DataKnobs

Decision frameworks

Five kinds of situation.
Five different first moves.

Most leadership advice assumes every problem is the same shape. The Cynefin framework starts from the opposite premise: that the right response depends on what kind of situation you are actually in — and that the most expensive mistakes come from applying a perfectly good method to the wrong domain.

Cynefin — a Welsh word, roughly “habitat” or the place of your multiple belongings. Developed by Dave Snowden in the late 1990s while at IBM, and refined continuously since.

Opening slide of a leadership presentation introducing the Cynefin framework
Title slide from the source deck. Slides carry their own text, so this one opens full size in a new tab rather than being read at page scale.

The core idea

It’s about constraints, not difficulty

The common misreading of Cynefin treats the domains as a difficulty scale — clear problems are easy, chaotic ones are hard. That reading makes the framework useless, because it tells you nothing you didn’t already know.

The useful reading is structural. Each domain describes how tightly the system is constrained, and constraint is what determines whether cause and effect can be known in advance. A tightly constrained system behaves predictably, so you can look it up. A system with no effective constraints has no stable pattern to look up at all.

This is why the framework's practical output is not a label but a sequence — the order in which you sense, analyse, probe, and act. Change the constraint structure and the correct first move changes with it.

The five domains

What each one is, and what it asks of you

Clear

Fixed constraints Best practice

Cause and effect are obvious to any reasonable observer, and the relationship is stable enough that the right answer can be written down in advance. Processing an expense claim, running a known safety checklist, handling a standard return.

Leadership here is largely about making the right answer easy to find and easy to follow. Codify it, train it, automate what you can, and get out of the way. The domain was originally called simple, then obvious, and is now usually written as clear — the renaming was partly to stop people hearing “simple” as “trivial”.

Failure mode

Complacency. A process that works reliably for years stops being examined, and the conditions it assumed quietly change underneath it. This is the drift that ends at the cliff.

Sequence
  • Sense
  • →
  • Categorise
  • →
  • Respond
Knowledge
The answer is already known and documented.
Authority
The procedure. Expertise is not required to apply it.
Good move
Standardise, automate, and schedule a review of the assumptions.

Complicated

Governing constraints Good practice

There is still a right answer, but it isn’t obvious — it has to be found. Cause and effect are separated by analysis, expertise, or time, yet the relationship holds still while you investigate it. Diagnosing an intermittent fault, structuring a financing deal, tuning a database.

Note the plural: complicated problems usually have several defensible right answers rather than one. That is why the practice type is “good” rather than “best”. Leadership here means convening the right expertise and, critically, listening to more than one expert.

Failure mode

Expert entrainment — specialists see the problem as an instance of what they already know, and reject readings that fall outside their training. Its twin is analysis paralysis, where the search for the best answer outlasts the window in which any answer would have helped.

Sequence
  • Sense
  • →
  • Analyse
  • →
  • Respond
Knowledge
Knowable, but only through investigation.
Authority
Expertise — ideally more than one school of it.
Good move
Get diverse experts in the room and set a decision deadline.

Complex

Enabling constraints Emergent practice

Here the right answer cannot be known in advance, however much expertise you assemble — not because the problem is harder, but because the system responds to your intervention. Cause and effect are only coherent looking backwards. Culture change, a new market, an unfamiliar organisational conflict.

The move that works is to run several small, safe-to-fail experiments in parallel, each with a way of telling whether it is working, then amplify what succeeds and dampen what doesn’t. The word that matters is parallel: a single pilot is a bet, not an experiment, because it gives you nothing to compare against.

“Safe-to-fail” is a specific bar and worth stating plainly — a failed experiment must be survivable and reversible, and you must be able to see the failure early enough to stop.

Failure mode

Reaching for analysis or best practice. Both assume a stable answer exists, so both produce confident plans that the system immediately invalidates — and the confidence makes the failure slower to detect.

Sequence
  • Probe
  • →
  • Sense
  • →
  • Respond
Knowledge
Only available in retrospect.
Authority
Evidence from your own experiments.
Good move
Several parallel safe-to-fail probes, each with a stop condition.

Chaotic

No effective constraints Novel practice

No stable pattern exists to observe, and waiting is itself a costly decision. The immediate job is not to find the right answer but to establish enough order that a pattern can form — to get the situation out of chaos and into a domain where the other methods apply.

Act first, then look at what your action changed, then respond to that. This is the one domain where decisive top-down direction is the correct instinct rather than a bad habit, which is exactly why leaders who are good at chaos are dangerous elsewhere: the move that makes them effective here is the move that destroys a complex situation.

Failure mode

Staying too long. Chaos is a transitional state, and the clarity of command is seductive. A related pathology is manufacturing crisis to justify authority that would not otherwise be granted.

Sequence
  • Act
  • →
  • Sense
  • →
  • Respond
Knowledge
None available in the moment.
Authority
Whoever can act now.
Good move
Stabilise, then deliberately exit into complex or complicated.

Disorder

Constraint unknown Also called confusion

The central domain: not knowing which of the other four you are in. It is where most contested decisions actually start, and it is not drawn in the middle by accident.

Its signature danger is that people in disorder fall back on the domain they are personally most comfortable with. The same situation gets read as complicated by the analyst, chaotic by the firefighter, clear by the administrator — and each of them is confident. When a leadership team is arguing about what to do and getting nowhere, they are often really disagreeing about what kind of situation this is, without ever surfacing that.

The way out is to break the situation into parts and place each part separately, because most real problems are not homogeneous. A product launch might have a clear compliance component, a complicated pricing component, and a complex adoption component — and treating those three the same way is what makes launches fail in surprising ways.

Failure mode

Never noticing you were in it. Disorder is invisible from the inside precisely because everyone has already resolved it — differently.

Sequence
  • Decompose
  • →
  • Place each part
Knowledge
Unclear which kind is even available.
Authority
Contested — that’s the diagnostic signal.
Good move
Ask the team to name the domain before debating the response.

The asymmetric boundary

The cliff between clear and chaotic

Cynefin’s boundaries do not all behave the same way. The line between complicated and complex is a gradient you can wander across and back. The line between clear and chaotic is drawn as a cliff or a fold, because crossing it is sudden and asymmetric.

The mechanism is complacency. A process in the clear domain works so dependably that it stops being questioned, while the conditions it quietly assumed keep changing. The gap between the procedure and reality widens invisibly, and then something loads the system and it doesn’t degrade gracefully — it collapses straight past complicated and complex into chaos.

Climbing back out is far more expensive than the drift down was. That asymmetry is the whole reason the boundary is drawn as a cliff rather than a line, and it is the strongest practical argument for periodically re-examining the processes that are working best.

Signature tool

What’s your first move?

Answer three questions about the situation in front of you. The panel narrows as you go — domains stay lit while they remain consistent with your answers, and the leading one shows the move it would have you make first.

1. If you asked five capable people for the answer
2. If you intervene, does the situation change shape?
3. What happens if you wait a week?
  • ClearSense first
  • ComplicatedSense first
  • ComplexProbe first
  • ChaoticAct first

All four are still consistent with what you’ve told me. Answer a question to narrow it.

Treat the result as a prompt for discussion rather than a verdict. If your team can’t agree on the answers, that disagreement is the finding — you were in disorder, and now you know it.

Getting it wrong

Four ways the framework gets misused

Drawing it as a 2×2

A matrix implies two independent axes and four equivalent cells. Cynefin has five domains, unequal boundaries, and one edge that behaves as a cliff. The grid version loses all of that.

Using it to categorise

It’s a sense-making framework, meaning the data precedes the frame. Sorting problems into pre-set boxes inverts it and turns a diagnostic into a filing system.

Assigning a domain permanently

Situations move, and often the whole point of your intervention is to move them. Chaotic is a state to exit; complex work that succeeds tends to become complicated, then clear.

Placing a whole project at once

Most real initiatives contain components in several domains simultaneously. Placing the project as one lump guarantees that some part of it gets the wrong treatment.

Questions

Common questions

Is Cynefin a 2×2 matrix?

No. It has five domains rather than four, the domains are not equal in shape, and the boundaries differ in character — the clear-to-chaotic edge is a cliff while the complicated-to-complex edge is gradual. A 2×2 implies two independent axes and four interchangeable cells, none of which applies here.

What’s the difference between complicated and complex?

A complicated situation has a knowable right answer that expertise can find; cause and effect are separated by analysis but the relationship holds still while you study it. A complex situation has no knowable right answer in advance, because the system changes in response to your intervention and cause and effect are only visible in hindsight. Complicated problems are solved by analysis. Complex ones are addressed by parallel safe-to-fail experiments.

What is the cliff?

The boundary between clear and chaotic, drawn as a fold rather than a line because crossing it is asymmetric. Complacency in a reliably working process lets assumptions age unnoticed, and the collapse when it comes skips the intermediate domains entirely. Recovery costs far more than the drift did.

What is the disorder domain?

The central domain — the state of not knowing which domain you’re in. Its danger is that people default to whichever domain they’re personally most comfortable with, so a team can argue about the response while actually disagreeing about the situation. Naming the domain explicitly, and decomposing the problem into parts that can be placed separately, is the way out.

Why does the first move change by domain?

Because each domain permits a different amount of knowledge before action. Clear and complicated situations hold still while you look at them, so you can sense first. Complex systems only reveal their behaviour in response to intervention, so you probe first. Chaotic situations offer no stable pattern to observe at all, so you act first and learn from what your action changes.