Key Takeaways
Treat taxonomy as negotiated infrastructure, not a tidy filing tree
More than filing. A taxonomy is a classification scheme with meaningful labels and explicit relationships. It acts as a map of a knowledge domain, helping people anticipate where useful information may be found. But organisational knowledge also lives in routines, tools, stories, and people, so arranging documents is only part of the task. The broader goal is to make knowledge easier to manage, connect, and apply.
A structure made together. The author calls the ongoing work of discovering vocabularies, negotiating categories, and keeping them relevant “taxonomy work.” Unlike a fixed scientific classification, an organisational taxonomy must reflect the people and circumstances it serves. Its categories can help users remember and connect ideas, but they also make some things visible while pushing others out of view.
What's striking is the shift from treating classification as a neutral filing decision to seeing it as part of an organisation's operating system. This connects to a wider insight in sociology: administrative categories do not merely describe people and work, they shape what institutions can notice and act upon. That makes taxonomy design consequential, even when it looks mundane. The practical restraint is equally important. A map can improve coordination, but it cannot contain the full territory of tacit skill, trust, or judgment. Its value depends on being useful, not exhaustive.
Choose lists, trees, matrices, or facets for the job they must do
Form follows purpose. A short list can be clearer than a complex tree. Trees show relationships such as whole-to-part or general-to-specific, but a tree can mislead when its branches quietly switch between different kinds of relationships. Strict hierarchies, where categories must be mutually exclusive and consistent, are rarely a good fit for messy organisational knowledge.
Use more than one view. Matrices compare items along two or three dimensions. Mendeleev's periodic table arranged elements by mass and recurring properties, making patterns and gaps visible. Facets divide content by distinct attributes, such as document type, activity, and location, so the same item can be found from different angles. Digital metadata makes this practical without storing multiple copies. The right structure depends on what users need to see and do.
This is a useful corrective to the instinct to make every knowledge system resemble a folder tree. In data design, the choice of representation determines which relationships are easy to see and which become awkward exceptions. A matrix can invite comparison, while a tree implies a single path through a domain. Still, flexibility has a cost: multiple views require reliable metadata and shared definitions. The best design is not the one with the most expressive structure, but the simplest representation that supports the decisions users actually make.
Use shared categories to connect teams without pretending they think alike
A bridge, not a merger. A boundary object is a shared reference point that different groups can use for their own purposes while still coordinating with one another. A faceted taxonomy can play this role: a shared document category anchors the system, while other facets preserve distinct departmental perspectives.
When information fails to travel. The book describes the case of Victoria Climbié, whose abuse was encountered by multiple agencies, hospitals, and social services teams. Different groups interpreted her situation through different categories, while fragmented records and unclear responsibility prevented them from forming a coherent picture. In another public failure, an Australian immigration department held eleven variations of Vivian Alvarez's name across databases. A thesaurus linking those forms could have helped connect records, though no classification tool could substitute for accountable action.
The cases show why interoperability is more than making databases technically compatible. People must also recognise that two records, labels, or accounts refer to the same underlying situation. Boundary objects help because they preserve differences instead of demanding that every profession adopt one worldview. The sobering limit is that shared categories are not a cure for institutional neglect. They can make connections easier to see, but authority, incentives, and responsibility determine whether anyone responds. Information architecture supports coordination; it does not create moral commitment by itself.
Use categories to expose waste, customer mismatch, and strategic gaps
Make decisions visible. Taxonomies can organise products, processes, customers, and resources, not just documents. Unilever reduced a portfolio of 1,600 brands to 400 that generated 90 percent of revenue. The simplification was intended to reduce complexity and save money, including through fewer procurement specifications. It was also a practical act of categorisation: deciding which brands merited investment.
Classify from the customer's side. Club Med found that more than 40 percent of customer dissatisfaction was linked to people choosing or being directed to resorts that did not fit their preferences. It developed customer archetypes, such as adventure-oriented or family-oriented travellers, to improve matching. The point is not that every customer fits a neat box. Categories are useful when they reveal actionable patterns and help an organisation respond more accurately.
These examples make classification feel less like clerical housekeeping and more like a management instrument. A portfolio map can expose where investment is concentrated, while customer archetypes can reveal mismatches hidden by broad demographic averages. This resembles decision science: a model is valuable when it improves a choice, not merely when it describes a population. The risk is reification, treating a useful profile as a permanent identity. Customer categories should remain hypotheses to test against behaviour and feedback, especially when the people being classified can change their preferences.
Match taxonomy rigidity to the environment, and keep complex maps provisional
Read the situation first. The Cynefin framework distinguishes environments where rules work, problems require specialist analysis, outcomes emerge unpredictably, or conditions are so chaotic that immediate action comes first. In stable settings, established taxonomies can organise routine work. In complex settings, categories should remain provisional and be revised as patterns emerge.
Beware the familiar label. Early in the SARS outbreak, the disease was treated in some places as ordinary viral pneumonia, although it did not behave as expected. The book describes Singapore's response as a shift from relying on standard assumptions to trying multiple interventions, monitoring results, and strengthening measures that appeared useful. The lesson is not to abandon categories, but to avoid mistaking an initial label for a reliable explanation when evidence contradicts it.
The framework offers a disciplined alternative to two common errors: demanding certainty before acting, and applying a familiar procedure to a genuinely unfamiliar problem. In complex systems, small experiments can reveal patterns that prediction misses. Yet experimentation also needs guardrails, particularly in public health or safety-critical work, where a failed probe can harm people. The practical question is therefore not simply whether a situation is complex, but which actions are reversible, how quickly their effects can be observed, and who bears the cost of being wrong.
Build from observed work, then negotiate the language people actually use
Start with the work. A taxonomy project should begin by clarifying its purpose, identifying affected groups, and mapping their activities and information needs. Interviews and concept maps help uncover possible stakeholders, but observation matters because official procedures do not always match daily practice. Process maps should capture how work actually happens, including workarounds and handoffs.
Find the hidden categories. In one organisation, managers believed customer calls followed a simple referral procedure. Interviews revealed that staff improvised when the manager was away, sometimes giving answers from outdated brochures or sending callers to the wrong person. Those variations exposed missing guidance and useful categories for routing enquiries. The project should test its early assumptions with stakeholders, revise its scope, and explain what it is trying to improve before settling on a design.
The insistence on observing real work echoes ethnographic research and human-centred design: people often cannot report routines they perform automatically, and official accounts tend to describe how systems are supposed to work. Mapping together also turns disagreement into usable evidence rather than a nuisance to suppress. The challenge is representation. A few interviews or a workshop can make one team's workaround look universal. Sampling across roles, locations, and levels of experience helps distinguish a recurring pattern from a memorable exception.
Design for human limits: short lists, shallow trees, and multiple facets
Keep choices manageable. The book draws on George Miller's influential discussion of limits in attention and short-term memory. Although the exact “seven plus or minus two” figure has been challenged, the practical point remains: long lists demand more effort, and familiarity changes what users can manage. The author recommends treating lists above roughly 15 items as candidates for subdivision and keeping ordinary trees shallow, ideally no more than three levels.
Let facets carry scale. A single deep tree becomes difficult to navigate as a collection grows. Several simple facets, each describing one attribute, let users combine perspectives without forcing every item into one long path. These are design heuristics, not laws. Experts may navigate deep structures they know well, while novices need plain labels, clear scope notes, and tested routes to common tasks.
The strongest point is not a magic number, but the connection between cognitive load and design choices. Each extra menu or ambiguous label consumes attention that could otherwise go to the task. Modern interface research likewise treats familiarity, context, and decision complexity as central to usability. Numerical limits should be handled cautiously, however: expertise, language, screen design, and consequences of error all matter. A taxonomy should be tested with actual users and content, not approved because its branches fit a memorised rule.
Make information useful at the moment of work, not merely searchable
From warehouse to workbench. A taxonomy and its metadata can help users locate content, but finding a document is not the same as using it. The book uses “taskonomy” for arranging resources around activities rather than dictionary-like subject categories. A blacksmith places tools where the job requires them, not merely alongside similar tools.
Build useful neighbourhoods. An information neighbourhood surrounds the main item a user needs with related resources, such as relevant expertise, history, locations, or next steps. A taxonomy supplies the underlying structure; metadata and a thesaurus help connect the pieces. This approach asks what the user is trying to accomplish, then places pertinent information and supporting material together. It also reduces the distance between a large repository and the working context where its contents can make a difference.
This distinction anticipates a persistent problem in digital workplaces: retrieval systems optimise for locating objects, while employees need support for completing tasks. Contextual arrangement can reduce the mental effort of stitching together scattered documents, tools, and contacts. It also changes the design question from “What category is this?” to “What else does someone need at this point in the work?” The caution is that task-based displays can become cluttered or overly tailored. Useful neighbourhoods need evidence about real tasks and a clear sense of what belongs nearby.
Pair controlled taxonomies with social tags and machine-readable ontologies
Different tools, different strengths. A folksonomy is a collection of user-chosen tags, made socially visible so people can discover items and the people who described them. Tags are quick to apply and reflect current language, but meanings can be inconsistent. An ontology is a formal model of concepts and their relationships, designed to help computer systems connect and analyse information across collections. It can express more precise relationships than a conventional thesaurus, but is costly to build and difficult for people to navigate.
Let the methods coexist. Formal taxonomies support predictable browsing and shared vocabulary. Social tagging can surface emerging terms and serendipitous connections. Ontologies can support machine processing across systems. None solves every problem. The balance depends on the size and diversity of the content, the users, and the need for precision, discovery, or coordination.
The book's portfolio view is more useful than the familiar contest between top-down control and bottom-up participation. Each approach allocates effort differently: experts design formal categories, users contribute tags, and technical teams encode relationships for machines. The resulting system can be more adaptive than any single method. But combination is not automatically harmony. Tags can reproduce bias, ontologies can encode narrow assumptions with technical authority, and formal terms can exclude everyday language. Governance must make disagreement visible and provide ways to revise the models as use changes.
Treat launch as the beginning of taxonomy governance, not project completion
Turn the design into practice. A taxonomy needs an implementation plan, a metadata framework, and a thesaurus that connects alternative user terms to preferred labels. Because tagging content takes time, systems should reduce repeated decisions where possible and show users a tangible benefit. Piloting with representative people and documents can reveal confusing categories before full rollout.
Keep checking fit. The book recommends a systematic review within six months of implementation and annually afterward. Reviews can examine search logs, user feedback, how evenly content is distributed, and whether new business needs require changes. A governance group should represent relevant stakeholders and have enough authority to resolve conflicts. Without ongoing maintenance, categories drift away from the work they were built to support.
This is a governance problem disguised as a design problem. Categories acquire value only when people apply them consistently enough to support retrieval and coordination, and that requires ownership, training, and revision mechanisms. The analogy to software maintenance is apt: a successful launch does not freeze the environment around the system. A useful review should also ask who is missing from the taxonomy and whose work remains invisible. Usage statistics can show where the system is strained, but they cannot by themselves explain why.
Analysis
The book's central contribution is to treat classification as organisational infrastructure rather than a technical prelude to search. That move connects information design to coordination, power, memory, risk, and everyday work. Categories make some relationships actionable: a shared label can connect a report to several teams, or a facet can let users approach the same document from different needs. But classification is never merely descriptive. The decision to include, exclude, merge, or privilege a category distributes attention and can shape which work is recorded and whose expertise is legible.
Lambe's strongest practical insight is that the taxonomy product cannot be separated from the social process that produces and maintains it. Observation, negotiation, testing, and governance are not project overhead. They are how a vocabulary becomes credible to its users and how the system keeps pace with changing practice. This also explains why technical sophistication alone underperforms: metadata and search cannot compensate for categories that do not match real work or for institutions that do not act on what they find.
The book's argument is especially valuable because it resists a false choice between order and openness. Trees, facets, controlled vocabularies, social tagging, and ontologies solve different problems and impose different costs. A healthy information environment can combine them. The unresolved difficulty is accountability for the design decisions. A flexible scheme may preserve multiple perspectives, but someone still decides which perspectives receive prominence, which terms become standard, and who can change them. Those are governance questions, not just usability questions.
In contemporary digital systems, the stakes are higher because classification is often automated or embedded in platforms that users cannot inspect. The book's principles therefore invite an audit trail: document why categories exist, test their effects across user groups, and make correction possible. A taxonomy should be judged not by elegance or completeness, but by whether it helps people coordinate without erasing the differences that make new knowledge possible.
Review Summary
Organising Knowledge receives mixed reviews, with an average rating of 3.95 out of 5. Readers appreciate the book's engaging examples and its ability to make taxonomies relevant. Some find it comprehensive and exciting, while others consider it dense and challenging to finish. The book is praised for its introduction to taxonomy thinking but criticized for lacking practical details. Some readers find parts of the content unrelated to the main concept, making it slow for non-taxonomists. Overall, it's viewed as a useful resource for website and intranet taxonomies.
Glossary
Taxonomy work
Ongoing category designThe activities through which an organisation discovers, describes, negotiates, represents, and maintains categories for its knowledge and information. It includes more than producing a finished classification: mapping vocabularies, agreeing shared meanings, testing the design, and revising it as work and knowledge change are all part of the process.
Faceted taxonomy
Multiple independent classification viewsA classification system built from separate dimensions, or facets, such as document type, activity, place, and time. Each facet describes one attribute, and users can combine them to locate content from different perspectives. In digital systems, metadata can associate several facet values with the same item without requiring duplicate copies.
Boundary object
Shared cross-group referenceAn object or structure that different communities can interpret in their own ways while still using it as a common point of reference. A taxonomy can serve as a boundary object when it offers shared categories for coordination but preserves the different perspectives of the groups that use it.
Common ground
Shared language and expectationsThe shared language, categories, routines, stories, history, and sense of purpose that help a group coordinate without explaining every assumption from scratch. A taxonomy can strengthen common ground by making terms and organising principles explicit, but trust, socialisation, and shared purpose also contribute.
Cynefin framework
Context-sensitive decision guideA framework for distinguishing environments that are clear and rule-based, complicated and requiring expertise, complex and shaped by emerging patterns, or chaotic and requiring immediate action. In the book, it helps determine whether a taxonomy should be stable and structured, or provisional and experimental.
Knowledge lens framework
Four domains of knowledge workA framework that distinguishes information, expertise and learning, collaboration, and organisational culture as interrelated areas of knowledge management. It uses the terms Logos, Sophos, Pathos, and Ethos. The framework helps show that taxonomy work can support more than document retrieval, including skill development, coordination, and shared identity.
Taskonomy
Work-centred resource arrangementAn arrangement of information and tools around the activities people perform, rather than around abstract categories alone. It reflects the context of use, placing resources together when they support the same task. The book contrasts this workface arrangement with a taxonomy that structures a larger collection for general retrieval.
Information neighbourhoods
Contextual sets of useful resourcesArrangements that place a user's primary information need alongside related materials and possible next actions. A neighbourhood may connect a main item to relevant expertise, people, places, history, or related topics. Taxonomies and metadata help locate the resources, while knowledge of users' tasks guides what should appear together.
Folksonomy
Socially shared user taggingA collection of content described with tags chosen by its users rather than assigned from a centrally controlled vocabulary. When the tags, tagged objects, and people applying the tags are visible together, users can discover related content and people with similar interests. Folksonomies are easy to contribute to but often ambiguous.
Ontology
Formal model of concept relationsA machine-oriented model that specifies concepts and the relationships among them, potentially using many relationship types rather than only broader, narrower, or related terms. Ontologies can help computer systems combine and analyse information across collections, but their complexity makes them less suited to direct human browsing.
ASHEN
Map of knowledge asset typesA framework for identifying knowledge in more forms than documents. Its letters refer to Artefacts, Skills, Heuristics, Experience, and Natural talent. The book extends the framework with relationships, recognising that access to another person's knowledge can itself be an important organisational resource.
The magical number 7 ± 2
Heuristic for cognitive capacityGeorge Miller's 1956 discussion of research suggesting that attention, immediate memory, and discrimination often work best with a limited number of items. The book treats the figure as a design prompt, not a fixed law: exact limits vary, and familiarity, chunking, and multiple facets can change how much users can manage.
Download PDF
Download EPUB
.epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.