IPGems Logo (3242 bytes) Articles and other writings
s_line.gif (886 bytes)
IPGems home page
s_line.gif (137 bytes)
Resources on Usability and Performance-Centered Design
s_line.gif (137 bytes)
Resources on Information and Knowledge Management
s_line.gif (137 bytes)
Contact IPGems
s_line.gif (137 bytes)

Knowledge Management in Pursuit of Performance: the Challenge of Context

by Duane Degler and Lisa Battle
Performance Improvement, ISPI, 39(6), July 2000

Read our additional thoughts on the subject, March 2002


The rabbit-hole went straight on like a tunnel for some way, and then dipped suddenly down, so suddenly that Alice had not a moment to think about stopping herself before she found herself falling down what seemed to be a very deep well.

Lewis Carroll, Alice in Wonderland


Knowledge Management (KM) has an increasingly visible profile within organizations, in part because of the growing use of intranets for communication and information retrieval. It has long been recognized that sharing knowledge and experience among people has the potential to improve the quality of work and the quality of decisions that are made – and the discipline of knowledge management has arisen to realize this value over increasing spans of time and distance.

However, the popularity of KM has led to some confusion in the industry about the difference between knowledge and information (Huang, Lee & Wang 1999; Glazer 1999). Much of the current knowledge management focus has been on the acquisition and storage of “knowledge resources,” although one of the principal values of KM has always been communication between individuals as a way to improve knowledge transfer (Nonaka & Takeuchi 1995; Davenport & Prusak 1998). Many knowledge management projects have created stand-alone information bases. While such repositories are a good start, they do not go far enough. Finding information is not the same thing as applying knowledge. It is not an end in itself.

This article explores opportunities for taking the next step and integrating KM seamlessly into business applications. What would applications be like if they were tightly coupled with KM? Information bases would exist, but the user would not need to interact with them directly. In an ideal world, business applications (such as inventory management, purchasing, and correspondence routing systems) would be designed to support the goals of the work, and users would be able to focus on doing their own work using the application as a tool. Whenever information could improve the outcome of the work, the application would identify the need, ask the information base for information, receive the information, package it, and present it to the users in whatever format was most appropriate. The users would get what they needed, whether they knew they needed it or not, and continue to work without interruption. The whole process of gathering information would normally be invisible to the users.

Of course, we know that this does not happen today. Most of the time, gathering information is painfully visible to the user (Frappaolo 1998; Henniger & Belkin 1994; Blair & Maron 1985). Often users have to leave their work and launch another application to seek information. They may have to type a syntactically correct query into a search engine that retrieves a long list of documents that may or may not be relevant. They then start a different search within each identified document to find data – or an idea – that helps them. It takes a lot of effort to describe the need and to evaluate the results.

As long as KM settles for stand-alone information bases that are not closely linked with business applications, it will not live up to its potential. As Michael Schrage recently said, "Knowledge is not the power. Power is power. The ability to act on knowledge is power" (Gurteen 1999). There is a need to focus on achieving successful performance outcomes. This is already the focus of Performance-Centered Design (PCD), an approach to designing business applications which has gained popularity in recent years. PCD applications typically aim to help inexperienced or infrequent users complete tasks successfully -- as if they were experts -- and to increase the overall quality of work products. The interfaces are easier to learn and use because they clearly match the goals and work processes of their users and provide timely information. However, many current designs rely on an information base that is relatively simple and static and tasks that are clearly specified. This does not work as well when designing for complex, changing environments or for knowledge workers whose responsibilities are less structured. The time has come to combine and extend the current PCD and KM approaches, with an emphasis on applying knowledge to support successful business outcomes.

Applying knowledge to support business outcomes

“‘Knowledge’ is generated when information is combined with context and experience.”

(Huang, Lee and Wang, 1999, p146)

In business terms, knowledge is reflected in the quality of an outcome – whether the task is operating machinery or deciding strategic direction. Knowledge benefits the organization when it is applied. Therefore, in order to reap the benefits, we need to find ways of bringing context, experience, and information into business applications to support successful performance outcomes.

Figure 1. Contributors to successfully applying knowledge.

Historically, PCD has emphasized the context in which a person performs and, through such mechanisms as role-based views and task-oriented interfaces, has attempted to provide "just enough, just in time" information appropriate to the context. KM practice has grown out of two main areas: information management (the acquisition, storage and dissemination of structured and unstructured information) and communication (represented as experience because the main goal of communication between people in KM terms is the sharing of experience to increase knowledge). We will briefly explore context, experience, and information in turn before going on to look at techniques for incorporating them into business applications.


Context is the collection of relevant conditions and surrounding influences that make a situation unique and comprehensible. The human cognitive and perceptual systems are designed to identify and use context automatically as we go about our daily lives (Anderson 1995; Hasher & Zacks 1984). It is harder than originally imagined to teach a machine to do the same. The field of Artificial Intelligence has been exploring the implications of context in areas such as natural language processing for decades and has had to readjust its initial high hopes for machines that can mimic human intelligence (Ashcraft 1994). Context is enormously complex. It involves numerous interacting factors that people do not necessarily even pay attention to on a conscious level, and many of which are outside the ability of machine input devices to capture.

Figure 2. Building a relevant view of business context.

There are two approaches to identifying context in business applications. Combining these approaches so that some attributes of context are drawn from the individual (description) and some from the computer (recognition) is often the best solution.

Context description means the person provides details that are outside the computer environment, particularly any implicit/subjective conditions. The way that someone describes context depends on how much they know about the subject, what they perceive to be the problem, and what kind of facilities they are given to describe it. Business applications should become more proactive in eliciting context from users in situations where the users need to ask for information.

Context recognition is an automated approach to context. Business applications should be familiar with their users, business processes, and subject domains. When the user asks for information, the application should be able to identify and bring forward the knowledge resource that best corresponds to the current context. When the user does not ask, the application should provide information anyway if necessary to achieve a successful outcome. A good understanding of context (on the application's part) is especially crucial in this situation, because "push" information and feedback can be extremely frustrating to the user if it is not relevant – the result being that the user ignores the information or turns the feature off entirely.


Identifying and describing context are not enough, of course. The user also needs to have a mental model of how to approach the task at hand. This typically comes from experience. There are three broad areas of experience that support successful performance.

Figure 3. Drawing on a wide range of experience.

Individual experience: An individual draws on personal experience both within the subject domain and outside of it. The depth of experience will affect the individual's approach to problem-solving and decision-making. Research in cognitive psychology has shown significant differences between novices and experts in the way they think (Tanaka & Taylor, 1991). For example, an expert has an enhanced ability to categorize circumstances and an extensive set of schemas for problem-solving, which helps mentally structure the need for additional information, but may also restrict the degree to which new information is sought (Feltovich et al., 1997).

Others’ experience: An individual can also call on other people's experience. Connecting with other people has a number of advantages. It allows for knowledge to be passed from person to person, which benefits the organization. It also increases the possibility that information and experiences are translated into tangible form – such as written in a document by one person for another – that can become part of the overall information available in the organization. Also, it challenges both individuals to unmask assumptions and misconceptions that can become a barrier.

Collective memory and patterns: If an application is able to capture best practices and patterns of expert behavior based on successful performance, then that information can be used as a model for future performance. The application can compare a current situation with a pattern of best practice to determine whether the user is "on track." If not, the application can provide suggestions based on the collective experience of the group of users.


The quality of information depends on its accuracy and its relevance to the context. Some valuable information management approaches, outside the scope of this article, are being explored in both KM and PCD disciplines (Yacci 1999). The work that is being done needs to be shared and enhanced to meet the growing needs of the merged disciplines. Areas of investigation include:

  • granularity of information – both in terms of storage and presentation
  • the importance of metadata, particularly as it relates to context
  • increasing the reusability of information, including the creation of ‘knowledge objects’
  • structuring information in ways that move away from the ‘document’ and 'page' metaphors

The ideal PCD/KM application should provide information that is both accurate and relevant without requiring the user to actively seek this information and determine its relevance. This will prevent information seeking from becoming a complex problem-solving activity of its own.

Techniques for merging PCD and KM

By bringing context, experience, and information together, we can begin to create more powerful, flexible, performance-centered business applications that support successful outcomes. These applications will assist the user by integrating dynamic knowledge repositories into the work environment, facilitating access to peers and experts as needed, and allowing both the human and the computer to do what they do best to achieve a business outcome. We have been exploring techniques for doing this, including:

  • Multi-layer context profiles
  • Proactive context-based information retrieval
  • Stealth knowledge management
  • Adaptive knowledge interfaces.

1. Multi-layer context profiles

To achieve any real benefits from automated context recognition, designers need to look for better ways of identifying context. We propose that context has multiple layers that overlap and intersect when describing a situation. These layers may include:

  • The user's current activities in the application
  • The user's role, background, working patterns, experience with the subject domain, previous use of information, and experience with relevant problem-solving or information-seeking strategies
  • Business rules, policies and processes, including aspects that are time-dependent (e.g. “this action must be completed every Thursday before 3pm”)
  • Patterns of use in the user's peer community, including what other people have done before in this context
  • System conditions, such as what other applications are running, the historic record of actions that led up to the current situation, what information resources are available and how recently the resources have been updated
  • Environmental conditions, including who else is working in the same subject domain, whether they are logged on, whether the current time is within their normal working hours, and whether they have other appointments (to determine whether people are able to be contacted to share their knowledge)

Techniques already exist for capturing much of this contextual information, but each piece of it comes from a different source, and they are not combined to achieve this kind of precision. It is increasingly important to do so. All the above information allows relevant knowledge to be provided to the individual and to be elicited from the individual for use by the organization. It also allows a dynamic PCD application and interface to effectively meet the needs of the user and the task being performed.

For example, multi-layer context profiles are being explored in the design of an inventory system for a large art sales event. Sales staff with a wide range of backgrounds and experience levels must be able to clearly inform potential buyers about both the work and the artist. Sales staff only approach the computer when a question arises – they do not regularly use the application. Ideally the inventory system should provide concise and accurate information while reflecting the experience and selling style of the salesperson. It could recognize and respond to:

  • User's domain knowledge (e.g. a ceramics expert will need more support to answer questions about 19th century oils)
  • User's past usage patterns (e.g. when talking with a potential buyer, one salesperson typically retrieves details on the associated school/style, while another prefers to see comparisons to similar artworks that he has learned about previously)
  • Time-dependent factors (e.g. if the event is currently in progress, it may check whether the item has been sold yet and recommend similar works to the buyer)
  • Geographical location of the event (e.g. at point of purchase, listings of local/regional information sources can be accessed based on the item purchased)

2. Proactive Context-Based Information Retrieval

Sometimes a knowledge worker needs access to information from multiple sources as part of a decision-making process. Instead of requiring the user to search each source individually, the business application can provide proactive context-linked information retrieval. This approach is proactive because it is initiated by the application, not the user. It is linked to the context because the facts of the current case trigger the retrieval of relevant information (including data, policy, and reference materials). The application performs the search in the background and presents a summary of its findings or relevant links embedded in the interface at appropriate points in the work context.

For example, for an insurance company's claims management system, we designed an interface to reflect each phase of analysis when handling insurance claims, in order to provide information relevant to that context. During each phase, the application would automatically retrieve information from multiple reference sources, as well as from past claims, and dynamically create links in appropriate places within the electronic claim file. When the claims specialist was evaluating medical information, the application would proactively identify relevant medical tests and reference topics related to the claimant's condition and the point in the decision-making process. The interface would also identify any applicable state laws and retrieve lists of previous cases with similar attributes to assist the claims specialist in following best practice for handling the claim.

3. Stealth Knowledge Management

We are beginning to develop an approach we call “Stealth Knowledge Management” – the ability to gain knowledge of best practices by capturing the way people perform tasks. The more computers are used to carry out job tasks and make decisions, the more there is a growing record of the way decisions are made – embodied in the way the applications are used. Unlike some expert system and knowledge management approaches, the rules of decision-making are not artificially defined in advance, but based on actual performance. They are as dynamic as the experienced people who perform the tasks day-in and day-out. The capturing of this experience (“knowledge from actions”) can be valuable to less experienced people performing the same tasks.

The key factors of Stealth Knowledge Management are:

  • the data is a representation of the business knowledge of experienced users
  • the knowledge is dynamic: it changes as the business context and application use change
  • decision-making patterns define rules that can be modified or weighted based on the qualitative outcomes of the tasks which generated the patterns
  • the knowledge can be made available immediately when a decision is to be made, presented as recommendations for action

For example, in one organization's correspondence routing system, all requests for information from the public were channeled through a central office. Experienced users in that office categorized the requests and assigned them to other departments to respond. The categorization and assignment of requests was a very subjective process, and when a less experienced person had to take over (for example, when primary staff were ill), they could not perform the task. One design approach was to compare the request description with the responses eventually received from the departments. The goal over time was to create patterns of ‘probable suppliers’ of requested information. When a user entered a new request, the application could recommend departments, along with a ranking to indicate higher or lower probabilities that they could successfully respond. There were two keys to this concept:

  • The possible routing paths needed to be presented as options (not as required rules for routing), because the user needed the flexibility to evaluate subjective factors and make the final routing decision
  • The application’s ranking system would constantly be fed with every routing decision made, giving it the ability to continue ‘learning’ over time.

The Stealth Knowledge Management approach is particularly exciting due to the invisibility of knowledge creation. This sort of application is a strong candidate for tools that work in the background, monitoring such things as message traffic between applications and the operating system, because they can be applied to existing applications without recoding. Also, the increasing availability of simple, plain-language neural network and fuzzy logic tools to support pattern identification and monitoring begin to make this a viable approach when developing new applications.

4. Adaptive Knowledge Interfaces

Providing information relevant to the current context is not enough to ensure successful performance. A person's ability to apply information depends in part on his/her level of existing knowledge. Instead of building one-size-fits-all interfaces, we may need to present information differently to different users based on their individual preferences and expertise. (This is similar in concept to "adaptive interfaces" in which a first-time user sees a limited set of features. Additional features become available later and support features disappear as they are no longer required.)

A system designed to do this would have both a knowledge layer and a context-aware presentation layer, which would evaluate the user's expertise and the work context when determining what information to provide. The interface could adapt its:

  • Content detail: To avoid overwhelming novices with detail, content could be regulated by identifying topics as appropriate for beginners or experts and displaying them accordingly.
  • Feedback style: An interface could provide more detailed feedback and do so more frequently if the user was unfamiliar with the subject domain, while an interface for experts could display limited feedback. The application could evaluate the business importance of the feedback, derived in part from best practices, in determining whether or not to interrupt the user's work.
  • Instruction: More instruction and more explicit labels could be shown if needed to clarify unfamiliar tasks. The instructional components of the interface could also be enhanced for contexts that are rare or in situations where the user appears to be deviating from best practices.
  • Structure and sequence: The application could provide increasing levels of decision support and structure for the work. It could help users with limited domain knowledge to categorize the problem, structure the problem space, and select the appropriate course of action, based in part on patterns and best practices drawn from previous user experience. Providing additional structure can help increase knowledge in the subject domain while at the same time leading to a successful performance outcome.

For example, in considering designs for an application to assist people with developing and cataloguing classroom training, we realized that some of the users were experienced instructors while others were novice instructors who had been recruited to teach because of their expertise in a subject area. The experienced instructors needed the ability to create and organize their own training materials, retrieving appropriate templates and checking the archives for previous materials that could be used as starting points. The novice instructors needed more structure to guide them through the process of preparing for a course, but fewer background materials on the subject itself. The interface for novices had to provide instructions on preparing course objectives, the amount of material to cover in the course, the appropriate types of exercises, what to include in handouts, and so on. This more structured approach would help the novice instructors to complete each step successfully.

Recognizing the limits of automated support

Although we have been talking about some of the advantages of automated context recognition, we also think it is important to acknowledge the limits. One of the things that KM brings to the merger is an understanding of corporate knowledge as a dynamic thing. A KM system quickly becomes obsolete if it does not continue to incorporate new information quickly and does not have a mechanism for removing out-of-date or contradictory information. This means that making assumptions based on patterns of use will be tricky. Patterns of use will change as circumstances change and knowledge evolves.

The implication for design is that a rigid system will frustrate the knowledge worker who is trying to adapt to new circumstances. For example, if the person who formerly advised on a subject has moved to another position, a hard and fast rule for communication as part of a knowledge management approach will create undue burdens. No matter how good the underlying patterns and knowledge base, there will always be a wrinkle that is not predicted.


Context is the key to more invisibly and dynamically supporting business performance – it is the area of greatest potential in the merger of KM and PCD. If the two approaches can jointly rise to the challenge of identifying context and responding appropriately, a great burden will be lifted from the ordinary user.

What we have come away from our experiences with are three simple rules:

  1. Pay attention to the many facets of context if you want to provide information to the knowledge worker.
  2. Design applications that learn from everyone’s experience over time – and keep learning.
  3. Provide support as suggestions – not answers – in complex, knowledge-rich environments so that the knowledge worker remains in control.

Return to the top of the page


Anderson, J.R. (1995). Cognitive Psychology and its Implications, 4th edition. New York: W.H. Freeman.

Ashcraft, M.H. (1994) Human memory and cognition. Second edition. New York: HarperCollins.

Blair, D.C., & Maron, M.E. (1985). "An evaluation of retrieval effectiveness for a full-text document-retrieval system." Communications of the ACM, 28(3), 289-299.

Davenport, T.H. & Prusak, L. (1998). Working knowledge: how organizations manage what they know. Boston, MA: Harvard Business School Press.

Feltovich, P.J., Spiro, R.J., & Coulson, R.L. (1997). "Issues of expert flexibility in contexts characterized by complexity and change." In Feltovich, P.J., Ford, K.M., & Hoffman, R.R. (Eds.), Expertise in Context, Menlo Park, CA: AAAI Press.

Frappaolo, C. (1998). "Search and retrieval lay the foundation to knowledge discovery." KMWorld, May 11, 1998.

Glazer, R. (1998). "Measuring the knower: Towards a theory of knowledge equity." Knowledge Management 2(2), 12-18. Reprinted from the California Management Review 40(3).

Gurteen, D. (1999). "The discipline of dialogue." Knowledge Management 3(4), 30.

Hasher, L., & Zacks, R.T. (1984). "Automatic processing of fundamental information: the case of frequency of occurrence." American Psychologist 39(12), 1372-1388.

Henniger, S., & Belkin, N. (1994). "Interface issues and interaction strategies for information retrieval systems." Proceedings of CHI '94 (pp. 387- 338).

Huang, K-T., Yang, W.L., & Wang, R.Y. (1999). Quality Information and Knowledge. New Jersey: Prentice Hall.

Koulopoulos, T. (1997). "The pieces of the knowledge management puzzle." KMWorld.

Lawton, W. (1999). Electronic Performance Support Systems and Knowledge Management - The Merging Ground. Dissertation, MSc Information Systems, University of Brighton.

Nonaka, I. & Takeuchi, H. (1995). The knowledge-creating company: how Japanese companies create the dynamics of innovation. New York: Oxford University Press.

Tanaka, J. W., & Taylor, M. (1991). "Object categories and expertise: is the basic level in the eye of the beholder?" Cognitive Psychology, 23, 457-482.

Yacci, M. (1999). "The Knowledge Warehouse: Reusing Knowledge Components." Performance Improvement Quarterly, 12(3), 132-140.

Return to the top of the page


Copyright, Reuse and Citation

The content of this article may be referenced with the appropriate citation information included (see below). The entirety of the article must not be reproduced without written communication with ISPI (www.ispi.org).   Also, we would appreciate your notifying us if you intend to use these concepts or images, as we are curious to know where they prove valuable.

To cite the material, please include the following information. We recommend the format: 

Degler, Duane and Battle, Lisa (2000). Knowledge Management in Pursuit of Performance: the Challenge of Context. Performance Improvement (EPSS Special Edition). ISPI, 39(6), July 2000. Online: www.ipgems.com/writing/kmcontext.htm.


| Welcome to IPGems | PCD & HCI | IM & KM | Contact IPGems |

© Duane Degler 2000