Cynefin vs Stacey Matrix: which complexity framework to use

The Cynefin Framework and the Stacey Matrix are the two models managers reach for when a project feels harder than its plan admits — and the two most often confused, because both draw a line between complicated and complex. The distinction is one of job. The Stacey Matrix diagnoses why a situation is uncertain: are you missing knowledge, or missing agreement? Cynefin prescribes what to do about it: follow a runbook, call in experts, run experiments, or stabilize first. Stacey reads the cause; Cynefin sets the response.

At a glance

What the Stacey Matrix is best for

Stacey earns its place when the useful question is what kind of uncertainty are we dealing with — because its two axes split "this is hard" into two very different causes:

• Methodology selection. Stacey maps cleanly onto delivery approaches: close to certainty and agreement → plan-driven (waterfall); far from both → iterative (agile); far on one axis only → resolve that axis first. It's the framework most often drawn on a whiteboard to justify "why we're running this one agile."
• Surfacing politics, not just unknowns. The agreement axis is Stacey's unique contribution. A project can be perfectly understood technically and still sit in the danger zone because stakeholders disagree on the goal. Stacey forces that onto the page; most problem-only models bury it.
• Diagnosing where the work is. Far from certainty means the bottleneck is knowledge — you need research, experts, or a spike. Far from agreement means the bottleneck is alignment — you need negotiation, a clearer mandate, or a decision owner. Naming which axis is far tells you which lever to pull.

What Stacey does not do: it stops at the diagnosis. It plots where you are but doesn't prescribe the operating rhythm once you know.

What Cynefin is best for

Cynefin earns its place when the question is what do we actually do now — it pairs each kind of situation with a response:

• Matching response to context. Each domain has a prescribed sequence: sense-categorize-respond (clear), sense-analyze-respond (complicated), probe-sense-respond (complex), act-sense-respond (chaotic). That last one — act first, then sense — is why Cynefin handles crises that Stacey's static map doesn't speak to. See the full breakdown in the Cynefin Framework for PMP guide.
• Beyond software. Cynefin covers incident response, policy design, and live crises, not just project delivery. When a system is on fire, "probe" is the wrong first move; Cynefin tells you to stabilize, then analyze — the same instinct a premortem builds in advance.
• Catching the misclassification trap. Cynefin's central "confused" domain is its sharpest idea: the most dangerous position is not knowing which domain you're in, because you default to the approach you're most comfortable with — usually treating a complex problem as a merely complicated one.

What Cynefin does not do: it won't tell you why you're uncertain. It names the domain but not whether the fix is knowledge or alignment.

The decision rule

Use the Stacey Matrix to diagnose the source of difficulty — a knowledge gap (low certainty) or a politics gap (low agreement). Use Cynefin to choose the response once you know the domain. When in doubt, run Stacey first, then Cynefin: read, then act.

The cost of skipping the diagnosis is concrete. In 2023, Klarna replaced roughly 700 customer-service agents with an OpenAI-trained assistant that soon handled two-thirds of all support chats. The decision was framed as complicated — a known efficiency problem with a known solution: automate the routine, cut the cost. On the Stacey axes it was nothing of the sort. It sat far from certainty (no one knew how the model would handle emotionally charged disputes, multi-step refunds, or confidently-wrong answers about fees and policy — a compliance risk in fintech) and far from agreement (customers, agents, and regulators were not aligned on AI handling money disputes). That puts it squarely in the complex zone, where Cynefin prescribes probe-sense-respond — safe-to-fail pilots, not a one-shot replacement of 700 people.

By mid-2025 Klarna reversed course and began rehiring humans; CEO Sebastian Siemiatkowski conceded the company "focused too much on efficiency and cost" and that "the result was lower quality, and that's not sustainable." Stacey's agreement axis would have flagged the stakeholder conflict the efficiency framing ignored; Cynefin would have set the cadence — experiment, don't replace. The two frameworks together describe exactly the mistake, which is the strongest argument for running them in sequence.

The Stacey Read, Cynefin Act Sequence

Here is the synthesis this comparison was built to support — a named way to chain the two frameworks so the diagnosis feeds the response instead of duplicating it. We call it the Stacey Read, Cynefin Act Sequence: Stacey reads the two axes to tell you which lever to pull; Cynefin acts by setting the operating rhythm.

The insight the table encodes is the one neither framework gives alone: Stacey splits uncertainty into two causes — a certainty (knowledge) axis and an agreement (politics) axis — that Cynefin collapses into a single "complex" bucket. So Stacey tells you whether to send in experts or a mediator; Cynefin tells you the cadence to run once they're in the room. Klarna's mistake reads cleanly on this table: it acted on the top row (Simple → run the playbook) while sitting on the third (Complex → experiment first).

Edge cases and combined use

• One axis far, not both. This is the most useful Stacey reading and the one teams skip. If you're certain how to do the work but stakeholders disagree on whether to, no amount of expert analysis helps — you have a Complicated-by-disagreement problem, and the fix is a mandate, not a spike. Cynefin alone would miss this, because it has no agreement axis.
• Chaotic vs Complex. Both frameworks warn against confusing these, but only Cynefin gives the rule: in chaos you act first (stabilize), in complexity you probe first (experiment). Running a safe-to-fail experiment during a live outage is the classic error.
• When neither is the right tool. For prioritizing a known backlog, both are too abstract — use RICE. For a specific risky launch, a premortem is sharper than a domain label. For surfacing the macro forces around a complex market rather than inside your project, that's PESTEL vs Porter's Five Forces territory.

In 30 seconds

Why is this hard — knowledge or agreement? → Stacey Matrix. What do I do about it — plan, analyze, experiment, or stabilize? → Cynefin. A genuinely uncertain initiative → run both, Stacey first, using The Stacey Read, Cynefin Act Sequence so the diagnosis sets the response. The framework you can name beats the one you only feel.

Want to run these on your phone? Framework for iPhone & iPad — fill in Cynefin, the Stacey Matrix, or any of 100+ frameworks with AI assistance.

Sources

• Ralph D. Stacey — "Complexity and Creativity in Organizations" (1996)
• Dave Snowden & Mary Boone — "A Leader's Framework for Decision Making," Harvard Business Review (2007)
• Entrepreneur — "Klarna Is Hiring Customer Service Agents After AI Couldn't Cut It on Calls"
• Digital Applied — "Klarna Reverses AI Layoffs: Why Replacing 700 Failed"
• FinTech Weekly — "Klarna Reverses Course on AI Customer Support, Resumes Human Hiring"