InformationWeek

March 16, 1998

Knowledge Management:
Great Concept...But What Is It?

InformationWeek Labs and Doculabs examine five products that try to help companies turn an abstraction into a reality

By Jeff Angus of InformationWeek Labs, and Jeetu Patel and Jennifer Harty of Doculabs

endors are doing their best to hawk a vast spectrum of products labeled "knowledge management." Combine this with an absence of common naming concepts and features, and it becomes difficult even for interested technology executives to get their arms around the subject.

But there's a monster of an underlying reason why knowledge management is tough to grasp. Like the Holy Grail, it's an idea worth battling for... that doesn't really exist.

Desktop applications are easy to understand and use because they're only trying to replace an existing machine with a computer. Word processing, for example, was a typewriter replacement; spreadsheets were a calculator replacement. More sophisticated applications are harder to understand and use because they replace a constellation of manual business systems and machines. Electronic document-management systems, for example, have designs based on known, understood business processes that occur without computers.

Knowledge management is still more difficult to understand because the business process it aims to computerize doesn't exist in the real world for most organizations. Unlike replacing a machine such as a typewriter or an office procedure such as document control, businesses have been trying unsuccessfully throu ghout the 20th century to make knowledge management a reality. With the specialized exception of units whose entire business is knowledge management, there's no obvious, proven model to follow. The results are product ivory towers built on conceptual quicksand.

We can, however, describe what knowledge management is by stepping back from technologies and products and taking a high-level view of the business issues before jumping into technology-based solutions.

The business problem that knowledge management is designed to solve is that knowledge acquired through experience doesn't get reused because it isn't shared in a formal way. Whether it's how to avoid remaking mistakes, to assure the reuse of proven best practices, or simply to capture what employees have learned about suppliers, customers, or competitors, knowledge management is the concept under which information is turned into actionable knowledge and made available effortlessly in a usable form to the people who can apply it.

Knowledge management is a way of doing business. In reality, it's more a business practice than a product. The products are what facilitate the practice of knowledge management--or at least specific facets of it--with the appropriate use of technology.

While some organizations may have proprietary systems that work for a single vertical market (most notably the consulting arms of Big Six accounting firms), no one has successfully created a reproducible system that others can follow with a reasonable chance of success. And there are vertical knowledge-management applets in departmental areas such as the help desk. But the ultimate goal isn't creating a departmental island of success recycling. It's giving the organization the capacity to be more effective every passing day with the gathering of institutional memory the way human beings have the capacity to become more effective and mature every day with the accumulation of thoughts and memories.

The system objectives that support the knowl edge-management goal are knowledge gathering, organizing, refining, and distributing. Each of those objectives has a host of enabling functions. Knowledge organizing, for example, happens through searching, filtering, cataloging, and linking, to name a few. Technologies (products or features of products) combined with business practices make these objectives achievable.

InformationWeek tested five products--Wincite 5.0, Intraspect 1.5, ChannelManager 2.0 (beta), BackWeb 4.0, and KnowledgeX 1.0 in conjunction with Doculabs Inc., a Chicago research and product-assessment firm specializing in electronic document-management and knowledge-management systems, to judge how well each product fulfilled its claims, and which parts of the knowledge-management map it has features to support.

Barriers To Success
Even if you can cover all the required functions with technologies, you're going to face intractable challenges you most likely have no authority to overcome.

Because of the downsizing cult's belief in the disposability of employees, many knowledge workers have lost a sense of loyalty to the organizations they work for. But the knowledge-management concept needs cooperation to work--employees have to trust their donations don't undermine their job security or, more commonly, their job competitiveness.

For example, in shops where evaluation, promotion, or compensation is based on relative numbers, individuals sharing their knowledge reduce their chances of success, and those holding back have a relative advantage, which discourages cooperative behavior. Organizations about to go through another downsizing are never likely to measure group contributions more highly than individual ones--the individual stars get to keep their jobs, the cooperators are lucky if they get a Miss Congeniality award with their pink slips. And the laid-off knowledge sharer is worth relatively less because his or her knowledge is now shared across a wider population. It's therefore important that the system be designed around financial incentives that support the knowledge-management way of doing business.

And, of course, there's the more routine challenge of committing to making the knowledge-management concept a reality. More often than not, people are hammering away as hard as they can to get their current workload processed. The rewards of knowledge management are all mid- to long-term.

The number of CIOs that have a lever, a fulcrum, and a platform on which to stand to move this set of problems aside is few. The only thing that makes it worth the effort is the reward: the chance to institutionalize important lessons and create the self-perpetuating cycle of increased competency.

Knowledge management is more a marketing slogan than a basis for buying a product. You should define your corporate goals in trying to implement the knowledge-management concept and lay out a set of supporting objectives. You should examine your organization's incentive systems for kinds of beh aviors and internal data storage and distribution models. You should also look at what functions a product supports and how easily one could integrate it with other products that support complementary functions.

If you're not going to treat the knowledge-management initiative as a strategic and high-priority effort, you shouldn't waste time trying to deploy it. And even if the deployment is purely departmental, some strategic changes might need to be made by the management of the department to address issues such as financial compensation based on contribution of knowledge to the corporate information store. This is because a knowledge-management implementation requires a shift in philosophy for most organizations--not only in how people work, but more importantly in how they behave and interact with each other. (For implementation ideas, go to techweb.cmp.com/iw/673/implement.htm .)

In the end, the key to whether you should embark on this journey is understanding whether your organization has the ability to make wholesale changes to become a business built around the knowledge-management concept.

Wincite 5.0
Wincite Systems' software is the most mature knowledge-management product in the group. It's been used for competitive intelligence and some other applications for about 10 years. This maturity is both an asset and a liability.

Wincite is designed to manage a shared repository of structured data and deliver it in forms that ease the analysis work. The program's design presumes most of the users are people who will analyze and contribute value to what's stored.

Wincite uses a group model, where putting content into the knowledge base is the work of one or a handful of "librarians," but users can suggest additions. Wincite users can add notes with proposed updates or new information. The librarians responsible for maintaining the knowledge can easily see these notes and incorporate all or part of t hem. While this model will be sensible for many organizations, the system's two-tier architecture makes it difficult to break out of the model and create a universal full-client deployment approach.

Librarians and users with rights can add information to the knowledge base by using its form. Each form contains defined fields, and one can add information manually, add as an attachment, or dynamically link to documents outside Wincite and on network drives, in dedicated databases, or on the Internet.

Wincite uses a multidimensional database model squeezed into a relational database. The stored information is organized into what the interface calls "notebooks," "topics," and "subjects." The naming convention can fool you. Subjects aren't hierarchically underneath topics--they're really "subject companies" or "subject people." So, for example, a "notebook" could have the title Knowledge Management and topics might include "customers" and "products," while subjects might be companies that make pr oducts and have customers. This structure replaces a fully hierarchical approach in which you could create multiple levels of subfolders to break information down into finer subcategories.

End users get information either by browsing the organized information, or by creating or using existing structured reports, which are a great strength of the program. The reports tend to be delivered in matrix form, with topics down one axis and subject companies or people down the other.

This form of delivery is best for competitive intelligence. You can also create "briefing books," structured sets of different reports on a series of topics or subjects. Reports can be output to paper or to word-processor or spreadsheet files.

The system tracks all changes in an audit data file so you can see both the evolution of the information and who was responsible for updating any field. The result is a structured, up-to-date library of gathered information with links to references and attachments you can consult.

The interface design is unexceptional but usable, without either charisma or obvious flaws. It definitely shows its maturity on both counts. Wincite has added a lower-powered browser client as a wide deployment view-only option.

Since the presentation of all knowledge is in a form, much of the librarian's set-up work revolves around form-building through a drag-and-drop screen-painting utility. It's easy to use and effective.

The program has no intrinsic search capability--Wincite recommends you use an existing over-the-counter product, such as the free version of AltaVista's engine. Nor is the product's security as flexible as it should be. You can't take advantage of operating system security or LDAP-enabled directories, and there's no intrinsic security scheme for protecting specific subjects--access is controlled user by user.

While Wincite 5.0 can easily integrate external files (both documents and URLs) into its knowledge base, it doesn't currently integrate with other applications such as groupware applications. With its focus on historic information, it doesn't push information at users, assuming instead that they'll find what they need. Its greatest conceptual strength is the multidimensional data design that lets a knowledge worker examine a slice of the information at any angle, potentially leading to more insightful observations.

Intraspect 1.5
Of the products we tested, Intraspect Software Inc.'s offering is one of the best designed for knowledge-management applications. It creates a "group memory" by storing information in maps customized to each end user and communicated through multiple mechanisms: intranet, networked files, or E-mail.

The program's design presumes most of the users are people who will both contribute to the group memory and consume information. It also assumes that "knowledge" is information made actionable by having an appropriate context.

The system's peer-to-peer model grants wide authority to contribute and inform. The benefit of this approach: It's likely your organization can collect more useful information. The cost: Because the system doesn't concentrate contribution authority, you may get a higher ratio of marginal information.

Given its peer-to-peer concept, it's logical that the Intraspect server is designed around E-mail. Every piece of information has an E-mail address. When you send information to the knowledge base, it appears on a user's desktop client software in a holding area called The Collection Bin. A user can add items from the bin into any of the folders in her personal knowledge hierarchy or set of contexts.

The product uses an object-oriented database to store information, and it's made more efficient by having options of how to receive a piece of information--a Web page, an E-mail message, a file, or other external reference--referenced by pointers to one single instance of the object.

An object can have multiple contexts. For example, a document w ith the Bahamian banking system regulations could have a global banking context, one on organized crime syndicates, and another on U.S. tax avoidance. If a user calls up that document, or discovers it through the bundled search engine, its metadata includes all of the different contexts it's stored in. A user can browse these contexts to find related information and discover new facets to the subject.

The user interface runs on a wide range of platforms. There are native clients for 32-bit and 16-bit Windows, existing SMTP-compliant standard E-mail clients, or a Web browser. The native clients use standard Windows Explorer concepts excellently, down to the function keys. The interface shows Collection Bin in one corner, Explorer-like hierarchy in another, and an embedded Web browser in a third pane.

There are threaded discussions, and users can post comments on any item. Intraspect tracks a lot of metadata that preserves the context. For example, comments on an item remain context-specific, so simply because an item is useful in a number of discussions doesn't mean people looking at comments get a cascade of verbiage about its other contexts. It also means search hits on the comments are more likely to be reasonably useful.

Administration works from the standard client, defining users and groups in a form-based model. The product doesn't currently support LDAP-compliant directory transfer, nor does it leverage information stored in network operating systems, so in large organizations, startup could be a significant manual process.

The program's bundled search function is from Verity Inc.; it supports compound Boolean searches and other flexible abilities and reports the context for the hit. Intraspect has a saved-search capability, too, allowing workgroups to leverage this form of reusing expertise.

Intraspect integrates with common desktop applications such as those in Microsoft Office or Lotus Notes. Integration of external files, though, is not bulletproof. For exam ple, if a file or URL is moved, it's not tracked by the system. Intraspect doesn't have an automated push system, relying on users' discerning items of value, but the system could be applied for both historical and emergent information. There's no API available for extending the system by integrating it with other programs such as push or mining tools.

ChannelManager 2.0 (beta)
ChannelManager 2.0 beta version is a set of tools designed to gather information and content from internal and external sources and use push technology to get it to the users who need it. Created with the assumption that the traditional executive information system provided too little, too late for too much money, and that overloaded Webmasters couldn't beat their existing schedule constrains, DataChannel Inc. designed this toolset around two major ideas.

The first is that no one should have to be responsible for converting to HTML documents to be shared, saving lag time. The second is that all files to be shared, internal to the intranet or file system, or external over the Internet, should share a common location description. In this case, each has a URL, even if they're on the file system.

The goal of the product is not to store information in a central repository for record-keeping or archival recall, but to turn available sources into channels and disseminate information in a timely basis to users (executives, customers, etc.) who need it. People who "own" content channels save their content to a specific folder, which saves that to the Web, automatically converting it into HTML readable form.

DataChannel distributes to a small degree the work of setting up and maintaining the system. An administrator sets up users and groups and sets up channels. You can import users from the Windows NT structure but the product's security is separate from that of the operating system. When you create a new channel, you base it on one of the bundled templates that are clever skeletal structures based on common themes such as "human resources" or "competitors." More often than not, these will save time spent creating channels.

People who have rights to publish channels can add information to them and give subscriptions to users. Subscriptions can be given in the form of permission (in which case the user has to do the subscribing) or forced by the administrator. Each approach will find favor with different kinds of organizations. Each user's desktop is customized with the specific channels he or she has subscribed to. We'd like to see better mechanisms for more peer-to-peer content addition, perhaps as simple as an intrinsic method for suggesting items.

Web Integration is a great strength, in that you can "save to the Web" from within your desktop applications. But later versions would be well-served by better integration with back-end systems such as Netcaster and an easier approach than programming to link it with enterprise resource planning and other systems that hold strategic data.

On the client's software side, options include a full Java application or a lighter choice, a Java-supporting browser or even an HTML client, guaranteeing an opportunity for every deployment style. The user interface design will be familiar to Windows 95 users, with tabbed dialog boxes and other familiar controls. It's very clean, though the hierarchy is limited to three levels which could be a challenge for a few applications.

On the server side, the system runs as a Windows NT service and can use any Open Database Connectivity-compliant database as the store for its information. ChannelManager doesn't hold the actual files and other data, just the pointers to it, so it's designed to be conservative with storage space. Another efficiency is missing--the ability to publish an item to multiple channels with a single push operation and index entry. The product has a published API so users can integrate it with other applications.

The administrator console is quite good, with care given to the interface. The biggest challenge is in some of the nomenclature, which is subtly different from some other products. "Publish," for example, doesn't mean create and then send --it means "make available." We were told the naming was still in flux in this beta version; it may be changed before the product is officially released.

The strength of ChannelManager is its model for rapid proliferation of organized information and low cost of ownership. This is magnified by its API, and the diversity of content you can put into channels.

BackWeb 4.0
BackWeb Technologies' eponymous BackWeb 4.0 can contribute to a range of knowledge-management applications, with its main area of strength being the ability to "push" channels of information to a wide range of desktop users.

Simply, BackWeb is a set of tools designed to gather information from any source (Internet news feeds and channels, internal users, the network file system) and broadcast it to the users who need it. BackWe b uses "push" channel technology as the mechanism to deliver emergent information in any file format. The goal of the product is not to store information in a central repository for record-keeping or archival recall, but to turn available sources into channels and disseminate information on a timely basis to users who need it.

BackWeb uses a hierarchical approach to delivery. A subset of people, administrators, control the channels users can access. This increases the focus of the available information, but by restricting contributions to an elite, it also limits the depth of what you should be trying to achieve with knowledge management.

An administrator gathers channels for a user type and creates a "workgroup," a prepackaged profile of channels for a user set. She then publishes these to the target users, creating a program users execute on their local machines, although once installed, changes made to the workgroup by the administrator are automatically reflected in each user's client so ftware. An optional publishing wizard can distribute channeling power.

BackWeb comes with 600 pre-configured channels that include Internet online journals and news feeds. You can, however, add other channels of your own design, based on Web pages or files.

The options for alerting a subscriber that there's new information are useful and flexible, including an omnipresent ticker-tape display in the current window title bar, a stand-alone ticker, a screen saver alter, or an Info-flash that appears in a pop-up window. Administrators (and users) can assign different alert models for different channels, which makes sense since different kinds of information have different levels of importance and urgency.

The administrator's console is fairly good, using existing Windows design standards, including useful wizards, but there's room for improvement. BackWeb has absorbed other companies, and while the product shows off all the functions of those different offerings, the interface still has multiple personalities reflecting their different ways of doing things. The company says it is working on refining the interface for a future version. There are also some conveniences that would be taken for granted in a more mature product category. For example, you can add only one file at a time to a channel, as opposed to multiple files or the entire contents of a folder.

The end-user interface is very good, with an embedded browser and Windows Explorer-like navigation system. BackWeb has user software available for most desktop users, including Windows 95, 3.11, NT, and Macintosh. The program's search capacity adds value to the system, though saved searches (searches you create once and then execute repeatedly, getting dynamic results each time) currently work for only one kind of channel. There should be no difference in the elementary features of different kinds of channels.

One of the main architectural strengths of BackWeb's design is its "polite" delivery protocol, which minimizes bandwidth use by controlling activity based on other network use and by providing a Zmodem ability to interrupt a transmission and pick the sending process where it left off.

Integrating BackWeb with other programs is possible, especially at the client level, where you can program to its ActiveX components.

KnowledgeX 1.0
KnowledgeX's product of the same name marches to the beat of a different drummer. It's not trying to manage facts alone, but to clarify the meaning of information by analyzing and presenting the relationships between facts.

KnowledgeX is designed to store a centralized repository of categorized information, then deliver it in forms that ease the work of analysis. The program is most effective when its use is centralized to a few experts who can master its concepts, then use its dissemination features to provide targeted, automated reports to a broader set of consumers.

The applications for the product seem very specific: competitive intelligen ce and governmental intelligence agencies. In part, this is because that's who currently uses the program, and in part because the underlying model is, unfortunately, alien to most businesses. The product is based on the idea that knowledge is not the facts themselves, but an understanding of the patterns that connect them into systems. The objects it's currently used to track are more often people or relationship definitions (such as "customer" or "peer" or "board member") or places. Current applications include a global petroleum giant's tracking of competitors' engineers and negotiators' visits to foreign capitals or embassies. This model could be used equally well for analyzing quality assurance results or optimizing service efforts, but people in those fields tend to limit themselves to numeric analyses and miss out on qualitative applications.

Because the traditional KnowledgeX customer is using it for a high-return, business-critical application, it hasn't mattered that the process of getting information into the system is clunky and slow. While the Windows native client to the program uses standard features such as drag and drop very well, the concepts are specific to this product. The traditional customer hasn't had much need for wide delegation of collection efforts, so there is no intrinsic model to "deputize" everyone to help gather system facts. On the other hand, the product is set up so different users see only specific sets of results. This is certainly part of a strong security scheme, but it's also a valuable recognition that attaining knowledge is as much the act of leaving out extraneous data as it is including relevant data.

Dissemination is productive. Users subscribe to topics or relationships between topics they're interested in and when something new comes in the server triggers a notification message through E-mail to subscribers. You can also send the relationship or objects to add or update that user's specific knowledge set.

The product communicates its resu lts via reports, primarily networked diagrams that indicate relationships. If you've entered all your salesfolks' affiliations and avocations and have collected the same kinds of intelligence on customers, you can ask the system to match them up based on commonalities it finds, such as being a Rotarian, a Cubs fan, or a techno-funk aficionado, relationship-reinforcers all.

The organizing of information happens in two ways. Once the extremely critical process of designing the categories and relationships has been produced, the people who have the KnowledgeX client can add information into forms. The product comes with a parsing utility that takes incoming files and tries to infer relationship data from them. This is most useful if the incoming data is based on a handful of basic forms, but even then, you want human intervention for polishing.

If you've designed the structure well enough, the system can reveal lots of knowledge--relationships you may not have known about. It can be tactical, l ike the sales-client connections discussed earlier, or strategic, letting the system build big networks you browse to try and find the answers to questions you didn't know to ask.

Beyond E-mail, the product currently is an island. Given its specialized function, this makes some sense. A beta version of 2.0, due soon, will provide Web clients and the ability to "steal" HTML forms you like off the Web. We didn't look at this beta version.

It's most notable strengths are that it is designed to deal with both emerging and historical information, and that the kind of knowledge you're most likely to acquire from it is something your competitors are unlikely to be getting.

Jeetu Patel is chief technical officer of Doculabs, and Jennifer Harty is a Doculabs analyst.

See related story, " Knowledge-Management Cosmology ."

See the table, " Which Product For Which Function? "