← Back to list
Project & Org Management
#커네빈#복잡계#의사결정#상황적 리더십#애자일
Last updated · 2026-11-16

The Cynefin Framework — Complexity-Based Decision Making

1. Overview

A. Definition

Cynefin is a sense-making framework proposed by Dave Snowden while at IBM (1999). It divides the context in which a decision is made into five domains — Clear, Complicated, Complex, Chaotic, and Disorder — and holds that because the nature of cause-and-effect (causality) differs in each domain, different modes of decision making must be applied. 'Cynefin' is Welsh for "the place of our belonging / habitat," implying that our judgment is rooted in countless experiences and contexts.

The core insight of Cynefin is that "applying a single best practice to every situation is dangerous." Many managers try to repeatedly apply the one approach they know well (e.g., detailed planning, expert analysis) to every problem, but when the causal structure of a situation differs, the same prescription produces the opposite result. In the Complex domain, for instance, what is needed is emergence through small experiments rather than up-front analysis; treating it like the Complicated domain with massive analysis and design only burns resources without yielding an answer. Cynefin helps you first diagnose "what kind of problem am I facing right now," so you can choose a mode of action that fits the context.

B. Background and Necessity

Cynefin was born within the currents of Knowledge Management and Complexity Science. In the 1990s, knowledge management tried to document and standardize all knowledge, but the tacit knowledge and context-dependent judgment of the field could not be handled that way. Snowden recognized that human organizations should be seen not as machines but as Complex Adaptive Systems (CAS), and that the very nature of causality differs between the ordered and un-ordered worlds. It later became widely known through the 2003 paper with Cynthia Kurtz and the 2007 Harvard Business Review article "A Leader's Framework for Decision Making," establishing itself as a theoretical foundation for context-appropriate leadership in Agile, DevOps, crisis management, and policy design.

From the perspective of an information management professional engineer, Cynefin matters because an IT undertaking is not a problem of a single nature. Legacy migration (Complicated), the market response to a new digital service (Complex), large-scale incident response (Chaotic), and routine operations (Clear) are all mixed within one project. The PM and architect must therefore selectively allocate predictive and adaptive (Agile) approaches per work area, and Cynefin provides the rationale for that allocation.

C. Characteristics

Cynefin is distinguished from other classification models in three ways. First, it divides domains by the nature of causality, not by coordinates, so the same problem can change domain depending on how it is handled. Second, it treats the boundaries and movement (dynamics) between domains as first-class elements, making transitions such as ordering and collapse objects of design. Third, it presumes a participatory, contextualized method in which field members read the situation together, rather than an expert classifying from above. These three traits make Cynefin a tool for designing decisions and leadership beyond a static diagnostic table.

2. Structure of the Five Domains

flowchart TB
  subgraph ORDERED["Ordered — causality repeats and is predictable"]
    CLEAR["Clear<br/>Known Knowns<br/>Sense -> Categorize -> Respond"]
    COMP["Complicated<br/>Known Unknowns<br/>Sense -> Analyze -> Respond"]
  end
  subgraph UNORDERED["Un-ordered — causality appears only in hindsight"]
    CPLX["Complex<br/>Unknown Unknowns<br/>Probe -> Sense -> Respond"]
    CHAOS["Chaotic<br/>Unknowables<br/>Act -> Sense -> Respond"]
  end
  DIS["Disorder<br/>unsure which domain"]
  CLEAR -. "cliff of complacency (ignored complexity)" .-> CHAOS
  CHAOS --> CPLX --> COMP --> CLEAR
  DIS -. "decompose and place into domains" .-> ORDERED
  DIS -. .-> UNORDERED
  style CPLX fill:#e8f0fe,stroke:#2f6fed,stroke-width:2px
  style CHAOS fill:#fde8e8,stroke:#e11d48,stroke-width:2px

Cynefin groups the five domains into the Ordered world, the Un-ordered world, and the central Disorder. The ordered world is where causality repeats stably so it can be analyzed and designed in advance, while the un-ordered world is where causality shifts constantly and cannot be specified beforehand. This distinction matters because the tools of the ordered world (detailed planning, standards, expert analysis) and those of the un-ordered world (experiment, emergence, rapid stabilization) are fundamentally different.

The epistemological names of the domains (Known/Unknown) split on the question "do we even know what we do not know?" In the ordered world the unknown is a knowable kind of unknown (resolvable by analysis and learning), whereas in the un-ordered world a kind of unknown where we do not even know what we do not know prevails. This epistemological difference explains why up-front planning is powerless in the un-ordered world. Notably, Snowden renamed the early "Obvious/Simple" to Clear because that label bred overconfidence — the attitude of "it is simple, so we are safe" is the first step toward the cliff of complacency.

A. Clear — Known Knowns. A domain where cause and effect are self-evident to everyone and repetitive. Since the answer converges to one, Best Practice and standard operating procedures (SOP) work. Examples: credit-card approval processing, call-center first-response scripts, server-restart checklists. The trap of this domain, however, is oversimplification and complacency — getting stuck in "the way we always do it" blinds you to environmental change.

B. Complicated — Known Unknowns. A domain where cause-and-effect relationships exist but become visible only through expert analysis. There can be several right answers (multiple "good ways"), and experts can make the correct choice. Examples: DB performance tuning, network capacity design, tax and legal advice. This domain is governed by Good Practice and engineering methodologies, but one must beware the "expert trap" of specialists locked into their own viewpoint and excluding new possibilities.

C. Complex — Unknown Unknowns. A domain where causality is explained only in retrospect; systems interact and outcomes emerge. Because up-front analysis cannot reveal the answer, one must discover patterns through Safe-to-fail experiments and then amplify or dampen them. Examples: user adoption of a new digital service, organizational-culture change, platform-ecosystem formation. Here the key is not a fixed answer but Emergent Practice. The right attitude toward the complex domain is not "perfect design, then execution" but "throw many small hypotheses at once, grow the ones that respond well, and quickly fold the bad ones." For example, when introducing a public-service complaint chatbot, rather than designing the whole scenario at once, piloting it on a few complaint types, learning the patterns that surface in the actual query logs, and then expanding scope is the typical complex-domain approach.

D. Chaotic — causality cannot be identified. A crisis domain where there is no time to find causality and immediate stabilization takes priority. Do not analyze — act first to stop the bleeding and create a thread of order. Examples: large-scale security breach, total service outage, disaster response. Here unprecedented Novel Practice is required. In the chaotic domain, even if the leader monopolizes the decision, they must create a foothold of order through decisive moves like halt, isolate, failover to backup, and switch to the complex domain for collaborative investigation the moment things stabilize. Lingering too long in chaos and becoming intoxicated with "strong command" creates another risk — the organization grows dependent on that mode of control.

E. Disorder. A state in which you do not know which domain you are in. People then quarrel to interpret the problem each in the way most familiar to them, so the first priority is to decompose the situation and place each part into the correct domain.

3. Decision Process by Domain

flowchart LR
  subgraph S1["Clear / Complicated (Ordered)"]
    A1["Sense"] --> A2["Categorize/Analyze<br/>Categorize·Analyze"] --> A3["Respond"]
  end
  subgraph S2["Complex (Un-ordered)"]
    B1["Probe<br/>safe-to-fail experiment"] --> B2["Sense"] --> B3["Respond<br/>amplify·dampen"]
    B3 -. repeat .-> B1
  end
  subgraph S3["Chaotic (Un-ordered)"]
    C1["Act<br/>stabilize now"] --> C2["Sense"] --> C3["Respond"]
  end

The prescription of each domain is summarized by a difference in the order of action. The ordered world (Clear, Complicated) reads the situation first (Sense), then categorizes or analyzes before responding; the complex world probes with a small poke (Probe) and then senses the reaction to respond; and the chaotic world acts first (Act) to create order, then senses and responds. The essence of Cynefin's prescription is that "which comes first, analysis or action" is inverted depending on the domain.

The complex world's safe-to-fail experiment is designed to be "small, parallel, and reversible so that even a failure does not bring down the whole system," which is the exact opposite of the ordered world's "fail-safe" design that bets everything on one big wager. For example, a public portal's new UX is first exposed to a small set of users via A/B testing and canary deployment — gathering reaction data and amplifying the patterns that show positive signals — instead of a big-bang full redesign. The case of a global commerce company in 2012 whose sales plunged after a full redesign and which then returned to an incremental experimentation regime is often cited as a typical failure of treating the complex like the ordered.

Domain Causality Decision Practice knowledge Leadership IT example
Clear self-evident·repetitive Sense->Categorize->Respond Best Practice direct·standardize SOP operations, batch jobs
Complicated visible on analysis Sense->Analyze->Respond Good Practice leverage experts performance tuning, capacity design
Complex explained only after Probe->Sense->Respond Emergent experiment·facilitate new services, org change
Chaotic cannot be identified Act->Sense->Respond Novel decisive command major outage·breach response
Disorder unknown decompose·re-place — clarify the situation complex issues of unknown cause

One must also remember that the mode of a leader's failure differs by domain. The leader in the Clear domain is prone to overcontrol and complacency; in the Complicated domain, to "analysis paralysis" and expert bias; in the Complex domain, to killing emergence by hastily forcing an answer. The leader in the Chaotic domain, conversely, insists on analysis when they should act and misses the golden time. Cynefin therefore makes you ask not only "which domain am I in now" but also "what trap am I prone to fall into in that domain." The flexibility to switch leadership style to fit the domain — directing -> leveraging experts -> facilitating -> commanding — becomes a core competency.

4. Inter-Domain Dynamics and Boundaries

If you use Cynefin only as a static classification table, you miss half its value. What Snowden emphasizes is movement (dynamics) between domains. First, the flow from Chaotic -> Complex -> Complicated -> Clear is the desirable ordering path. After stabilizing a crisis (Chaotic), you find patterns through experiment (Complex), organize those patterns into a form experts can analyze (Complicated), and finally solidify them into standard procedure (Clear). The process by which an organization absorbs a new technology and settles it into an operational standard follows this path exactly.

Second, the most dangerous is the "Complacent Cliff" from Clear to Chaotic. If work classified as Clear is in fact no longer simple due to environmental change, yet you cling to "the way we always do it," hidden complexity accumulates and at some point plunges into chaos as if falling off a cliff. Incidents in which an old legacy system is left as "it just runs" until it meets a total outage, or a pipeline deployed repeatedly without verification escalates into a major incident, are examples. Cynefin demands that in the Clear domain you continuously check "is this assumption still valid?"

Third, Disorder sits at the center of all boundaries. In an event where many characters are mixed, like a composite outage, people try to drag the problem only through the lens of their own specialty, but the leader's role is not to rashly force a single approach — it is to decompose the event into parts and place each piece into the correct domain. A payment outage, for instance, can be split into "infrastructure restart (Clear) + transaction analysis (Complicated) + experiment on user-churn response (Complex) + emergency blocking of some functions (Chaotic)."

5. Applying Cynefin — the Sense-Making Workshop

Cynefin shows its power when used not as a tool for an individual to classify in their head, but as a participatory method in which a group reads the situation together (sense-making). Snowden recommends a "contextualisation" approach that does not hand out abstract definitions first but lays out concrete items first and lets people place the domains themselves. In practice it is run in the following four steps.

A. Collect items. Write down the tasks, risks, and decision items handled in the project and operations as small units at the level of sticky notes, concretely. The key here is to break them into actionable units — not an abstract slogan like "strengthen security" but "verify user adoption of the new simple-pay integration," "nightly settlement batch processing," "apply the legacy WAS patch." The more abstract they are, the more everyone drags them into the domain convenient for them, and disorder grows.

B. Place into domains. Place each item onto the five domains. Reach agreement by using questions such as "is causality self-evident (Clear)? visible on analysis (Complicated)? knowable only by doing (Complex)? must it be stopped immediately (Chaotic)?" Items on which opinions diverge are put in the Disorder zone, revealing why people see them differently — this disagreement itself is the most valuable information.

C. Choose prescriptions and allocate resources. Once placement is done, apply the per-domain prescription. Assign SOP/automation to Clear, experts/design review to Complicated, safe-to-fail experiment budget and feedback loops to Complex, and emergency authority/runbooks to Chaotic. The actual deliverable of a Cynefin workshop is that governance and budget are differentiated to fit the domain at this step.

D. Design transitions and re-check. Finally, design the movement paths between domains. Set a plan to order from Complex to Complicated and then to Clear once a pattern is confirmed in a complex experiment, and set a re-check cycle for the "complacent cliff" on Clear items. Because placement changes when the environment changes, the workshop is repeated on a quarterly/release cadence. There is a reported case of a financial-IT organization that ran this workshop semi-annually to reclassify its task portfolio and assigned Agile squads only to tasks judged complex, lowering its transition-failure rate.

6. Comparison with Similar Models

Cynefin is often compared with the Stacey Matrix. Stacey divides into simple, complicated, complex, chaotic, and anarchy along two axes — "degree of agreement on requirements (certainty)" and "technical certainty." Both models share the message that "the greater the complexity, the more adaptation over planning," but the difference lies in their philosophical basis. Whereas Stacey deterministically positions a domain by the coordinates of two variables, Cynefin sees a domain not as a fixed coordinate but as the nature of causality and the movement of boundaries. That is, Cynefin places more emphasis on the dynamism that "the same problem changes domain depending on how it is handled." In practice, then, Stacey is better for quick intuitive classification, and Cynefin for transition/boundary management and leadership design.

The criticisms and limits of Cynefin also deserve a balanced view. First, domain judgment has no objective measurement criterion and depends on an expert's intuition, so reproducibility across organizations and people is low. Second, there are points where the practical distinction is ambiguous, such as the boundary between the complex and chaotic domains, so a shallow understanding of the theory risks degenerating into hasty labeling like "it's complex, so just go Agile." Third, the framework itself is continually revised (terms and domain names change), carrying a learning cost. Nonetheless, Cynefin is widely used because its value lies not in being a perfect classifier but in being a thinking tool that forces the perspective shift of "do not use one approach for every problem."

Linkage with software methodologies is also important. Waterfall (Predictive) suits the Clear/Complicated domains where requirements and technology are stable, while Agile (Scrum·Kanban) is a "probe-sense-respond" engine optimized for the Complex domain where requirements emerge. Iterative sprints, small increments, and adaptation through retrospectives have the same structure as the complex domain's safe-to-fail experiments. Thus the precise implication of Cynefin is not "Agile is always right" but Agile is right when the task is complex. Conversely, handling a settlement batch with clear rules in Agile only adds overhead. This is why large public SI projects adopt a hybrid that goes predictive for the architecture/regulation areas and Agile for the user-service areas.

Domain Suitable methodology Planning·budget mode Governance focus
Clear waterfall·automation·SOP fixed budget, repeated execution standards·efficiency·anti-complacency
Complicated waterfall·design review·experts estimate-based plan (WBS) architecture·securing expertise
Complex Agile (Scrum·Kanban)·Lean experiment budget, incremental investment safe-to-fail·fast feedback
Chaotic emergency response·runbook emergency reserve·delegated authority rapid stabilization·post-transition

The mapping above holds the principle that "the greater the complexity, do not nail down the budget in advance but release it in stages." The Clear/Complicated domains have stable scope, so a fixed budget and estimate-based plan work, but the complex domain emerges in outcome, so executing a large budget all at once only grows sunk cost. Hence in the complex domain, option-like, incremental investment — executing a small experiment budget divided into many rounds and adding investment only to patterns whose results are confirmed — is rational.

7. In Depth — Recent Trends and Practical Application

Cynefin is not a fixed theory but keeps evolving. In 2020 Snowden refined the framework with the so-called "St David's Day model," explicitly introducing liminal (transitional) areas between domains. The "complex-complicated transition," for example, is a transitional period before a pattern found through experiment has yet been standardized, carrying a warning that pushing this zone prematurely into the ordered world stops emergence. He also split the central disorder into Confused and Aporetic (which admits paradox and inquires into it), re-evaluating "the state of admitting what one does not know and inquiring into it" as a healthy starting point. This revision expanded Cynefin from a classification tool into "a tool for designing flow."

Practical application cases are abundant. First, crisis/disaster management: health authorities in several countries used Cynefin language to define early pandemic response as Chaotic (act now) and then transition to Complex (region-by-region experiment and data-driven adjustment). Second, DevOps/SRE incident response: the Incident Management procedure that handles an outage along the flow "Chaotic (stabilize) -> Complex (hypothesis-based investigation) -> Complicated (root-cause analysis) -> Clear (runbook-ization)" meshes exactly with Cynefin dynamics. Third, digital-transformation portfolio: a company maps its tasks onto the four domains and differentiates investment and governance into "to standardize (Clear), to entrust to experts (Complicated), to experiment (Complex), to block urgently (Chaotic)." In this way Cynefin serves as a meta-framework that threads strategy, operations, and crisis management into one common language.

The implications are large for public/financial IT environments too. E-government and public-platform projects, with their strong regulatory and audit requirements, have traditionally been run predictively, but the acceptance of a citizen-facing digital service is inherently complex. Behind the repeated cases of trying to fix all requirements in advance only to be ignored after launch lies an approach that mistook a complex task for a complicated one. For such projects Cynefin provides the rationale for a divide-and-conquer approach of managing areas separately as "predictive for regulation/architecture, experiment-based for the service experience," and it connects naturally with the recently spreading adoption of Agile and lean-startup methods in the public sector.

The effect is visible in concrete numbers too. One global financial firm mapped about 120 digital tasks onto Cynefin's four domains, then reorganized its portfolio by concentrating cross-functional Agile squads and more than half of the total experiment budget on only the 30% of tasks classified as complex, while handling the remaining Clear/Complicated tasks through a standard pipeline. As a result, the early-abandonment rate of new services that used to fail with "big analysis then big-bang launch" dropped, the automation rate of operational tasks rose, and overall lead time shortened, it is reported. The key is that it did not apply a specific methodology uniformly across the enterprise but differentiated resources and governance to fit the causal nature of the task, showing that Cynefin is a tool that affects real decisions and budgets beyond mere classification.

8. Considerations and Implications (Professional Engineer's View)

  • Application strategy — per-task domain mapping: Do not shove the whole project into one methodology; diagnose the domain per WBS/service unit and mix predictive and adaptive allocations. Hybrid governance that designs architecture/compliance as Complicated and user-value verification as Complex lowers the failure rate.

  • Trade-off — subjectivity of classification and the cost of misclassification: Cynefin has no quantitative criterion, so domain judgment is subjective, and the cost of misclassification is asymmetric. Mistaking the complex for the complicated burns cost through over-design, while seeing the complicated as complex squanders a verifiable problem on needless experiments. Judgment risk must be managed with multi-perspective workshops and post-hoc verification.

  • Outlook — combining with adaptive governance: As environmental change accelerates, Cynefin will develop into the backbone of dynamic governance, combined with OKRs, lean startup, a safe-to-fail experimentation culture, and AIOps-based real-time sense-making. In particular, institutionalizing the "experiment-measure-amplify" loop becomes the task for undertakings whose outcomes emerge, like generative-AI adoption.

  • Related technologies·methodologies: Closely tied to Agile (Scrum·Kanban·[[waterfall-vs-agile]]), CCPM ([[critical-chain-project-management]] Critical Chain), risk response ([[it-project-risk-response]]), DevOps incident management, and complex-adaptive-system theory. On the leadership side, combining with situational leadership ([[tuckman-ladder]]) lets you design context-appropriate leadership that switches among directing, expert, facilitating, and commanding styles by domain.

  • Internalizing it as an organizational capability: Do not end Cynefin as a one-off workshop; it yields lasting effect only when embedded as a common language in existing governance procedures such as Incident management, portfolio review, and architecture decision records (ADR). In particular, unless a "probe-sense-respond" loop and a failure-tolerant budget are institutionally guaranteed for complex tasks, the organization — lacking psychological safety — avoids experiments and over-designs everything as if Complicated.

  • Constant watch on the complacent cliff: The more an operational task is classified as Clear, the more it needs a governance device (periodic reviews, chaos engineering, observability) that periodically re-verifies "is the assumption still valid?", to prevent hidden complexity from plunging into chaos.

References


In one line: Cynefin divides the decision context into Clear, Complicated, Complex, Chaotic, and Disorder, and is a complexity-based sense-making framework that guides you to apply "sense-analyze-respond" versus "probe/act-sense-respond" differently according to the nature of causality.