IM Engineer Study Notes
← Back to list
Management & Strategy
#KMS#SECI모델#암묵지#형식지#지식경영
Last updated · 2026-10-08

Knowledge Management System (KMS)

1. Overview

A. Definition

A Knowledge Management System (KMS) is both an information system and a management framework that supports the systematic creation, storage, sharing, and use of the tacit knowledge scattered in the minds of an organization's members and the explicit knowledge captured in documents and data. Beyond a mere document repository, it is defined as an organizational capability infrastructure that underpins the entire knowledge lifecycle — creation, accumulation, diffusion, and reuse — across the triad of people, process, and technology.

The fundamental insight of KMS is that "an organization's competitiveness comes not from the amount of knowledge it holds but from its ability to make that knowledge flow." No matter how many brilliant experts there are, if their knowledge is locked inside individuals it never becomes an organizational asset, and it vanishes the moment a person retires or changes departments. KMS "capitalizes" such easily-lost knowledge as an organizational asset, reducing dependence on specific individuals, shortening the time to solve recurring problems, and raising the quality of decision-making.

The key here is the DIKW hierarchy running from Data to Information, Knowledge, and Wisdom. Raw facts (data) acquire context to become information; information combines with experience and patterns to become knowledge that captures "why it is so" and "what to do"; and knowledge sublimated into insight and judgment becomes wisdom. The essence that KMS addresses is precisely this "knowledge above information" layer, and therein lies its decisive difference from an EDMS (electronic document management) or DW (data warehouse) that merely searches and stores: KMS manages context, experience, and judgment together.

B. Background and necessity

As economies shifted to the knowledge-based economy in the late 20th century, intangible assets such as patents, brands, know-how, and human capability came to account for a far larger share of corporate value than tangible assets like land and equipment. Yet traditional information systems stayed confined to transaction processing (TPS) and structured-data management, failing to capture an organization's most precious asset — "people's experience and know-how." The problem awareness that "knowledge that is neither measured nor shared is not managed" was the starting point for the emergence of KMS.

A classic example is NASA in the United States. After the engineers who had worked on the Apollo program retired en masse, much of the detailed know-how behind the early crewed-spacecraft design was left undocumented, so follow-on projects had to re-verify the same engineering judgments from scratch — a fact frequently cited as a classic lesson in how costly "the loss of knowledge that was never made explicit" can be. It symbolizes how, without a mechanism to systematically capture and preserve knowledge at the organizational level, knowledge assets accumulated at enormous cost perish along with the people who held them.

The real-world necessity is more concrete. First, the risk of core-personnel knowledge loss due to aging and turnover has grown. When the equipment-failure response know-how that a skilled engineer accumulated over decades disappears at retirement, the organization repeats the same trial and error. Second is the inefficiency of repetitive work and duplicated investment. The "reinvention of the wheel" — one department solving from scratch a problem another has already solved, unaware of it — happens throughout the organization. Third, as remote and distributed work became routine, the physical points of contact for collaboration and tacit-knowledge transfer shrank, raising the need for systems to replace the knowledge sharing that once occurred naturally in hallways and meeting rooms. KMS is a managerial and technical solution that seeks to resolve these problems of loss, duplication, and disconnection through the making-explicit and circulation of knowledge.

2. Types of knowledge and the SECI knowledge-conversion model

The theoretical foundation of KMS is Ikujiro Nonaka's SECI model. The model divides knowledge into tacit and explicit, and explains that the mutual conversion between the two repeats in a spiral, expanding and deepening organizational knowledge. Tacit knowledge is embodied knowledge hard to express in words or documents — experience, intuition, mastery — while explicit knowledge is knowledge codified in language and symbols, such as manuals, reports, and formulas. The diagram below shows the overall structure in which the four knowledge-conversion modes circulate and amplify knowledge.

flowchart LR
  T1["Tacit (individual)"] -->|"Socialization<br/>sharing·observing experience"| T2["Tacit (collective)"]
  T2 -->|"Externalization<br/>codify via metaphor·dialogue"| E1["Explicit (concept)"]
  E1 -->|"Combination<br/>combine·systematize explicit"| E2["Explicit (system)"]
  E2 -->|"Internalization<br/>embody via learning·practice"| T3["Tacit (deepened)"]
  T3 -.->|"spiral repetition expands org knowledge"| T1
  style T1 fill:#e8f0fe,stroke:#2f6fed,stroke-width:1px
  style E2 fill:#fef3e8,stroke:#ed8f2f,stroke-width:1px

A. Socialization — from tacit to tacit

Socialization is the process in which tacit knowledge transfers to another person's tacit knowledge by sharing experience together. Apprenticeship-style transfer, on-site observation (OJT), mentoring, and informal conversation are its representative means. For instance, the fingertip sense of a skilled welder is hard to put into a manual, so a novice watches and imitates for hundreds of hours beside them, learning with the body. From a KMS perspective, socialization is the hardest area to codify, but the opportunities for tacit-knowledge transfer can be expanded by increasing points of contact through video collaboration, real-time communities, and mentor-matching features.

The practical implication of this stage is that "knowledge sharing is a matter of relationship and trust before technology." No matter how good the system, if experts lack the motivation and sense of safety to put forward their own know-how, socialization does not happen. Therefore the design of socialization must bundle internal communities (CoP, Community of Practice) with reward and recognition systems.

B. Externalization — from tacit to explicit

Externalization is the process of "drawing out" tacit knowledge in the mind through metaphor, analogy, dialogue, and modeling, converting it into language, pictures, and rules as explicit knowledge — the most creative yet difficult stage of the SECI cycle. Checklists written from expert interviews, troubleshooting guides organizing incident-response cases, and decision trees that turn know-how into rules fall under this. For example, documenting a veteran agent's "knack for handling difficult customers" as a Q&A scenario makes it an organizational asset that even newcomers can use.

The quality of externalization determines the success or failure of a KMS. Because if tacit knowledge is clumsily codified it becomes a "dead document" stripped of context and goes unreused. Therefore one must record together the background of "why it is done this way" and the conditions of application (when it works and when it does not), and recently generative AI has greatly lowered the cost of externalization by summarizing and structuring expert conversations.

C. Combination — from explicit to explicit

Combination is the process of collecting, classifying, combining, and reconfiguring scattered pieces of explicit knowledge to create a more systematic and comprehensive new explicit knowledge. A standard methodology synthesized from multiple project reports, a knowledge system (taxonomy·ontology) bundling individual technical documents, and an analytical report synthesizing data are representative. It is the area to which databases, search engines, and document management systems contribute most directly, and the stage where the "systemic" character of KMS shows most clearly.

The core task of the combination stage is the design of the classification scheme (metadata·tags·ontology). If knowledge is poorly classified, it is not found and merely accumulates without being used. For example, if a manufacturer tags and systematizes 50,000 quality-defect cases over ten years by process, part, and cause, then when a new defect arises it can immediately find similar cases and cut response time from several days to several hours.

D. Internalization — from explicit to tacit

Internalization is the process of learning explicit knowledge and, through practice, embodying it again as an individual's tacit knowledge. "Learning by doing" — reading a manual and repeatedly practicing until it is in the body — simulation training, and e-learning fall under this. The individual who has gone through internalization adds new experience to create yet another tacit knowledge, which again becomes the starting point of socialization, completing one turn of the knowledge spiral.

An important implication of this cycle is that a KMS does not work if it supports only one of the four stages. Many KMS failure cases were the result of concentrating only on externalization and combination (document accumulation) while neglecting socialization and internalization (people's learning and sharing). In other words, a KMS must be the design of both a technology platform and a learning organizational culture at the same time.

Another point worth noting is that the SECI spiral expands by broadening its spatial field (Ba) from individual → group → organization → inter-organization. Nonaka called the contextual space in which each conversion occurs "Ba": socialization occurs in the face-to-face, on-site "originating field," externalization in the field of dialogue, combination in the field of cyberspace, and internalization in the field of practice. Designing a KMS ultimately means providing these four fields both physically and digitally, and therefore a good KMS takes the shape not of a single repository but of an ecosystem in which meetings, communities, collaboration tools, and learning platforms are organically connected.

3. KMS architecture and components

A KMS is broadly composed of a multi-layer structure — knowledge-source layer → collection·processing layer → storage·classification layer → use·service layer — and a governance·security axis cutting across them. The architectural detail diagram below shows the data·knowledge flow in which knowledge streams in from sources inside and outside the organization, is processed and stored, and is ultimately delivered to users.

flowchart TB
  subgraph SRC["Knowledge sources"]
    S1["member tacit knowledge·CoP"]
    S2["documents·email·collab tools"]
    S3["core systems·DW·external data"]
  end
  subgraph PROC["Collection·processing"]
    P1["collect·cleanse·add metadata"]
    P2["classify·ontology·index"]
  end
  subgraph STORE["Storage·knowledge base"]
    K1["knowledge repository"]
    K2["search engine·vector DB"]
  end
  subgraph SVC["Use·service"]
    U1["search·recommend·Q&A"]
    U2["expert locator·collab·community"]
  end
  SRC --> PROC --> STORE --> SVC
  GOV["governance·security·access·quality"] -.-> PROC
  GOV -.-> STORE
  style GOV fill:#fdecea,stroke:#d93025,stroke-width:1px
  style STORE fill:#e8f0fe,stroke:#2f6fed,stroke-width:1px

The knowledge-source layer spans members' tacit knowledge and communities of practice (CoP), the explicit knowledge in documents·email·collaboration tools, and core systems·data warehouses·external data. The collection·processing layer gathers knowledge from these sources, cleanses duplicates and errors, adds metadata, and indexes it according to a classification scheme (ontology). The storage layer's knowledge repository and search engine make cleansed knowledge storable and searchable, and recently a vector database for semantic search is integrated. The use layer delivers knowledge to the field through search·recommendation·question-answering, an expert locator, and collaboration·community features.

The governance·security axis cutting across all these flows is the key to keeping a KMS from degrading into a "garbage repository." It includes knowledge-quality management that verifies the accuracy and currency of registered knowledge, security that controls access rights and confidentiality grades, and reward·evaluation systems that promote knowledge registration and use. The table below organizes the components along the three axes of people, process, and technology (an aid to the prose explanation).

Axis Components Role
People CKO·knowledge managers·CoP·experts Agents of knowledge creation·verification·sharing; cultural rooting
Process knowledge-cycle procedures·quality mgmt·reward systems Institutionalizing create→store→share→use
Technology repository·search·collaboration·AI·vector DB Infrastructure for accumulating·finding·delivering knowledge

Among the three axes, especially important roles on the people axis are the Chief Knowledge Officer (CKO) and communities of practice (CoP). The CKO is an executive-level officer who aligns knowledge management with enterprise-wide strategy and oversees budget·reward·culture, serving as the hub that keeps a KMS from staying an IT department's technology project and makes it a management agenda. A CoP is an informal community in which members with the same interests·work voluntarily exchange knowledge across formal organizational boundaries — the field where socialization and internalization actually occur. No matter how excellent the technology and repository are, if this people axis is empty the KMS ends up an empty vessel that goes unused, as countless adoption-failure cases commonly attest.

4. Build types and strategy comparison

KMS build strategies are broadly divided into two branches, and which one is chosen fundamentally changes the character of the system. The distinction between the codification strategy and the personalization strategy proposed by Harvard Business School researchers is representative. The codification strategy is an approach that codifies knowledge into documents, accumulates it in a repository, and reuses it in a "people-to-document" manner, suited to work high in standardization and repetition (e.g., consulting methodologies, call-center responses). By contrast, the personalization strategy is an approach that leaves knowledge with people and transfers it through "person-to-person" dialogue, fitting work that demands high expertise and creativity (e.g., strategy consulting, R&D).

The key is that the two strategies are not mutually exclusive, but a primary-to-secondary ratio of about 8:2 must be made clear. Pursuing both half-and-half in a half-hearted way often leads to failure in both document accumulation and community activation. The reason is that the two strategies require different organizational cultures, reward systems, and directions of IT investment. The codification strategy rewards "the person who registered many good documents," while the personalization strategy rewards "the person who helped colleagues well." For example, one global consulting firm invested in codification for the business unit whose share of reselling standard solutions was large, and in personalization for the unit that formulated bespoke strategy, aligning business characteristics with knowledge strategy.

As an actual application case, a domestic manufacturing conglomerate, to solve the generational discontinuity of equipment-maintenance know-how, codified skilled-technician interviews into troubleshooting cases (codification strategy) and simultaneously ran an intergenerational mentoring community (personalization strategy) to promote socialization and internalization. As a result, the mean time to repair (MTTR) for new equipment failures is reported to have shortened significantly, which is assessed as a success factor of combining the two strategies to fit work characteristics.

To understand KMS correctly, one must clarify its differences from seemingly similar adjacent systems. KMS is often confused with an electronic document management system (EDMS), a data warehouse (DW), and business intelligence (BI), but the objects and purposes they address are fundamentally different. Whereas an EDMS manages "the storage·versioning·approval of documents (explicit knowledge)," a KMS connects the "context·experience·know-how" contained in those documents and even the "people" who hold it. Whereas DW·BI aggregate and analyze structured data to show "what happened," a KMS handles the judgmental knowledge of "why it is so" and "what to do." In short, a KMS is not a replacement for other systems but closer to a higher-order concept that integrates the knowledge layer on top of them. The table below compares and organizes those differences.

Category Primary object Core purpose Nature of knowledge
EDMS electronic documents storage·versioning·approval·retention explicit-centric
DW/BI structured data aggregation·analysis·status visualization information layer
KMS tacit+explicit+people creation·sharing·reuse·judgment support knowledge·wisdom layer

The practical implication of this comparison is that even an organization already running an EDMS or BI finds that these alone do not manage "people's know-how." Therefore, adopting a KMS is more cost-effective when approached not as building a new system but as placing a knowledge-connection layer of expert locator·community·semantic search on top of existing assets. Conversely, if one overlooks this difference and builds a KMS merely as an "advanced document repository," it ends up becoming yet another EDMS and fails to achieve its original purpose of tacit-knowledge sharing. This is the structural cause of why many firms' first-generation KMS failed, and the reason one must first clarify "what to manage" when formulating a knowledge strategy.

5. (In depth) Combining generative AI·RAG with KMS

The biggest recent change in KMS is its combination with generative AI·large language models (LLM). The limitation of a traditional KMS was that "it can search but does not give answers." The user had to know the right keywords and read and interpret the found documents themselves. By contrast, a RAG (retrieval-augmented generation)-based KMS takes the user's natural-language question, finds semantically similar content in the knowledge base via vector search, and on that basis has the LLM generate an answer that cites its sources. That is, it evolves from a "system that finds documents" into a "system that answers questions."

The change on the standards side is also noteworthy. In 2018 the International Organization for Standardization established ISO 30401 (Knowledge management systems — Requirements), framing knowledge management — until then done differently at every firm — within the framework of management-system standards (the same family as ISO 9001 for quality and ISO 27001 for security). This standard specifies as requirements the knowledge lifecycle (acquiring·developing·retaining·transferring·applying) and supporting culture·leadership, guiding organizations to run knowledge management not as a one-off campaign but as a continuously improving management system (PDCA). Now that generative-AI adoption is accelerating, such a standard is, if anything, growing in importance as a reference point that guarantees "the quality·governance of the knowledge AI will handle."

This change accelerates each stage of the SECI model. An LLM lowers the cost of externalization by auto-summarizing and structuring expert conversations, automates combination by synthesizing scattered documents into new knowledge, and supports internalization learning through personalized question-answering·tutoring. For example, an AI assistant that takes internal rules·past projects·technical documents as its knowledge base answers a new employee's questions instantly with sources, shortening the onboarding period.

As a concrete industry application, a global IT services company vectorized hundreds of thousands of past incident tickets and technical documents to build a RAG-based support assistant, so that when an engineer asks in natural language it presents similar incident cases and solutions with sources. By this, the rate of resolution at first response rose and average handling time fell, and the knowledge-utilization rate is reported to have improved greatly versus a traditional keyword-search KMS. This is a typical case in which accumulated knowledge that "went unused because it was not found" finally began to flow through generative AI.

That said, combining generative AI carries clear risks. First, the hallucination by which an LLM plausibly fabricates groundless content can inject wrong knowledge into organizational decision-making, so citing sources and presenting evidence must be enforced and important answers must be verified by humans. Second, to prevent confidential·personal information from leaking to an external LLM API, one must consider search that reflects access rights (permission filtering) and an in-house closed-network model·private deployment. Third, if the knowledge base's quality is low, "Garbage In, Garbage Out" means AI answer quality collapses with it, so one must heed the paradox that AI adoption in fact raises the importance of knowledge-quality management even further.

6. Considerations and implications

From a professional engineer's perspective, for a KMS the operational design that "keeps it continuously alive and moving" determines success more than adoption itself. Strategy should be formulated around the following four points.

  • Culture·reward take priority over technology: The majority of KMS failures stem not from technical defects but from a knowledge-monopoly mindset — "if I share knowledge my standing weakens" — and the absence of sharing motivation. One must link knowledge registration·use·mutual evaluation to the HR·performance system and build a sharing culture in which management leads by example. Change management must accompany it, on the premise that technology only "enables" sharing but cannot "make" it happen.

  • Knowledge quality·lifecycle management: If knowledge is only accumulated and never updated, a KMS soon becomes a "knowledge graveyard" piled with old and wrong knowledge. One must assign an expiration date and an owner to knowledge and institutionalize a lifecycle that periodically verifies·updates·discards it. Because obsessing only over quantitative metrics like registration count sacrifices quality, one must evaluate with use-centered metrics such as usage frequency, reuse rate, and contribution to problem-solving.

  • Balance between security·governance and sharing: A trade-off exists whereby the wider knowledge sharing spreads, the greater the risk of confidential leakage. One must classify the confidentiality grade of knowledge and apply role-based access control (RBAC) and data classification·masking, yet find the balance point of "the minimum control necessary" so that excessive control does not suffocate sharing. In particular, when combining generative AI, governance to prevent the learning·exposure of personal information·trade secrets is essential.

  • Phased build and ROI proof: Rather than an enterprise-wide big-bang adoption, a realistic strategy is to start with a pilot in the core work areas where the pain of knowledge loss·duplication is greatest, quantitatively prove the results (shorter search time, less rework, shorter onboarding, etc.), and scale out. Because a KMS's benefits are intangible and thus hard to measure as ROI, one must design measurable performance metrics from the start of adoption to secure the justification for investment.

  • Outlook on integration with related technologies: A KMS is no longer a standalone system but is evolving into a knowledge layer integrated with collaboration tools·search·data platforms·generative AI. Through linkage with data governance·master data management (MDM)·knowledge graphs·vector DBs, it is expected to develop into a foundation for organizational intelligence encompassing both structured data and unstructured knowledge.

  • Metric and maturity management: A KMS's performance must be measured not by input metrics like registration count but by outcome·use metrics such as knowledge reuse rate, search success rate, the share of knowledge-based decisions, and the shortening of new-hire onboarding. Furthermore, the key to sustainable operation is to diagnose the organization's knowledge-management level with a maturity model such as "initial–repeatable–defined–managed–optimizing" and to establish a roadmap that raises people·process·technology investment in balance stage by stage.

References

In one line: A Knowledge Management System (KMS) is an integrated people·process·technology framework that creates·shares·uses tacit and explicit knowledge through the SECI cycle; its success is decided by the design of culture·quality·governance, and it is lately evolving, combined with generative AI·RAG, into a "knowledge infrastructure that answers."