a central RDBMS, a planned fu- ture version of the Coq should be able to manage the complexity inherent in a large system devel- opment project while allowing multiple partici- pants concurrent access.

7. Conclusions

The Coq developed to support quality function deployment over the system life cycle, is intended

to provide this support for general systems. The specific examples used in the illustrations were geared for an Air Force system and addressed logistics concerns; however, the Coq is adaptable and can be modified for other types of system development projects, such as consumer product development or software system development.

To achieve quality in a system, a number of decision-makers and technical experts must be involved, and their contributions recorded. Over several years, a retrievable record of decisions made and their rationales becomes vital. Paper documents are deficient in maintaining a good record of group interactions and providing facili- ties to easily retrieve, for example, all documents relating to a specific user requirement.

The Coq addresses some of these issues. It makes acquisition and retrieval of the informa- tion a simple matter of a few mouse clicks and text string entries. It also allows for recording models and explaining all decisions. In addition, the Coq provides a low-cost system to support certain kinds of group interactions, thereby pro- viding some GDSS support to lower and middle managers.

Thus, the Coq could provide the structure and technology io make this information available to the general managers throughout the system life cycle. The system could provide support for strategic planning at the inception of each major system, and support for requirements manage- ment, coordination, and control throughout the development process.

Acknowledgements

I wish to thank the Air Force Office of Scien- tific Research for sponsorship of this research, and the Logistics Research Acquisition Branch of Armstrong Laboratory for their cooperation.

Many people were very helpful, and made this research project both enjoyable and productive. Special thanks are due to Capt Hill, Ms. Edly, Capt Moen, Ms Peasant and Ms Stiller for their assistance with my research in various ways. Thanks are also due to Ms. Campbell and Mr. Hoffman for their concern and interest in my project.

Finally, I have to thank Ms Wheeler for her invaluable editorial assistance, without which I

Page 20: Development of the city of quality: A hypertext-based group decision support system for quality function deployment

318 Michael Wolfe / Development of the city of quality

could not have completed the writing of this paper.

References

[1] Y. Akao. Foreword. in: B. King, Better Designs in Half the Time. Methuen, MA: GOAL/QPC; 1989.

[2] Akscyn, Robert M., et al., KMS: A Distributed Hyperme- dia System for Managing Knowledge in Organizations. Communications of the ACM. 31(7) 820-835, 1988.

[3] K.J. Arrow. Social Choice and Individual Values (2nd Ed.). Cowles Foundation for Research in Economics at Yale University: John Wiley and Sons Inc., 1963.

[4] C.C. Biddle. Using Interactive Management (IM) For Defining and Solving Complex Problems. Presented at the Government Group Decision Technology Confer- ence, Defense Systems Management College, 1991.

[5] A-M. Chang, C.W. Holsapple, A.B. Whinston, Decision Support System Theory. West Lafayette, IN 47907: Man- agement Information Research Center Krannert Gradu- ate School of Management Purdue University, 1988.

[6] D. Clausing and S. Pugh. Enhanced Quality Function Deployment. In Design and Productivity International Conference. Honolulu, HI, 1991.

[7] J. Conklin and M. Begeman. glBIS: A Hypertext Tool for Exploratory Policy Discussion. ACM Transactions on Of- fice Information Systems, 6(4), 303-331, 1988.

[8] D.J. Duke, G.M. Nijssen, and S.M. Twine. The Entity- Relationship Data Model Considered Harmful. In the Sixth Symposium on the Empirical Foundations of Infor- mation and Software Science, 1988.

[9] J.S. Dyer. Remarks on the Analytic Hierarchy Process. Management Science.; 36(3), 249-258, 1990.

[10] D.A. Garvin. Competing on the eight dimensions of quality. Harvard Business Review, 65(6), 101-109, 1987.

[11] R.F. Hales. Quality Function Deployment (QFD). Pre- sented at Quality Function Deployment Presentation and Demonstration, 1991.

[12] J.R. Hauser and D. Clausing. The House of Quality. Harvard Business Review, 66(3), 63-73, 1988.

[13] J.D. Hill, and J.N. Warfield. Unified Program Planning. IEEE Transactions on Systems, Man, and Cybernetics, 2(5), 610-621, 1972.

[14] B. King, Better Designs in Half the Time. Methuen, MA: GOAL/QPC, 1989.

[15] M. Kogure and Y. Akao. Quality Function Deployment and CWQC in Japan. Quality Progress, 16(10): 25-29, 1983.

[16] G.A. Maddux et al. Organizations Can Apply Quality

Function Deployment As Strategic Planning Tool. Indus- trial Engineering, 23(9), 33-37, 1991.

[17] D. Michie. Inductive Rule Generation in the Context of the Fifth Generation. In Proceedings of the International Machine Learning Workshop, 65-70, 1983.

[18] T.H. Miller. Quality Force Deployment. Program Man- ager, 32-37, September-October, 1990.

[19] A. Panday and D.P. Clausing. QFD Implementation Sur- vey Report. Cambridge, MA: Laboratory for Manufactur- ing and Productivity, 1991.

[20] G.L, Rein and C.A. Ellis. rlBIS: a real-time group hyper- text system. International Journal of Man-Machine Stud- ies, 34(3), 349-367, 1991.

[21] T.L. Saaty. The Analytic Hierarchy Process. New York: McGraw-Hill, Inc., 1980.

[22] M. A. Schubert. Quality Function Deployment-A Com- prehensive Tool for Planning and Development. In NAE- CON 89 (CH2759-9/89), 1498-1503. Piscatawny, NJ: IEEE, 1989.

[23] J.B. Smith and S.F. Weiss. An Overview of Hypertext. Communications of the ACM, 31(7), 816-819, 1988.

[24] M. Stefik et al. Beyond the Chalkboard: Computer Sup- port for Collaboration and Problem Solving in Meetings. Communications of the ACM, 30(1), 32-47, 1987.

[25] N.P. Suh. The Principles of Design. New York: Oxford University Press, 1990.

[26] L.P. Sullivan. Quality Function Deployment. Quality Progress, 19, 39-50, 1986.

[27] L.P. Sullivan. The Seven Stages in Company-Wide Qual- ity Control. Quality Progress, 19, 77-83, 1986.

[28] E. Triantaphyllou and S.H. Mann. An Examination of the Effectiveness of Multi-Dimensional Decision-Making Methods: A Decision-Making Paradox. Decision Support Systems, 5, 303-312, 1989.

[29] E. Tse and W.E. Cralley. Management of Risk and Uncertainty in Product Development Processes, Draft Final Report, Contract Number MDA 903 84 C 0031, Task Order T-D6-553, 1988.

[30] T. Vaughan. Using Hupercard ®. Carmel, IN: Que ® Cor- poration, 1988.

[31] J.N. Warfield. [Letter to the Editor, Harvard Business Review]. Institute for Advanced Study in the Integrative Sciences, George Mason University; 1989.

[32] K. Watabe, C.W. Holsapple and A.B. Whinston. Coordi- nator Support in a Nemawashi Decision Process. Gradu- ate School of Business, The University of Texas, Austin, TX, 1988.

[33] R.T. Watson, G. DeSanctis, and M.S. Poole. Using a GDSS to Facilitate Group Consensus: Some Intended and Unintended Consequences. MIS Quarterly, 12(3), 463-477, 1988.