Knowledge Management in the Service and Support Business Kemal A. Delic Umeshwar Dayal Hewlett-Packard Hewlett-Packard kemal_delic@hp.com umesh_dayal@hp.com measurement methodology, and there are few reliable reports of the impact of knowledge management on Abstract business metrics [Del00]. As we are entering the New Millenium, we are We believe that knowledge management (KM) today witnessing a global shift toward a society of [Dav00] plays a crucial role in the IT support and service services. The evolution of the world's workforce environment. It directly impacts the productivity and structure indicates clearly the magnitude and training of support personnel, and this in turn justifies importance of this mega-shift. The introduction investments in knowledge management systems, programs of new, exciting technologies has led to a kind of and associated processes. To give an idea of the potential New Economy that is based on the huge and market size, industry estimates indicate that for each rapid flows of data, information and knowledge. dollar spent on hardware and software products, seven to The Internet has had a most profound impact on fourteen dollars are spent in associated training, support business and society, enabling the quick spread and services. of service industries and technologies. We observe that cost reduction and acceleration of From the business point of view, KM impacts profit by business processes are the most obvious reducing the costs of doing business and creates new consequences of this singular phenomenon. In streams of revenues through the introduction of new, such an environment, the management of IT innovative services. From a strategic point of view, it is support services is becoming critical for business seen as a potentially strong business growth generator (> profitability. 50%). Unfortunately the KM field is saturated with hype and buzzwords, so that real, documented KM success stories are rarely published. We will focus our attention 1 Introduction on deployment of KM in IT support services as one of the most important practical domains, and we will describe The management of support services includes the efficient our experience at Hewlett-Packard with the deployment of deployment of people, processes and technologies so as to one KM system for IT support services. improve operational business parameters and financial indicators. Knowledge Management (KM) is a general IT support and services are typically organized as a multi- umbrella term that encompasses several techniques and tiered operation, consisting of help desks or service desks, processes whose main common objective is to deal and operational service centers supporting the IT successfully with various inefficiencies in operational infrastructure and delivering various services. Customers business processes and to create business value. obtain support and services by contacting the first line service desk. Problems that cannot be resolved in the first Intuitively, it is clear that knowledge plays a crucial role line are escalated to higher, but more expensive, levels of in human business activities, and that it has significant expertise. In addition to telephone or email-based help monetary value to an enterprise. However, measuring the desks, enterprises are moving towards Web-based service impact of knowledge management has proved to be portals, where customers can directly obtain access to IT difficult, since there is no commonly accepted support and services. In this paper, we will show in more detail how knowledge management can enhance the The copyright of this paper belongs to the paper’s authors. Permission to copy efficiency of traditional IT operations and enable the without fee all or part of this material is granted provided that the copies are not delivery of new types of services. made or distributed for direct commercial advantage. Proc. of the Third Int. Conf. on Practical Aspects of In the second section of this article, we introduce Knowledge Management (PAKM2000) Knowledge Management. The third section deals Basel, Switzerland, 30-31 Oct. 2000, (U. Reimer, ed.) specifically with KM in the Information Technology (IT) http://sunsite.informatik.rwth-aachen.de/Publications/CEUR-WS/Vol-34/ domain. In the fourth section, we outline the architecture of an IT KM system being deployed within Hewlett- K. A. Delic, U. Dayal 7-1 Packard Corporation. In the fifth section, we sketch the ontology consisting of contexts, subjects, and topics. The evolutionary architecture of IT support and service third stage includes a battery of automated refinery systems, concluding with a few insights about the future processes that include on-line analytical processing role of KM in the service industry. (OLAP), data mining and knowledge discovery, and information visualization techniques [Fay96, Min99]. The results of this stage are usually in the form of actionable 2 Knowledge Management Defined knowledge that can be disseminated, shared, and delivered to knowledge workers for decision making. Today, much of this dissemination is Web-based. Knowledge is primarily embodied in human expertise and experience. It has first to be captured and expressed We will try to classify typical KM scenarios as reported in explicitly, then transformed and represented in various industry and academia. The most frequent situation is repositories, then disseminated and shared by knowledge known as Knowledge Sharing (Figure 2. - left) in which workers who exploit it for making business decisions, and KM is seen as the technique to transfer expertise from the finally these actions may in turn lead to the creation of top performers or domain experts (20%) to the rest of the more knowledge. Knowledge Management can be defined population (80%). In the IT support and services context, as a set of generic processes aligned with the four for example, this involves the transfer of problem solving principal transformation phases: gathering, organization, knowledge from the more proficient service desk agents to refining, and dissemination of knowledge (Figure 1). For other agents, thus improving the productivity of the whole the purpose of this article, we will consider Knowledge population. Management as a combination of several disciplines and techniques. The next common scenario is known as Knowledge Harvesting (Figure 2. - right) where we collect knowledge In Figure 1, we summarize a generic KM methodology codified in various repositories. A user harvests and process that can be applied to the majority of known knowledge by querying the various repositories. In KM approaches in various domains such as consulting, response, a list of solutions and answers is offered to the customer service & support, training, human resources, user, who will thus experience augmented problem- finance, product creation, sales and marketing, and strategic management. L H KNOWLEDGE HARVESTING L KNOWLEDGE H KNOWLEDGE Collect SHARING PROBLEM SHARING SOLVING POWER Data & Introduce Context, PRODUCTIVITY Information Subjects, Topics 20 % 1 1 2 QUERY KNOWLEDGE 2 MANAGEMENT Gathering Organizing ORDERED LIST - SOLUTIONS 3 File System, Indexing, 80 % - ANSWERS Database, Classification, - ADVICES Spreadsheets Categorization - HINTS USER 4 POPULATION REPOSITORIES Web, Java, Data Mining, OLAP, Messaging, Knowledge Discovery Brokering Refining Disseminating solving power. 3 4 Pack & Deliver Measure & Adapt Discover Relationships, Figure 2. Knowledge Sharing and Knowledge Harvesting Synthesize, Abstract, Aggregate In the first two scenarios, the knowledge management system does not differentiate among users. However, the Figure 1. Generic Knowledge Management Process ultimate objective of IT tools is to be able to learn and adapt to individual users’ habits, preferences and evolving The first stage is the gathering or capture of raw data or needs. There are several products that are able to capture information collected from operational processes. This interactions with users via sophisticated adaptive information may be in the form of structured data (e.g., algorithms in which the key asset is Captured Knowledge relational databases, log files, event traces), or in the form (Figure 3, left). Since this knowledge is captured in a of semi-structured or unstructured documents. The second suitable user model, it can be used to provide a user with stage organizes the information through indexing, solutions that are customized to his needs, and hence categorization and classification into a domain-specific directly impacts the user's productivity. K. A. Delic, U. Dayal 7-2 insights from knowledge gained over a huge population of L H IT systems and over very extended periods of time creates KNOWLEDGE DISCOVERY wisdom, e.g., “It is cost-effective to invest in a high L DECISION MAKING KNOWLEDGE KNOWLEDGE SHARING CAPTURING H QUALITY availability solution with server fail-over capability for the PERSONAL production department, because a server outage there PRODUCTIVITY BUSINESS INTERACTIONS CALL LOGS CALL LOGS INTELLIGENCE results in a loss of 500 person days of work.” Business enterprises that are able to efficiently manage these kernel DATA MINING WEB LOGS entities (data, information, knowledge, wisdom) are ADAPTATION LEARNING CUSTOMER typically market leaders and consistent business winners. RELATIONSHIP INTERACTION MANAGEMENT LOGS SYSTEM - TOOL FRONT END BACK END Business enterprises are huge generators and consumers of data, information, and knowledge. Terabytes of data are Database Figure 3. Knowledge Capturing and Knowledge GROSS AVERAGES Number of Events within Discovery Non Human Enterprise per Day Trigger 5000 - 15000 Events - Burst Low Disk Space $ 100s Finally, the most recent KM paradigm is known as Knowledge Item Created $ 10s Knowledge Discovery (Figure 3, right) wherein many Event Data Item $ 1s Information Knowledge Document Item Created different types of information about user interactions (e.g., Blue Screen Created Trouble Ticket Case $ 1000s Death transaction logs, case histories, web logs, call logs, traces Human Wisdom Service Initiated Sizing & Architecture of problem solving sessions) are amassed and analyzed Number of Objects Creation & Delivery GROSS AVERAGES Number of Events with sophisticated large-scale algorithms. These create 1000 PCs having Number of Users & Operators Topology Knowledge Base 40 Problems per Day causing Data Store insights and recommend optimal business actions aimed at 120 Calls improving the quality of decision making. 3 IT Knowledge Management Figure 4. Event and Value Chain within Enterprise IT Different types of knowledge are encountered in IT created by machines and humans in their daily operational domains: product knowledge, procedural knowledge, legal tasks. Generic event creation rates within the knowledge, behavioral knowledge, customer knowledge infrastructure of a typical enterprise are illustrated in and topological knowledge Within the domain of IT Figure 4 in the context of PC support and service. One support and services, knowledge management can be must take into account these figures when planning and regarded as a process that impacts productivity and sizing an enterprise architecture. From experience, we can learning. It is typically achieved through the capture, attach a dollar value to each item in the chain and talk articulation, and reuse of relevant domain knowledge. about its positioning in the value chain. So, if we spend $1 This strategy fits domains in which the problems to capture event traces, then additional processing might encountered are simple, repetitive in nature, and for which bring us into the $100 range per knowledge item (e.g., standard solutions exist. document) if we are able to create one in a cost efficient manner. Some companies will still earn money by For the purpose of this article, we will define key IT exploiting even a simple data capture process. Others will terms: data, information, knowledge and wisdom. In the cash in on the wisdom acquired from careful knowledge typical IT service domain, data is a collection of observed management and transforming knowledge into the ability facts or events, such as "3 disk errors from server xyz to deliver superior end-to-end services. have been recorded in the last 10 minutes". Information is derived from data by summarizing or aggregating data In the generic IT landscape, one typically encounters a from several sources and over a period of time, e.g., "the multi-tiered operation consisting of help desks or service failure rates of systems with a configuration similar to desks, and operational service centers supporting the IT server xyz is 5% over a year; or, server xyz has been infrastructure and delivering various services (Figure 5). down 20% of the time in the past 3 months, and such failures affect 250 users in the production department". Activities in the help desk and service desk require Knowledge is in the form of business rules or patterns knowledge sharing and reuse. One analyst usually covers derived from large collections of data and information, 200-300 desktops, handles daily 20-30 phone calls, whose e.g. "3 disk errors within 15 minutes from systems similar duration is typically 20 minutes and whose average cost is to server xyz are predictive of server failure with 90% around $50. Service Desk analysts are focused on simple confidence”. Deriving actionable business decisions and information inquiries, desktop problems, and application K. A. Delic, U. Dayal 7-3 usage problems. Their work is usually supported by a discovered knowledge is to create business insights (for problem-solving knowledge base. This knowledge base example, why are users from one line of business more typically consists of a large collection of documents (case efficient than others). We believe that certain core histories of problems encountered before and their technologies such as data mining and machine learning, resolution, product documentation, and the like), and Bayesian reasoning, neural networks and genetic other troubleshooting and diagnostic tools. algorithms, information retrieval and natural language processing, information visualization, and intelligent Operational centers within IT departments in big agents, some of which have being actively investigated enterprises take care of more complex problems, for during the last 50 years, will play an important role in the which unique solutions are created through the future development of decision support services. communication and cooperation of several human experts. Accumulated experience and insight from these large (A typical enterprise has $3 billion in annual revenues, fields of research are beginning to appear in innovative 9000 employees, 500 IT professionals, 4300 desktops, applications and tools. The notion of "knowledge" 470 servers spread over 40 sites.) Knowledge provides the glue for combining and exploiting techniques management here usually involves indexing of human from these different research disciplines. expertise to enable collaboration. IT personnel in the operational center are usually focused on more complex Enterprise information systems should be designed with domains such as network management, server the user model explicitly included into all aspects of the management, and system management. They normally design. The primary point of focus for user interaction is a encounter complex but non-repetitive type of problems user view or portal, which provides the working context that are frequently specific to the particular set-up and the for the user, guides the user through his tasks, serves up particular enterprise. relevant data, information and knowledge to aid him in decision making, and (ideally) evolves and adapts to the All IT processes are supported by computerized systems, user’s needs. The machine emulates intelligent activities so that huge volumes of data are created and stored daily. such as answering questions, giving advice, solving The process of transforming data into information and problems, and playing what-if scenarios. knowledge is not completely automated, and recent investigations in text mining and business intelligence Frequent, Tactical, $ Rare, Strategic, $$$ represent efforts in that direction. Knowledge is extracted Top Management and captured in a model able to emulate certain human Decisions behaviors. Such knowledge could be embedded in Senior Management Knowledge products (giving differentiating or innovative features) or Data Mining it could drive various processes. Management Decision Support Document Collections System Document Knowledge Workers Management Work Desk Net Server Text Collections Contact Control Console Application Service Supported Population Intangible Desktop Support Collections Management Service Events Operational Services Calls Service Center Strategy Tools Efficiency Infrastructure Problem Resolution Figure 6. Cooperative Systems: Users and Collections Dashboard Time Information Service Desk 3 In Figure 6, we sketch segmented user populations and Tangible Management repositories used to support decision making. All users Calls are considered to be knowledge workers, the principal Enterprise : US$ 3 bn in annual revenues, 9000 employees, 500 IT Efficacy difference among them being the frequency of decision 2 professionals, 4300 desktops, 470 servers Problem Resolution making and the associated value-at-risk. Typically, we spread over 40 sites 1 Rate provide decision support systems for two main areas: first, for users who have to make frequent and not too risky or costly decisions (service-desk agents, for example); and Figure 5. Generic IT Landscape: Service Desks And second, for users who have to make infrequent but risky or Operational Centres costly decisions (a CIO, for example). Strategic managers and senior managers are typically supported with knowledge automatically extracted from different repositories deployed within a decision support system [9]. For these managers, the primary purpose of 4 An IT Support & Service Architecture K. A. Delic, U. Dayal 7-4 At Hewlett-Packard, we have implemented an IT information from the user’s environment can enable the architecture for support and services that is based on the expansion of another line of support services (e.g., principles of knowledge management outlined above predictive support). [Del98]. This architecture is depicted in Figure 7. In their daily work, our help desk analysts use HP WiseWare1, a Realizing that the integration of the whole support knowledge based system that contains various types of landscape (self-healing, self-support, technical support knowledge documents covering well over 70 standard and administrative support) is an obvious future necessity, products, as well as customer specific problems, for a total we designed the architecture to cover all relevant problem of about 150,000 solutions. Approximately 70 to 80 areas with competent user support aimed at the integration percent of the analysts use the WiseWare knowledge base and correct management of data, information, and regularly. To illustrate the problem-solving power of knowledge. WiseWare, let us mention that it is equivalent to a library containing 2000 books of 100 pages, which could be packed onto 20 bookshelves, each containing 100 books. From the word count perspective, WiseWare contains circa 47 million words, which is more than any currently available on-line encyclopedia. A search engine enables pinpointing of the relevant content with 2,3 word long query (long-term average length of the query), which is equivalent to finding in the above library the appropriate subchapter (10-15 pages) or even a single page in a few seconds. On average, 8-15 percent of the users provide feedback (annotation) about content usage, and this feedback drives content management tasks. HP RuleWare is an extension of WiseWare that provides procedural knowledge about what a customer is entitled to and procedures for handling customers. Currently, HP RuleWare contains circa 6000 rules, which capture Figure 7. Layered Knowledge Management Architecture knowledge about customers. The upper layer of the architecture contains a front-end These two knowledge repositories represent the middle system (which in our architecture is a Web-based portal layer of our architecture. All user interaction traces are called Nimbus) that delivers support and simple advice captured in the back-end system, Search & Access Mine. enabling end-users to self-solve their problems, while which is a huge repository of interaction traces containing gathering information from the user's environment and several million sessions that are collected from several capturing usage patterns. All user interactions with the global servers, pre-processed, transformed and stored in knowledge-based system are captured. These include all the appropriate format. Such transformations typically queries launched, documents retrieved and accessed with create tables, charts and graphs, and drive the computation user ratings of the documents. of various business indicators. The back-end layer of the architecture contains traces (i.e., The design of HP WiseWare combines Web technology logs) of all user interactions with the knowledge-based with an indexing engine to enable good coverage of system. This enables the profiling of individual users, Wintel-related problems for help desk analysts. A well drives the adaptation (personalization) of the system’s controlled and ISO 9001-certified knowledge responses to meet each user’s needs, and the reporting and management process (Knowledge Refinery) enables a alerting functions that are provided to the business constant feed of relevant source material for HP managers. Profiling knowledge is stored in database tables WiseWare. As the user population has shifted from help and associated procedures. desk analysts towards service desk and channel partners, the nature and complexity of the service calls and support To summarize, this layered architecture contains problem has changed as well. For instance, there is an increased solving knowledge in various document repositories; emphasis on self-support. The productivity of end users knowledge about user behavior is captured in user models; can be greatly increased by providing assistance for and interaction models provide the basis for strategic simple, repetitive problems. Exploiting real-time decision making. Judicious management and exploitation of knowledge enables an evolution from problem solving 1 HP WiseWare, HP RuleWare, HP Search & Access Mine are assistance toward a decision support system [Tur98] for internal HP products K. A. Delic, U. Dayal 7-5 users and managers. Consequently, IT support processes will be governed by business metrics and measurements, We see this message-brokering paradigm evolve into a instead of rough indicators and intuition. knowledge brokering paradigm, where components can publish and subscribe to knowledge (IT knowledge, 5 Enterprise Service Architecture business knowledge, legal knowledge, financial knowledge), not just to messages or events. Ultimately, IT support services are just one of the touch points that this will lead to a dynamic marketplace for electronic typical enterprises have with their customers and partners. services [Qim99] within an enterprise, and even across In general, within an enterprise we find several different multiple enterprises, in which service providers advertise repositories containing data, information, and knowledge their capabilities and the types of knowledge they can about different aspects of the business. For instance, provide, and consumers can dynamically find service document repositories contain project, product, process, providers that can satisfy their requirements. and workflow information. Structured databases contain financial data, accounting, sales and marketing figures, 6 Conclusions operational and business tables. Messaging systems contain communication and collaboration traces. Knowledge is a crucial ingredient for enterprise IT service Interaction histories contain log files and web-access & support. Knowledge may appear explicitly in interaction databases. Various data & knowledge feeds documents, it may be embedded in algorithms, tools and come in and out of the enterprise. Ideally, we would like processes that assist users in automatic problem solving. to create an integrated architecture that will augment the Knowledge Management is still more an art than an support services we described earlier with many other established scientific theory or commonly accepted types of decision support services for various corporate methodology. It combines several techniques and users and for various applications such as customer technologies aiming to capture, articulate and disseminate relationship management, marketing campaign design and knowledge [IDC99, Gar99]. In different areas, it takes optimization, consulting, and so on. different forms, but its absence will always be very palpable. Companies that are able to transform data and Knowledge information (costs) into knowledge assets (values) can Document Database Base Systems Systems Base Systems Workflow Systems claim that they have successfully applied Knowledge Management. I - Engine Integration Bus References Data Information Interaction [Dav00] Thomas H. Davenport, Laurence Prusak, Knowledge Feeds Message Histories & Systems Working Knowledge, Harvard Business School Press, Stores & Systems May 2000. [Del00] Kemal A. Delic and Birgit Hoellmer, Knowledge-Based Support in Help-Desk Environments, Figure 8. Integrated Enterprise Service Architecture IEEE IT Professional, Vol. 2, No. 1, pp. 44-48, January/February 2000. This requires an architecture that supports the integration of the various repositories, the creation of business [Fay96] Fayyad, U., Piatetsky-Shapiro, G, Smyth, P., and intelligence from the information captured in these Uthurasamy, R., (eds.), Advances in Knowledge repositories, and the creation of services that exploit this Discovery and Data Mining, AAAI Press/ MIT Press, business intelligence. We believe that basing such an 1996. enterprise integration architecture on a publish-subscribe paradigm offers important benefits. In this paradigm, the [Min99] Hao, Ming; Dayal, Umesh; Hsu, Meichun; different components communicate via messages. Baker, Jim; D'Eletto, Robert, A Java-based Visual Components can subscribe to particular types of messages Mining Infrastructure and Applications, HP Labs (events) published by other components. The publish- Technical Report HPL-1999-49 available at subscribe middleware provides message brokering, i.e., http://www.hpl.hp.com/techreports/1999/HPL-1999- transparency between publishers and subscribers, 49.html asynchronous delivery of messages, scalability and high availability. K. A. Delic, U. Dayal 7-6 [Del98] Kemal A. Delic and Dominique Lahaix, Knowledge Harvesting, Articulation, and Delivery, HP Journal, Vol. 49 No. 2. pp. 74-81. May 1998. [Qim99] Chen, Qiming; Hsu, Meichun; Dayal, Umeshwar; Griss, Martin, Multi-Agent Cooperation, Dynamic Workflow and XML for E-Commerce Automation, HP Labs Technical Report HPL-1999-136 available at http://www.hpl.hp.com/techreports/1999/HPL-1999- 136.html [IDC99] IDC, Bulletin "Knowledge Management Factbook", September 1999. [Gar99] Gartner Group, Conference Presentation, Knowledge Management - Everything and Nothing, 1999. [Tur98] Efraim Turban and Jay E. Aronson. Decision Support Systems & Intelligent Systems (5th ed., Prentice Hall, 1998; ISBN 0-13-740937-0) K. A. Delic, U. Dayal 7-7