CONTENTS

1. Unilever Context

1.1 From Knowledge Engineering to Knowledge Management

1.2 From KBS Development to Knowledge Mapping

2.1 Visualising Knowledge in Graphical Format

2.2 The QFD Knowledge Framework

2.3 A Causal Knowledge Framework

3.1 Conventional Knowledge Acquisition

3.2 Group Knowledge Acquisition

3.3 Group Knowledge Acquisition in Knowledge Workshops

4.1 Knowledge Acquisition Worksbenches

4.2 PC PACK

4.3 Meta PACK

4.4 Advantages and Limitations of PC PACK and Meta PACK

5.1 Knowledge Workshop Deliverables

5.2 How to run a Knowledge Workshop?

5.3 Sharing Tacit Knowledge

5.4 Conclusions and Lessons

6.1 Strengths/Weaknesses Knowledge Map

6.2 Opportunities/Threats Knowledge Maps

6.3 Conclusions Knowledge SWOT Analysis

Acknowledgments

References
 

1. THE UNILEVER CONTEXT  [Table of Contents]

1.1 From Knowledge Engineering to Knowledge Management

Unilever is an Anglo-Dutch fast moving consumer goods company with headquarters in London and Rotterdam. Its portfolio covers foods as well as home and personal care products. With manufacturing operations in more than 90 countries and about 300,000 employees, the company has a turnover of more than $50 billion. In 1996, investments in basic research and product innovations exceeded $934 million, leading to the filing of more than 400 patent applications.

This company with its wealth of knowledge forms a challenging arena for knowledge-focused applications. As a result, a lot of attention has been paid within Unilever to KBS development in the last decade. In particular, many diagnosis and assessment systems were developed in various application areas, ranging from manufacturing to product and process development (see for example (Speel & Aben, 1997; Speel & Aben, 1998)). These applications were developed using knowledge engineering methods, techniques and tools for the acquisition, modelling, representation and use of knowledge (Schreiber et al., 1999).

As knowledge is the key source for sustainable competitive advantage (Nonaka & Takeuchi, 1995; Davenport and Prusak, 1998), Unilever is paying increasing attention to knowledge management. This means optimizing creation, dissemination and exploitation of both explicit and tacit knowledge to create business value and compete with knowledge. For example, the company stresses the importance of knowledge sharing which is mentioned explicitly in Unilever's Corporate Purpose Statement:

Our deep roots in local cultures and markets around the world are our unparalleled inheritance and foundation for our future growth. We will bring our wealth of knowledge and international expertise to the service of local consumers — a truly multi-local multinational.

1.2 From KBS Development to Knowledge Mapping

In the last decade, many KBSs have been developed at Unilever. One of the key activities of KBS development is knowledge structuring on a conceptual level (Schreiber et al., 1999; CommonKADS web site). We define knowledge structuring as the techniques and tools that bring order in a knowledge area through a representation in a proper, systematic form. Knowledge structuring results in knowledge models in which knowledge has been transformed into a more explicit form.

In order to validate knowledge models and pass them through to KBS designers, it is important to present them in a comprehensive yet transparent form. A graphical format has proven to be very helpful for this purpose. In the last decade of KBS development at Unilever, these graphical knowledge presentations also turned out to be very useful in their own right. It was not unusual for intermediate deliverables to become even more important than the KBS end deliverable. Therefore increasing attention has been paid to the creation of graphical knowledge models, also called knowledge maps. In particular, high-level, business-focused knowledge maps have become popular in Unilever.

2. KNOWLEDGE MAPPING  [Table of Contents]

2.1 Visualising Knowledge in Graphical Format

Presentation of knowledge in a graphical format is also called knowledge visualisation (e.g., Tufte, 1983; Tufte, 1990) . The field of knowledge modelling has a long tradition in visualising knowledge; for example, the universally quantified graphs of KL-ONE (Brachman & Schmolze, 1985), conceptual graphs (Sowa, 1984) and inference structures (Clancey, 1985). These graphical formats are very useful for communication purposes between human beings. Note however that graphical representations may also cause confusion when their semantics are not properly defined (Woods, 1975).

Visualised knowledge models are used by knowledge engineers to discuss modelling techniques, to validate knowledge (with experts), and to specify functional requirements of KBSs (with KBS developers). Additionally, in our experience certain visualised knowledge models can be very useful for business managers as well. Since management is usually more focused on high-level issues (as opposed to detailed problem solving in KBSs), high-level models of knowledge areas have been produced increasingly. Knowledge models in graphical format are also called knowledge maps.

We define knowledge mapping as the techniques and tools for visualizing knowledge and relationships in a clear form such that business-relevant features are clearly highlighted (see also Vail III 1999]). Knowledge maps are created by transferring certain aspects of (tacit or explicit) knowledge into a graphical form that is easily understandable by end-users, who may be business managers, experts or technical system developers.

For end-users, the highlighted aspects of knowledge maps form indicators/performance measures (of strengths, weaknesses, complexity, etc.) that can be used to improve business processes (Vail III, 1999). Knowledge maps can also be used to search for knowledge (O'Leary, 1998). Knowledge maps form the interfaces that refer you to knowledge directly or to experts who have knowledge in a tacit form.

In the remainder of this section, we discuss two knowledge frameworks which form the basis for knowledge mapping later on in this article.

2.2 The QFD Knowledge Framework

Quality Function Deployment (QFD) is a communication technique for product and process change programmes (Hauser & Clausing, 1988; Cohen 1995). QFD is mainly focused on consumers as it guides translations of the implicit and imprecise ‘voice of the consumer’ into explicit and more precise product requirements (in QFD terminology called the WHAT). QFD also guides the construction of an inventory of technology options to make the products (called the HOW).

For optimal communication across functions, a shared language needs to be available. QFD provides the framework in which multi-disciplinary teams can agree upon a common terminology by defining the precise product attributes and technology options. In Section 5, we will explain how tools can be used to support the process of defining and agreeing a common terminology. Note that this QFD perspective relates closely to the notion of ontologies (Gaines, 1997).

Moreover, QFD helps to structure the WHAT versus the HOW. Using the QFD framework, the consumer requirements can be systematically translated into technology options to make the products. However, QFD is not focused on the nature and underlying knowledge of these relations. We have added this knowledge perspective in order to make the QFD structures useful for knowledge mapping. Thus, this QFD framework forms a business-oriented template of the CommonKADS Knowledge Model.
 
The essentials of the relationships between the WHAT and the HOW are shown in the form of a ‘QFD House,’ shown in Figure 2.1. In a matrix form, it lists the consumer requirements as product attributes on one dimension and technology in the form of processes and raw materials on the other dimension. In the matrix, knowledge of the impact of every process and raw material on every product attribute can be captured systematically. For example, knowledge about the impact of free range eggs on the taste of a mayonnaise can be captured in the QFD framework. If no knowledge is available, then a knowledge gap has been identified.

In Section 5, we demonstrate how the QFD framework has been used to map knowledge in several Unilever areas. The resulting knowledge maps present both what is known and what is not known in these areas. These knowledge maps are used by business management to exploit knowledge strengths and to direct research.

The construction of an adequate knowledge framework is a complex activity which requires knowledge engineering skills. The knowledge engineer ensures that the different perspectives on the knowledge area are incorporated in the knowledge framework, and that it is compatible with other knowledge frameworks in related areas (or the area itself). Domain knowledge modelling and task analysis (included in the CommonKADS knowledge engineering methodology (Schreiber et al., 1999)) forms the basis for this activity. Two issues are particularly important when constructing the knowledge framework:

• scope: e.g. which products and product attributes should be included, which technology options should be excluded, etc.
• level of detail: should knowledge only be described in qualitative terms, or should the more detailed quantitative form be used

In the knowledge structuring phase during the development of a diagnosis KBS, relevant causal knowledge is acquired and structured. This causal knowledge includes causal relations between problems, their causes and their solutions. For an overview of this causal knowledge, a knowledge framework in the form of a Problem-Cause-Solution diagram could be used (Figure 2.2). Behind the problem, cause and solution statements as well as the causal relations, very detailed and complex knowledge might be available. Thanks to the high-level presentation of the causal knowledge in the graphical presentation of the Problem-Cause-Solution diagram, management gets a transparent overview of the knowledge which could lead to strategic decisions about developing the knowledge area further (see Section 6).

3. KNOWLEDGE ACQUISITION  [Table of Contents]

The acquisition of knowledge to feed into knowledge models and knowledge maps can be facilitated by the use of well-known techniques from the field of knowledge acquisition.

3.1 "Conventional" Knowledge Acquisition

Conventional knowledge acquisition (KA) is designed to operate in a default environment of a knowledge engineer directing a session to acquire knowledge from a domain expert. Hence most KA techniques need someone with knowledge or experience of the use of the technique in a session. It is usual for there to be a single expert to be the subject of the KA session, so conventional KA is one-to-one.

KA techniques can be divided into two rough groups (Shadbolt & Burton, 1990). Natural techniques are methods of acquiring knowledge which are not very different from general social interactions in which the expert might expect to take part. Contrived techniques are methods which present the expert with unfamiliar situations and get him/her to react in ways that may be novel.

One example of a natural technique is interviewing, where the knowledge engineer asks questions of the expert. Interviews can be structured, where a fixed set of questions is asked by rote, unstructured, where the knowledge engineer asks whatever questions seem appropriate at the time, and semi-structured, where there is a fixed set of questions to be asked but the knowledge engineer is allowed to diverge. Another example is shadowing, where the knowledge engineer watches the expert’s problem-solving and where necessary prompts the expert for a commentary.

Contrived techniques are intended to make the expert ‘see’ his/her expertise in novel ways. Laddering involves the expert creating a hierarchical structure of the objects and classes in the domain. Card sorts involve the expert putting cards representing objects in the domain into different sets of piles to show different ways of categorising the domain. Repertory grids, by prompting experts to make distinctions explicit that were previously only tacit, are intended to uncover the personal constructs with which the experts conceptualise the domain. Most well-known contrived techniques have been implemented as software tools. Note also that such contrived techniques generally have a strong connection to a particular method of representing and visualising the domain (e.g. as a laddered grid, as piles of cards, as dendrograms); thus developments in knowledge acquisition go hand in hand with developments in knowledge mapping.
 
3.2 Group Knowledge Acquisition

As an extension of conventional KA, there is the fairly natural idea of group KA, where, maybe under the supervision of a knowledge engineer, knowledge is acquired from a group of experts. The advantages of group KA are first, that time and resources can be saved by acquiring knowledge from several experts at once with a relatively small overhead, and second, that a consensus can be achieved early. One problem with conventional KA is that experts often disagree about fundamental points, and the knowledge engineer finds it impossible to decide which account of the domain to adopt. In group KA, any lack of fit between individual accounts should be noticed by the experts in the session, who should then apply their expertise to reaching a consensus. Group KA techniques, therefore, should aim to facilitate such discussion.

New problems can be created by the extra ambition. As (Cochran et al., 1990) note, relying on single experts can result in idiosyncratic representations of the domain, which in turn can cause problems with end-user agreement. Using multiple experts should get round this problem, but the difficulty then is to integrate the knowledge of the various experts and to develop a seamless KB.

One way to address this problem is to support consensus-finding. As an example, the Group Elicitation Method (GEM; (Boy, 1996)) is a brainwriting technique augmented by a decision support system for constructing a shared account of a domain. 7-10 experts comment on a ‘seed’ set of questions/statements about the domain in writing, and then comment in turn on everyone else’s comments (anonymously). These various viewpoints are reformulated into concepts, and everyone asked to fill in a matrix form to describe the relative priorities of the concepts. Once the information is in matrix form, a consensus can be derived using an algorithm to analyse the matrices. Critical analysis of the consensus achieved, according to Boy, tends to be constructive and to reinforce the consensus.

This and similar Group Support Systems (GSS) are improving co-operation and productivity by dramatic leaps (Nunamaker, 1997). Tools in GSSs are intended to focus team efforts in some particular way, e.g. by neutralising the effects of dominant team members, allowing teams to interact across space and time, or helping achieve consensus by getting team members to diverge from customary thinking patterns. (El-Shinnawy & Vinze, 1997) discuss the effects of GSS on group decision making. See also (Vinze, 1997) for a collection of papers on GSS.

3.3 Group Knowledge Acquisition in Knowledge Workshops

At Unilever, we have developed the concept of knowledge workshops. The goal of a knowledge workshop is to capture and analyse Unilever’s knowledge in a specific business-relevant area. A knowledge workshop is highly interactive, with 10 to 15 world-class specialists in the selected area. Knowledge sessions follow a structured knowledge framework using an interactive knowledge mapping tool in combination with group facilitation techniques. The main deliverables are a common terminology and an overview of "what is known" and "what is not known" in the area.

A key element in knowledge workshops is the QFD knowledge framework (Subsection 2.2). A carefully constructed knowledge framework provides guidance and structure during the workshop, and forms the basis for visualisation of all knowledge captured. The framework covers different perspectives on a particular knowledge area, which forces participants to look at their own knowledge in different and usually new ways. The use of a structured knowledge framework is the key differentiating factor between knowledge workshops and ‘standard’ workshops.

Filling the framework with knowledge is a process of group KA. The conventional group KA process has been tailored to include interactive software tools (Section 4). The interactive tool, the KA workbench Meta PACK (Subsection 4.3), can produce graphical and textual overviews of the knowledge that is captured. These overviews are constructed during the knowledge sessions (Figure 3.1), and projected on a big screen. The first knowledge sessions are focused on the definition of a shared terminology in order to form a common language. Following the dimensions of the QFD knowledge framework, this means that the product attributes (What) and the technology options (How) will be defined and agreed. The next knowledge sessions are intended to capture "what is known" and "what is not known". In these sessions, the knowledge framework will be filled.

Group knowledge acquisition in knowledge workshops using an interactive tool allows the interactive validation of intermediate results of the discussions. The validation of knowledge by the whole group can aid consensus. Furthermore, knowledge workshops not only bring together knowledge dispersed across individuals and sites; total knowledge can increase as participants build on each other’s experience and understanding. Furthermore, the electronic format of the results allows participants whose native language is not English to follow and remain involved in intensive debates; the collective format also allows instant hardcopies to be made of the results, without an extra stage of the circulation of and agreeing on minutes of the meeting.

4. SOFTWARE TOOL SUPPORT  [Table of Contents]

As mentioned above, knowledge acquisition and modelling techniques are increasingly supported by software. For many years, such support remained in academic prototypes, but now knowledge management support tools are making their way to the marketplace. Knowledge acquisition tools are not as commercially exploited, but examples do exist. In particular, Unilever has applied commercial tools that have their roots in academic research, Epistemics Ltd’s PC PACK and Meta PACK (which was developed with assistance from Unilever).  

4.1 Knowledge Acquisition Workbenches

Typically, KA techniques or tools are targetted at particular types of knowledge (often within particular task types). Experimental evaluation has shown that different techniques can uncover a range of types of knowledge (Shadbolt et al., forthcoming). This leads onto the idea of presenting KA techniques as a connected series of tools in a workbench (Anjewierden et al., 1990). The ideal workbench structure would include a common underlying knowledge representation language which could be used to amalgamate the output of all the different KA tools into a representation of a single knowledge base. In a workbench context, KA can be directed by a model of the problem-solving of the domain which can suggested types of knowledge that should be acquired, and maybe even which tool is best placed to acquire it.

In a knowledge management context, this implies that the model of the problem-solving, and the knowledge base or knowledge bases acquired, can be input into the management process as a representation of the current state of knowledge within the organisation. The KA workbench concept helps integrate the output of a range of tools for presentation in the management context.

Of course, problems such as contradictory output from different tools or different experts will always have to be dealt with; a facilitator with knowledge engineering experience may be needed to ensure that the workbench delivers a consistent or adequate model. Workbenches can aid this process by including such functions as consistency or redundancy checking.
 

4.2 PC PACK

PC PACK (O’Hara et al., 1998) is a PC-based portable package of tools for knowledge acquisition and engineering, and requirements capture. It is an example of a workbench of KA tools, in which more than a dozen tools are integrated with a common knowledge representation language. This means that, for example, if new knowledge is acquired using one tool, that knowledge will immediately be represented in the visualisations of the other tools in the workbench.

Most tried and tested knowledge acquisition techniques are represented in PC PACK, including a machine learning tool, tools for analysing and annotating protocols, tools for acquiring knowledge of the objects in the domain and tools for describing the various relations in the domain. The effect is of a suite of tools that provides comprehensive cover within a compact and easy-to-use structure.

Unilever uses PC PACK for various Knowledge Mapping activities. Recently, PC PACK has been extended by its developers in order to speed up Knowledge Mapping activities and improve the quality of the Knowledge Map. In this section, we present the use of laddering, matrix and hypertext tools of the extended PC PACK to construct Knowledge Maps.

Knowledge acquisition tools like PC PACK, and its successor Meta PACK, are providing a solution to the so-called knowledge acquisition bottleneck. The tools support the knowledge engineer by automating registration and providing efficient tools for identification and categorisation of knowledge elements, and tools to relate them to reasoning structures. Practice shows that knowledge engineers most often rely on document analysis and (un)structured interviewing only. Tools for laddering, and relating knowledge elements tend not to be used despite the fact that they can be very effective and efficient. Engineers need to know when these tools can be applied and for what purpose.

4.1.1 Laddering Tool

The PC PACK laddering tool is one of the central tools of the workbench. Laddering, as noted above, is the process of putting objects into a hierarchical form. The PC PACK laddering tool provides a simple interface to allow the expert (with a knowledge engineer's guidance, naturally) to determine concept classes, with inheritance of attributes, including multiple inheritance. Attribute values can be filled in by clicking on particular objects or classes. The tool also supports requirements laddering, to create efficient representations of decision-making processes, such as requirements, arguments for and against, positions, issues, etc.

4.1.2 Matrix Tool

The matrix tool is an alternative route to the adding of values to concepts. It uses a spreadsheet-style interface whose structure is determined by the concept structure of the domain in the KB (which may have been acquired using a different tool, e.g. the laddering tool). Inheritance and conflict detection are both supported. There are also sorting options which allow areas of sparse data to be represented transparently.

4.1.3 Hypertext Tool (Hyperpic)

The hypertext tool in PC PACK allows the user to annotate any object or concept in the KB, and to create hypertext links between objects. Annotations are generally text-based, and can be exported as text or HTML.

Between them, the laddering tool, matrix tool and hypertext tool are capable of being used to develop a picture of the objects in a domain and their hierarchical structure and attributes, to display them in either a tree (ladder) format or as matrices, to make textual annotations to them, and to represent relationships as links across a hypertext structure.

4.3 Meta PACK

Meta PACK, a knowledge management tool enabling the construction of flexible ontologies which can be populated at a later time on the fly, is the latest development on the PC PACK concept. Flexibility is achieved by allowing all of the object classes and relations used in the acquisition and data-entry tools to be user-defined. Ladders can be constructed using any number of object classes and relations. Ontology rules are used to detect 'unreasonable' laddering operations, which may be allowed as exceptions. A matrix tool allows data entry with user-defined forms, and presentation of information in multiple formats. As with PC PACK, a hypertext tool allows extensive annotation of objects and concepts.

Meta PACK was originally developed for workshops. It is very suitable for workshops based on QFD or similar techniques. The tool provides more benefit if the structure of the workshop is (at least partially) determined before the workshop starts: it is not always easy to transform ladders and matrices on the spot.

As with PC PACK, the main applications for aiding knowledge elicitation are the laddering tool and the matrix tool. Both of these tools operate on a single shared knowledge base; this means that the tools can both can be used independently, but they can also be used in combination. The shared knowledge base is stored as a standard (MS Access) database, which means that knowledge from Meta PACK can easily be exported to other applications.

In contrast to its predecessor PC PACK, multiple types of objects and relations can be depicted in the same ladder. If predefined ladders are used, using Meta PACK does not require much training. However, constructing new types of ladders is a bit more complicated. Due to its central knowledge base, Meta PACK is also suitable for knowledge acquisition in which causal trees, linking problems, causes and solutions, are acquired as input to a KBS. Since the knowledge is stored in the form of database tables, it can be exported easily.
 

4.4 Advantages and Limitations of PC PACK and Meta PACK

There were a number of reasons for selecting PC PACK and Meta PACK, but the most convincing was the pragmatic argument: there are no commercially available KA workbenches with the coverage of PC PACK (fifteen tools).
As noted in Subsection 4.1, the workbench architecture means that knowledge of several different types can be acquired using it. Knowledge can also be imported and exported easily in the standard formats PC PACK recognises. Meta PACK was designed for use in QFD contexts, which makes it ideal for Unilever's knowledge workshops which use the QFD framework. Working with an alternative tool would mean in all likelihood sacrificing the breadth and integration that a workbench would provide. Working with a single tool would mean that only a limited range of knowledge could be acquired. Working with multiple tools would mean problems integrating knowledge bases with incompatible representational formats (as well as more pragmatic difficulties such as the necessity of confronting experts with several different interfaces).
Experience of working with PC PACK in knowledge workshops has uncovered some limitations, however. For example, it is not yet possible to include the graphics of PC PACK and Meta PACK in Internet applications; this is unfortunate because the formats devised for the presentation of knowledge (e.g. a laddered hierarchy, a matrix) are straightforward and efficient. In the next version of PC PACK current under development, much more thought will be given to the management problem of publishing knowledge as well as acquiring it; in particular, it should be possible to export knowledge content as web pages.
Another limitation is that the user base of PC PACK is not particularly large. The tool itself has not been commercially available for very long, and although there is a small number of large bluechip organisations which use the technology, nothing like a PC PACK user community has developed.
A third limitation is that PC PACK is designed for a centralised user — a knowledge engineer — which makes it difficult for groupwork. Later versions of PC PACK are intended to work on a client/server architecture so that multiple/simultaneous views of the domain are possible.

Having described the technology in use in Unilever’s knowledge mapping initiative, as well as the underlying principles, we now go on to present two studies of the initiative in action.

5. CASE 1: KNOWLEDGE WORKSHOPS AT UNILEVER  [Table of Contents]

In many areas, expertise and experience is fragmented and scattered across Unilever and often located in the heads of specialists. It is clear that leveraging this knowledge is crucial to getting a business edge, but this is easier said than done. The intuitive solution would be to capture the knowledge of the individual experts and distribute it e.g. through books or electronic systems. This would, however, be extremely time-consuming. Moreover, the specialists would remain isolated, unable to question or build on the collective knowledge. When specialists are brought together, "individual practices" can be combined to an agreed "common practice" and their knowledge captured in a structured way so that it can be shared, developed and applied globally.

In the following sections, we first present the key deliverables of knowledge workshops. Then, we explain how to run knowledge workshops. Before presenting conclusions, we emphasise the relevance of sharing tacit knowledge during knowledge workshops.
 
 
 
5.1 Knowledge Workshop Deliverables

The main deliverable of a knowledge workshop is an overview of "what Unilever knows" and "what Unilever does not know" in the selected knowledge area. However the total list of deliverables is much larger (see also Figure 5.1):

Knowledge Mapping:

A fast, accurate and cost effective way of making explicit and recording knowledge held by experts in Unilever has been developed: the knowledge workshop. In previous sections, we have discussed the key elements of knowledge workshops:

• The QFD knowledge framework provides guidance and structure in defining a common language and mapping key knowledge (Subsection 2.2)
• Group knowledge acquisition processes contribute to efficient capturing and validation of knowledge of 10-15 world-class specialists (Subsection 3.3)
• The laddering and matrix tools of Meta PACK make interactive knowledge validation possible as well as dynamic creation of new knowledge (Subsection 4.3)

• Introductory presentations
• Terminology sessions to structure and agree on the terminology as described by the dimensions of the knowledge framework of the workshop, e.g. what the quality attributes of a product are, how can they be broken down, and how can they be measured
• Knowledge mapping sessions to capture "what is known" and "what is not known" in this area, using the knowledge framework as well as the outcome of the terminology sessions. For example, examining the impact of a unit operation on a quality attribute of a product
• Knowledge sessions to capture diagnostic knowledge about the causes of problems in quality attributes of a product, using the predefined knowledge framework
• Knowledge validation sessions to validate the captured knowledge both individually and groupwise
• Sessions to analyse and prioritise "what is not known" (i.e. the knowledge gaps) in the area
• One or more non-workshop but related activities, such as a factory visit to get a practical view on the knowledge area
• Concluding activities including the creation of recommendations of follow-up activities to act upon knowledge exploitation and bridging knowledge gaps in order to continue the momentum of this network of participants
• Evaluation of the workshop to improve the quality of future knowledge workshops

• Developing the knowledge framework for the knowledge workshop (mainly in advance, but also during the workshop)
• Planning knowledge acquisition sessions with respect to both the knowledge framework and time (mainly in advance, but also during the workshop)
• Preparing and distributing pre-reading material
• Stimulating discussions
• Guiding and directing discussions during knowledge acquisition sessions
• Capturing and structuring knowledge from the discussions interactively using Meta PACK

Knowledge workshops were initially designed to capture and map knowledge in an explicit format. Moreover, we have noted that a lot of tacit knowledge has been shared amongst participants as well, in particular during breaks, lunch, dinner and in the bar. Although knowledge capturing is a very important knowledge management activity, we will only able to capture a fraction of all the available tacit knowledge. Therefore, activities where experts can share tacit knowledge will be very important as well. We believe that a balance needs to be found between capturing knowledge in an explicit format on the one hand and sharing tacit knowledge at the other hand. One of the reasons for the success of knowledge workshops within Unilever is the creation of an open environment where tacit knowledge can both be captured and shared among the participants (Von Krogh, 1998).

Workshops are not always the most congenial environments for knowledge sharing, at least initially. During the workshops, then, it is useful to try to get the participants to share knowledge whenever possible. The following points are suggestions as to how this might be done.

1. The workshop should be structured and run so that there is a clear, direct and immediate benefit to all participants
1. The knowledge framework used during the workshop should help to neutralise competitive view points and allow different knowledge perspectives to be covered, rather than working towards a single "best" view
1. The facilitation team has to help all participants to contribute, and has to ensure that no professional intimidation takes place
1. The participants should be explicitly recognised as being the best Unilever experts on the workshop topic
1. The workshop should not be a one-off event, but part of a series of knowledge activities supporting the business strategy

Over 30 knowledge workshops have been held so far — ranging from week-long knowledge mapping exercises to shorter meetings, for example to debrief projects and capturing lessons learned. More than 300 different people from around the world have been involved so far. The success of knowledge workshops has meant that a bottleneck was created by the scarcity of knowledge workshop facilitators. As a result, an internal transfer programme was set up, including the production of knowledge workshop guidelines, a video, PR material and training courses. More than 40 people have followed the training course so far and half of them have been involved in facilitation work already.

We have learned that the organisation of follow up activities is crucial for sustainability and medium to long term benefits. A knowledge workshop is a superb way of beginning the process of leveraging knowledge. For example, activities should be organised to develop the knowledge map further and exploit the leading edge knowledge throughout the company. In addition, research & development should be encouraged to focus on top priority knowledge gaps.

Unilever has recognised that the speed of developing, sharing and applying knowledge is crucial to meeting its strategic goals. Knowledge workshops have brought together experts from around the world to share and capture vital knowledge. We expect that the knowledge workshop concept will be a key building block in many knowledge management activities to come in the near future.

6. CASE 2: KNOWLEDGE SWOT ANALYSIS  [Table of Contents]

In this second case study, we start with a knowledge model that was created for the development of a KBS in the area of oil processing. We will demonstrate how this knowledge model can be used as the basis for a set of knowledge maps that were highly relevant for management in the area of research and development.

In this case, the causal knowledge framework of Section 2.3 was filled twice. First, knowledge about oil processing problems, their causes and solutions was captured. Using the structured interviewing technique (Section 3.1), it took 20 sessions of about 1.5 hours to capture the key knowledge of 2 experts in this field of oil processing. This knowledge forms the core of the diagnosis KBS that was developed (Speel & Aben, 1997; Speel & Aben, 1998).

In addition, knowledge maps were created by filling the causal knowledge framework with numbers of literature references. Counts of both internal and external literature references (including reports, patents, memoranda and letters) were assigned to the problems, causes and solutions. Note the danger of assigning numbers of references alone since one document may contain more knowledge than ten others. Therefore, the validity of the indications must first be confirmed by experts in the field.

• The number of internal references provides indications about strengths and weaknesses of the research field. This means that these knowledge maps provide interesting indications on how well the internal knowledge base provides solutions for the problems and causes appearing in the field.
• Comparisons between the quantities of internal and external references give indications for opportunities and threats of the research field. This means that these knowledge maps provide interesting indications on how this internal knowledge base relates to external knowledge bases.

The "Strengths/Weaknesses Knowledge Map" can be used to visualize the internal field of knowledge. The PCS-diagram is taken as the horizontal grid of the three-dimensional map. Every causal relation cited by the internal literature is denoted on the grid. The criterion for including an internal reference as a citation of a certain causal relation is that it has to quote a complete sequence of one unique problem, one unique cause and one or more solutions. Every new citation is added to the previous ones and the number is visualized in the third dimension. In Figure 6.1, an example knowledge map contains a two-dimensional PCS-grid and a third dimension which is the number of citations on a certain causal relation (between problem and cause or cause and solution).

From the map in Figure 6.1, one can determine whether the current set of solutions is sufficient to neutralise causes and to solve problems. It fixes the areas of attention by visualizing the knowledge gaps. This knowledge map can support the discussion and formulation of a research strategy for the research field. The following observations can be made:

• Problem 2 frequently occurs. There is one cause of this problem for which a range of solutions exists, namely cause B (the grey bars). This range is probably sufficient to address this cause when it appears. There is another cause to the problem, namely cause D (the white bars), for which very few solutions exist. It is not a coincidence that this cause is the subject of a project currently running.
• Apparently, the solutions presented to solve cause D (the white bars) are generally insufficient. Therefore, the ‘If all Fails’-cause (the black bars) is often invoked and solutions 14, 15 and 16 applied. These solutions are less desirable than others, since it turns out that they lead to relatively high processing costs. It is also a goal of a current research program to create solutions which are less expensive.

After assigning all relevant internal and external literature to the problems, causes and solutions in the causal knowledge framework, it is very tempting to compare the profiles. For example, in Figure 6.2, the solutions-profiles from the internal and external literature are compared. We call the result an opportunities/threats knowledge map.

When considering the example opportunities/threats knowledge map in Figure 6.2, the following observations are worth noting:

• There seems to be a big difference in the application of solution 2. Apparently, there are many external activities in this area, while this solution is not often cited internally. If this difference is significant, it may justify the fact that there is a project running dealing with this solution.
• Also solutions 7 and 13 are cited externally much more than internally. Again, it needs to be noted that solutions to certain problems have been found externally.
• There are three solutions which are much more cited internally than externally (4, 8 and 15). This may indicate a strength. It might be worthwhile to protect this strength by applying for (even more) patents.

First, the reservation needs to be made that the knowledge maps of the SWOT analysis are not sufficient to define a research strategy. The indications must first be confirmed by experts in the field. We consider expert confirmations as an obligatory part of the knowledge SWOT analysis.

The visualised SWOT analyses can be related to the research strategy in this area, and management can take various knowledge management decisions informed by them. In addition, decisions how to disseminate, exploit and protect available knowledge can be taken, having isolated gaps and areas of strength through these analyses.

7. CONCLUSIONS  [Table of Contents]

In this article we have demonstrated that knowledge mapping techniques can be applied successfully in an industrial environment in order to manage key knowledge appropriately. Knowledge mapping forms a key technique for knowledge management initiatives, since key competitive decisions can be taken based on the resulting transparent overviews of knowledge maps.

When creating knowledge maps, we have learned the following:

• Knowledge modelling techniques form the basis for structuring and visualising knowledge in order to create knowledge maps. In particular, visualisation of relations between knowledge items seems to bring new insights to experts and managers. Knowledge modelling techniques also help to decide what is relevant and can be focused on business-relevant knowledge only.
• Knowledge Acquisition techniques form the basis for identifying and capturing business-relevant knowledge. The application of group knowledge acquisition techniques helps achieve consensus on the knowledge maps.

For example, Tufte has performed a series of studies on information design, looking at three separate problems. (Tufte, 1983) examines the process of representing statistical data pictorially; for instance how representing statistics about the rates of diseases in various parts of the United States as a coloured map can (a) make a large data set coherent, (b) encourage the eye to make broad comparisons, and (c) highlight trends that would be invisible on virtually all numerical representations. (Tufte, 1990) performs a similar study on qualitative data, while (Tufte, 1998) looks at pictorial ways of representing process or narrative.

These studies are largely focused on case-by-case analyses, but there is a wealth of hints about design, colouring, layering, interaction effects etc, that can affect the ability of a visual representation to convey information (or prevent the representation appearing to convey something untrue). There are many important lessons to be learnt from work such as this for the field of knowledge mapping and knowledge acquisition, for many experts routinely use diagrams to pass on or acquire knowledge (or merely to crystallise a particular piece of reasoning) as part of the everyday exercise of their expertise (Cheng, 1996).

On a more pragmatic level, we learned that during breaks and bar sessions in knowledge workshops, experts share large amounts of tacit knowledge. Since it is impossible to acquire all business-relevant knowledge in an explicit form, we recognise the importance of sharing tacit knowledge. We believe that a balance needs to be found between capturing knowledge in an explicit format on the one hand and sharing tacit knowledge at the other hand. In knowledge workshops, knowledge mapping techniques to capture knowledge also help to share tacit knowledge. Another technique we are investigating at the moment is establishing communities of practice. Important for these techniques are cultural issues: how best to establish a dynamic group of devoted people who are willing to share knowledge openly? Experiences from knowledge workshops provide a first step in that direction.

Knowledge engineering provides very useful techniques and tools for knowledge management. However, knowledge engineering is only a part of knowledge management. We see knowledge management as a combination of three pillars: knowledge processes, technology enablers and organizational and cultural alignment. Although knowledge engineering provides a solid basis for knowledge processes and technology enablers, not enough attention has been paid to organizational and cultural alignment. Our current aim is to improve these organizational and cultural issues in order to reach a balanced view on knowledge management.

ACKNOWLEDGMENTS

We would like to thank the following people for their contributions: Manfred Aben for his support in KW and Meta PACK development; Iain Ritchie and Manfred Aben for being champion in organizing KWs; Chris van der Touw for his very useful feedback on various topics; Aafke Keizer for improving KW concept from lessons learned and organizing transfer of technology; and Steve Swallow for developing Meta PACK

REFERENCES  [Table of Contents]

Anjewierden, A., Wielemaker, J. & Toussaint, C. (1990). SHELLEY: Computer Aided Knowledge Engineering, in B. Wielinga, J. Boose, B. Gaines, G. Schreiber & M. van Someren (Eds.), Current Trends in Knowledge Acquisition, Amsterdam: IOS Press

Boy, G. (1996). The Group Elicitation Method: an introduction, in N.R. Shadbolt, K. O’Hara & A.Th. Schreiber (Eds.), Advances in Knowledge Acquisition, 290-305, Berlin: Springer

Brachman, R.J. & Schmolze, J.G. (1985). An overview of the KL-ONE knowledge representation system, Cognitive Science, Vol. 9, 171-216

Cheng, P.C.-H. (1996). Diagrammatic knowledge acquisition: elicitation, analysis and issues, in N.R. Shadbolt, K. O’Hara & A.Th. Schreiber (Eds.), Advances in Knowledge Acquisition, 179-194, Berlin: Springer

Clancey, W.J. (1985). Heuristic classification, Artificial Intelligence, Vol. 27, 289-350

Cochran, E.L., Bloom, C.P. & Bullemer, P.T. (1990). Increasing the end-user acceptance of expert systems by using multiple experts: case studies in knowledge acquisition, in K.L. McGraw & C.R. Westphal (Eds.), Readings in Knowledge Acquisition, 73-89, New York: Ellis Horwood

Cohen, L. (1995). Quality Function Deployment: How to Make QFD Work For You, Reading, Mass: Addison-Wesley

Davenport, Th.H. & Prusak, L. (1997). Working Knowledge: How Organizations Manage What They Know, Harvard Business School Press

El-Shinnawy, M. & Vinze, A.S. (1997). Technology, culture and persuasiveness: a study of choice-shifts in group settings, International Journal of Human-Computer Studies, Vol. 47, 473-496

Gaines, B.R. (1997). Special Issue on Using Explicit Ontologies in KBS Development, International Journal of Human-Computer Studies, Vol. 46, No. 2/3, Jun 1997

Hauser, J.R. & Clausing, D. (1988). House of Quality, Harvard Business Review

Krogh, G. von (1998). Care in knowledge creating, California Management Review, Vol. 40, No. 3, 133-153, spring 1998

Nonaka, I. & Takeuchi, H. (1995). The Knowledge-Creating Company, New York: Oxford University Press

Nunamaker Jr., J.F. (1997). Future research in group support systems: needs, some questions and possible directions, International Journal of Human-Computer Studies, Vol. 47, 357-385

O’Hara, K., Shadbolt, N.R. & van Heijst, G. (1998). Generalized directive models: integrating model development and knowledge acquisition, International Journal of Human-Computer Studies, Vol. 49, 497-522

O'Leary, D.E. (1998). Enterprise Knowledge Management, IEEE Computer, Vol. 31, No. 3, 54-61, March 1998

Schreiber, A.Th., Akkermans, J.M., Anjewierden, A.A., Hoog, R. de, Shadbolt, N.R., Van de Velde, W. & Wielinga, B.J. (1999). Knowledge Engineering and Management: The CommonKADS Methodology, to be published by MIT Press. See also the CommonKADS web site

Shadbolt, N.R. & Burton, A.M. (1990). Knowledge elicitation, in J.R. Wilson & E.N. Corlett (Eds.), Evaluation of Human Work: A Practical Ergonomics Methodology, 321-345, London: Taylor and Francis

Shadbolt, N.R. & Milton, N. (1999). From knowledge engineering to knowledge management, to appear in special BAM'98 (British Academy of Management) conference issue of the British Journal of Management, spring 1999

Shadbolt, N.R., O’Hara, K. & Crow, L. (forthcoming). The Experimental Evaluation of Knowledge Acquisition Techniques and Methods: History, Problems and New Directions, International Journal of Human-Computer Studies

Sowa, J.F. (1984). Conceptual structures: information processing in mind and machine, The systems programming series, Reading, MA: Addison-Wesley

Speel, P.-H. & Aben, M. (1997). Applying a library of problem-solving methods on a real-life task, International Journal Human-Computer Studies, Vol. 46, 627-652 (also in Proceedings of the Tenth Knowledge Acquisition for Knowledge-Based Systems Workshop, Banff, Canada, November 9-14, 1996)

Speel, P.-H. & Aben, M. (1998). Preserving conceptual structures in  design and implementation of industrial KBSs, Special issue on Problem-Solving Methods for the International Journal of Human-Computer Studies, Vol. 49, 547-575

Tufte, E.R. (1983). The Visual Display of Quantitative Information, Cheshire, Conn.: Graphics Press

Tufte, E.R. (1990). Envisioning Information, Cheshire, Conn.: Graphics Press

Tufte, E.R. (1998). Visual Explanations, 3rd printing with revisions, Cheshire, Conn.: Graphics Press

Vail III, E.F. (1999). Mapping Organizational Knowledge, Knowledge Management Review, Issue 8, May/June 1999, 10-15

Vinze, A.S. (1997). Special Issue Group Support Systems, International Journal of Human-Computer Studies, Vol. 47, No. 3, 355-511

Wielinga, B.J., Sandberg, J. & Schreiber, A.Th. (1997). Methods and Techniques for Knowledge Management: What Has Knowledge Engineering to Offer?, Expert Systems With Applications, Pergamon, Elsevier Science Ltd, Vol. 13, No. 1, 73-84

Wigg, K.M., Hoog, R. de & Spek, R. van der (1997). Supporting Knowledge Management: A Selection of Methods and Techniques, Expert Systems With Applications, Pergamon, Elsevier Science Ltd, 1997, Vol. 13, No. 1, 15-27

Woods, W.A. (1975). What's in a link: foundations for semantic networks, in Representation and understanding: studies in cognitive science, D.G. Bobrow & A.M. Collins (Eds.), 35-82, New York, NY: Academic Press. See also in Readings in Knowledge Representation, R.J. Brachman and H.J. Levesque (Eds.), 217-241, Los Altos, CA: Morgan Kaufmann Publishers, Inc.