Complete firstIceberg Model
There are signs of recurring patterns or structures that go beyond individual events.
Without: Without signs of structural patterns, the CLD becomes a causal chain without feedback and is redundant to 5 Whys.
RunsheetSystems Thinking
Causal Loop Diagram
ComplexityMedium
Time1-3 h
Participants2-8
FormatWorkshop
MaturityEstablished
01
Prerequisite
What needs to be finished first
Complete firstVariable listnot in catalog
A first list of relevant variables with a clear direction (more or less) is prepared, so the model starts with substance.
Without: Without variables, the workshop has to do definition work, and diagram quality stays low.
02
Preparation
What needs to be ready before start
Materials
Whiteboard or Miro board; arrows with polarity (+/-) and loop labels (R for reinforcing, B for balancing); notation cheat sheet visible; prepared variable list.
People / roles
One facilitator with a systems-thinking background; 3-6 participants with different system knowledge; one model writer who sets arrows consistently; optional domain expert.
Pre-read
Data points or reports that show the suspected behavior over time; relevant hypotheses about causes; existing quick fixes and their (failed) success.
Time needed
2-4 h
Setup
Notation legend visible: arrow with + means same direction (more A -> more B), arrow with - means opposite direction. R loops reinforce, B loops balance. Rule: every variable is a measurable quantity, not an adjective.
03
Core question
The one question this method answers
Which feedback loops create the observed behavior, and which leverage point inside the system shifts it sustainably?
04
Flow
Marker: Phase
| Step | Duration | Action | Hint |
|---|---|---|---|
1Phase 1: Behavior over time | 30 min | Sketch the reference mode: show the behavior of the key variable over time as a graph (past plus expected future). Draw both observed and desired behavior. | If nobody can sketch the behavior over time, data is missing. A model without a reference mode becomes theory without grounding. |
2Phase 2: Collect variables | 30 min | One card per variable. Variables are measurable (count, rate, share), not qualities. 'Motivation' becomes 'approval rate per survey'. About 8-15 variables. | If variables are not formulated in a measurable way, the model produces unclear arrows. Minimum standard: I could theoretically quantify the variable. |
3Phase 3: Causal arrows | 60 min | Draw arrows between variables, each with a sign. Test: if A increases, what happens to B, all else equal? Only direct causality, no multi-step jumps. | If an arrow is uncertain, mark the assumption with a question mark. Multiple steps between A and B often hide variables that need to be added. |
4Phase 4: Identify loops | 45 min | Find closed loops. For each loop multiply the signs: even number of minus signs = R (reinforcing), odd number = B (balancing). Note loop label and character (growth, stabilization, delay). | Identify the central loops when there are several. If only reinforcing loops appear, balancing mechanisms are missing and the system would explode or collapse. |
5Phase 5: Leverage points and interventions | 30 min | Discuss leverage points per top loop (Donella Meadows: parameters, loop strengths, loop structure, goals, paradigms). Sketch one concrete intervention per leverage point. | Parameter changes are the lowest leverage level. If all interventions are parameters, the facilitator asks for structure- or goal-related leverage points. |
05
Artifact
What comes out at the end
Form
Causal Loop Diagram as an image plus a Markdown companion document with reference mode, variable list, arrow list with signs, loop description (R/B, effect), identified leverage points, and intervention sketches.
Versioning / ownership
One version per iteration with date. Document model changes with justification. Mark hypothesis arrows separately from validated arrows so model quality stays visible.
Tool alternatives
- Miro or FigJam with a custom notation
- Kumu for interactive loop visualization
- Vensim or Stella for formal simulation
- Loopy (ncase.me/loopy) for animated models
- draw.io or Lucidchart with a system dynamics library
causal-loop-diagram-working-template.md
Compact working template for Causal Loop Diagram with context, input, output artifacts, and next step.
Causal Loop Diagram Canvas
Context
What is this method used for?
Core question
Which question should be answered at the end?
Input
Which data, observations, or materials are available?
Working area
- Area 1:
- Area 2:
- Area 3:
- Relationships / patterns:
Output artifacts
- Causal Loop Diagram:
- Feedback Notes:
- Leverage Points:
Open questions
- ...
Next step
Owner, date, success signal.
06
Example output
Concrete filled scenario, fictional example
causal-loop-diagram-beispiel.md
Concrete filled scenario, fictional example
Causal Loop Diagram - Onboarding workspot, May 2026
Observed behavior: Activation rate rises briefly after a marketing push, then falls below the prior period. The reference mode shows a sawtooth pattern (spike, drop, spike, drop).
Variables (excerpt)
- A: New sign-ups per week
- B: Onboarding wait time for Customer Success (hours)
- C: Activation rate (value per sign-up)
- D: Month-1 churn rate
- E: Customer Success capacity
- F: Marketing budget per quarter
Causal arrows
- F -> A (+): More budget means more sign-ups.
- A -> B (+): More sign-ups lengthen wait time (CS capacity fixed).
- B -> C (-): Longer wait time reduces activation.
- C -> D (-): Lower activation increases month-1 churn.
- D -> F (+): Higher churn creates pressure to increase marketing for replacement users.
Loops
- R1: F -> A -> B -> C -> D -> F. Reinforcing loop 'churn spiral': more marketing creates more churn, which justifies more marketing. Main cause of the sawtooth.
- B1: E -> B (-). Balancing, but E does not change.
Leverage points
- Weaken loop strength: couple CS capacity (E) to A in a flexible way.
- Change the goal: not 'more sign-ups' but 'more activations'.
- Structure: make onboarding self-serve so B does not depend on E.
Interventions
- CLD-01: Dynamically couple CS slots to sign-ups. Owner: @lisa.
- CLD-02: Build self-serve onboarding. Owner: @ben, spike by 30.06.
- CLD-03: Shift the North Star to activation instead of sign-ups. Owner: @julia (Head of Growth).
07
Pitfalls
Recognize symptoms and steer against them
Trap
Adjectives instead of variables
Symptom
Cards contain 'good mood' or 'complex architecture' without a measurable lever.
What to do
Name a measurement path for each variable, even if rough. If it is not measurable, split the variable. 'Architecture complexity' becomes 'number of modules' or 'cycle time per deploy'.
Trap
Arrows without signs
Symptom
Connections look plausible, but loop analysis is impossible.
What to do
Notation is strict. An arrow without + or - is not accepted. If the direction is unclear, mark a double arrow with a note about delay.
Trap
Model without delays
Symptom
The model suggests immediate effects, and the real dynamics with delay disappear.
What to do
Use delay notation (||) on arrows with a significant time lag. The reference mode often shows delays as waves, which motivates the notation.
Trap
Too many variables
Symptom
The model has 40 variables and 80 arrows, and nobody can find the loops.
What to do
Reduce to 10-15 central variables. Move detail variables to submodels or a notes appendix. Model quality beats completeness.
Trap
Leverage point at parameter level
Symptom
All interventions change only values (more budget, more people), not structure or goals.
What to do
Walk through Donella Meadows' leverage points. Start at least one intervention at the structure or goal level.
08
Stop criteria
Done signals checkable in under a minute
The problem is linear and one-off, with no feedback visible.
Variables cannot be formulated in a measurable way, so the model remains a word cloud.
The reference mode cannot be sketched and the behavior over time is unknown.
The workshop is shorter than 90 minutes, so loops cannot be identified cleanly.
Participants insist on solutions, and the model becomes justification instead of diagnosis.
No willingness to address structural leverage points, only parameter optimization is wanted.
Finished the runsheet?
Go to the profile for purpose, similar methods, and sources or continue to the next method in the catalog.