Cynefin Framework

Decide how to act by first asking what kind of causal situation you are in. A routine procedure, an expert diagnosis, an uncertain market experiment and an active crisis should not be managed with the same decision logic.

In one minute

Cynefin is a sense-making framework. It distinguishes contexts in which cause and effect are:

• Clear: stable, repeatable and widely understood; sense, categorise and respond with established practice.
• Complicated: discoverable through analysis or expertise, with more than one plausible answer; sense, analyse and respond.
• Complex: shaped by interacting agents so useful patterns emerge through action; run safe-to-learn probes, sense what happens and respond.
• Chaotic: not usable in time for analysis because the situation is unstable; act to contain harm, sense the new state and respond.
• Confused or aporetic: the context is not yet placed, or uncertainty is held deliberately to challenge assumptions; gather, decompose and avoid forcing one answer.

The domains describe a decision context, not a permanent label for a project, company or person. One initiative can contain a clear compliance step, a complicated technical diagnosis, a complex adoption problem and a chaotic incident at the same time.

Best for: deciding which management response fits the causal conditions of a bounded issue.
Avoid when: the group wants a decorative quadrant, uses “complex” to mean merely difficult, or lacks authority to stabilise immediate harm.

The problem it addresses

Teams often apply their preferred response everywhere. Experts analyse a situation whose patterns can only emerge through interaction; experimenters probe a safety-critical routine that already has a validated procedure; crisis leaders keep commanding after stability returns. The mismatch creates delay, false confidence or unnecessary risk.

Cynefin creates a structured conversation about the relationship between evidence, causality and action. It does not calculate the correct domain and it does not prove that a chosen response will work. Classification remains a hypothesis to test against what happens next.

When to use it

Use Cynefin when:

• people disagree about whether a situation is routine, expert, emergent or unstable;
• one programme contains work that needs different decision modes;
• a novel product, policy or AI-enabled service has uncertain behavioural effects;
• leaders need to distinguish experimentation from analysis;
• a crisis response requires a planned transition back to normal governance;
• repeated “best practice” deployment fails in different local contexts.

When not to use it

Do not use Cynefin:

• to rank people as simple or complex;
• to call every strategic issue complex without examining constraints and evidence;
• to avoid expertise where a complicated problem is analysable;
• to run uncontrolled experiments where harm can be irreversible;
• to justify command-and-control after a chaotic condition has stabilised;
• as a substitute for a risk, legal, safety or incident-management process;
• to claim empirical certainty from a workshop classification.

Use Scenario Planning when the decision depends on several plausible external futures. Use Premortem to challenge a concrete plan. Use PDCA/PDSA to structure a bounded change test.

Inputs required

Prepare:

• a bounded decision, event or operating condition;
• observable facts separated from interpretation;
• constraints, irreversibility, time pressure and potential harm;
• differences between subproblems rather than one average description;
• people with operational, technical and affected-user knowledge;
• authority for containment, analysis or safe-to-learn probes;
• a review point for reclassification.

Step-by-step process

1. Bound the situation

Name the decision, affected population, time window and excluded questions. “Our transformation is complex” is unusable; “how to introduce AI-assisted scheduling across five clinics without reducing access” can be examined.

2. Separate facts from the story

Record observed events, known rules, expert disagreements, emerging patterns and urgent harms. Mark what is assumed, inferred or unknown.

3. Decompose the issue

Split the situation where different causal conditions coexist. Authentication may be clear, model integration complicated, customer adoption complex and a current outage chaotic.

4. Place each component provisionally

Ask whether the cause-effect relationship is repeatable, analysable, emergent or unavailable in time. Keep disputed items in Confused rather than resolving them by hierarchy.

5. Test the classification

Look for disconfirming evidence. If experts cannot reproduce the relationship, the issue may not be complicated. If a probe has an irreversible downside, it is not safe-to-learn.

6. Choose the response pattern

• Clear: apply and monitor an established rule; watch for complacency and context drift.
• Complicated: commission analysis, compare expert explanations and make assumptions visible.
• Complex: run several small, coherent probes with amplification and stopping rules.
• Chaotic: contain immediate harm, establish constraints and move deliberately toward a more stable domain.
• Confused: collect evidence, decompose and expose competing interpretations.

7. Define limits and authority

For every action, state who may act, what must not be harmed, how much can be spent, which signals stop the action and who reviews the evidence.

8. Sense the result

Use evidence appropriate to the domain. Compliance may be audited in Clear; models and tests support Complicated; portfolios of probes reveal patterns in Complex; stabilisation indicators matter in Chaotic.

9. Reclassify

Domains can change. A complex practice may become repeatable; a clear process can fall into chaos when assumptions fail. Record the trigger, not only the new label.

AI automation lens

AI can organise authorised observations, compare alternative classifications, detect contradictory evidence and monitor agreed signals.

It must not:

• assign a domain from hidden behavioural profiling;
• convert a probabilistic model output into causal certainty;
• invent safe-to-learn conditions or acceptable harms;
• execute chaotic containment without authorised incident control;
• suppress minority interpretations because a majority selected one domain;
• recommend experimentation on people without consent, safeguards and oversight.

Human owners remain accountable for decomposition, safety, decision rights, classification and the transition between response modes.

Visual model

Text alternative: a bounded situation is decomposed and provisionally placed according to whether cause and effect are repeatable, analysable, emergent, unstable or unresolved. Each condition uses a different response pattern and is reviewed for reclassification.

Interactive example

Scenario

A hospital launches an AI-assisted appointment service. Password resets follow a stable procedure. Model latency has several diagnosable technical causes. Patient trust and channel switching differ unpredictably across groups. A regional outage has stopped urgent bookings.

Your move

Classify the four components and choose the next response for each.

Worked answer

• Password reset is Clear while the procedure remains valid: use the standard workflow and monitor exceptions.
• Latency is Complicated: compare logs, architecture and expert hypotheses.
• Trust and channel behaviour are Complex: use small, safeguarded service probes with affected groups and preserve human access.
• The outage is Chaotic: restore a safe booking route first, then diagnose after containment.

Calling the whole programme complex would hide the urgent containment and the routine work.

Facilitation notes

• Classify components, not personalities or entire organisations.
• Ask participants to place evidence before labels.
• Let minority classifications remain visible until tested.
• Invite safety, accessibility and affected-user perspectives.
• Make the cost of a wrong classification explicit.
• Time-box Chaotic authority and define the return to normal governance.

Expected output

• a bounded and decomposed issue;
• a provisional domain for each component;
• evidence and uncertainty behind each placement;
• a domain-appropriate action pattern;
• constraints, owners and stopping rules;
• observation signals and a reclassification date.

Common mistakes

1. Complex means difficult. A difficult calculation may still be Complicated.
2. One label for everything. Mixed contexts require decomposition.
3. Probe means pilot anything. A Complex probe must be coherent, bounded and safe-to-learn.
4. Chaos becomes permanent command. Containment authority needs an exit condition.
5. Consensus becomes evidence. Workshop agreement does not prove causal structure.
6. Framework as recipe. Cynefin selects a response logic; it does not supply the substantive solution.

Quality checklist

• The situation, decision and time boundary are explicit.
• Mixed components have been separated.
• Observations, assumptions and interpretations are distinguishable.
• Domain placement has disconfirming tests.
• Complex probes have safeguards and stopping rules.
• Chaotic action has an exit and review condition.
• Expert, operational and affected-user knowledge is represented.
• A reclassification trigger and owner are recorded.

Template

Use the structured Cynefin workspace to preserve decomposition, evidence, response logic and reclassification.

Knowledge check

Question: A team can analyse a technical failure through known physics, logs and several expert models. Which domain and response fit best?

A. Clear; apply one universal checklist.
B. Complicated; sense, analyse and respond.
C. Complex; run uncontrolled experiments.
D. Chaotic; act without review.

Answer: B. The relationship is not obvious, but disciplined expertise can analyse it.

Related tools

• Scenario Planning explores choices across plausible external futures.
• Premortem elicits failure mechanisms for a proposed plan.
• PDCA/PDSA Cycle structures a bounded learning test.
• Stakeholder Mapping identifies affected groups and power before probes.
• Bow-Tie Analysis specifies barriers around a defined top event.

References

1. The Cynefin Company. “The Cynefin Framework.” Official overview (opens in a new tab). Accessed 22 September 2026.
2. Snowden, David J., and Mary E. Boone. “A Leader’s Framework for Decision Making.” Harvard Business Review, November 2007. Article (opens in a new tab).
3. Kurtz, Cynthia F., and David J. Snowden. “The New Dynamics of Strategy: Sense-making in a Complex and Complicated World.” IBM Systems Journal 42(3), 2003. DOI (opens in a new tab).
4. Elford, Douglas R. “The Cynefin Framework: A Tool for Analyzing Qualitative Data in Information Science?” Library & Information Science Research 35(4), 2013. DOI (opens in a new tab). Independent critical application; it does not establish universal predictive validity.

Method profile

• Primary output: decomposed context assessment and domain-appropriate action portfolio.
• Decision level: operational, programme or strategic component.
• Evidence strength: established sense-making practice with context-specific applications; domain placement remains a judgement to test.
• Review trigger: new evidence, stabilisation, pattern emergence, procedure failure or change in reversibility and harm.