Causal Loop Diagram

Thinking Tools3.3 Mapping the System Around the ProblemAlso known as CLD

First drawn by Magoroh Maruyama (The Second Cybernetics, 1963), taken up in the system dynamics of Jay W. Forrester (Industrial Dynamics, 1961), and carried to managers by Peter M. Senge (The Fifth Discipline, 1990) and Donella H. Meadows (Thinking in Systems, 2008).

Links the variables in a situation with arrows marked same or opposite direction, then names the reinforcing and balancing loops that drive its behavior over time.

Hiring grows a consulting firm, but every newcomer first draws on the seniors who do the billable work

Each arrow is marked + where the two move the same way and − where they move opposite ways. Select R1 or B1 to light a loop, then set the hiring push and the senior time each newcomer takes, and Play: billable work dips while the balancing loop bites, then the reinforcing loop takes over. Illustrative.

Select a loop to trace it, or a box to read it.R1, Growth. More consultants bill more work, more work earns more revenue, and revenue pays for more consultants. No negative (−) links: an even count, so the loop reinforces.B1, Onboarding drag. Every hire adds to the onboarding load, which pulls seniors off billable work and slows the revenue that pays for the next hire. One negative (−) link: an odd count, so the loop balances.Consultants. Everyone on the payroll who can bill, seniors and newcomers alike.Billable work delivered. Consultant-months billed to clients. Newcomers bill a little; seniors bill less while they coach.Revenue. Billable work times the rate. Revenue above today's level pays for extra hires.Onboarding load. Newcomers still learning the work. Each one takes senior time for about three months.
4
100%

Newcomers still onboarding

0: 0.01: 0.02: 4.03: 6.74: 8.45: 9.06: 9.27: 9.58: 10.29: 11.210: 12.411: 13.812: 15.20.010.020.030.001234567891011120: 0.01: 0.02: 4.03: 6.74: 8.45: 9.06: 9.27: 9.58: 10.29: 11.210: 12.411: 13.812: 15.20.010.020.030.00123456789101112

Billable work, consultant-months

0: 30.01: 30.02: 30.03: 26.84: 26.05: 26.86: 29.27: 32.08: 34.89: 37.510: 40.011: 42.812: 45.80.020.040.060.080.001234567891011120: 30.01: 30.02: 30.03: 26.84: 26.05: 26.86: 29.27: 32.08: 34.89: 37.510: 40.011: 42.812: 45.80.020.040.060.080.00123456789101112

Month 12 of 12

Newcomers still onboarding
15.2
Billable work, consultant-months
45.8

R1, Growth
More consultants bill more work, more work earns more revenue, and revenue pays for more consultants. No negative (−) links: an even count, so the loop reinforces.
B1, Onboarding drag
Every hire adds to the onboarding load, which pulls seniors off billable work and slows the revenue that pays for the next hire. One negative (−) link: an odd count, so the loop balances.
Consultants
Everyone on the payroll who can bill, seniors and newcomers alike.
Billable work delivered
Consultant-months billed to clients. Newcomers bill a little; seniors bill less while they coach.
Revenue
Billable work times the rate. Revenue above today's level pays for extra hires.
Onboarding load
Newcomers still learning the work. Each one takes senior time for about three months.

An example

A thirty-person operations consultancy had more work than it could staff, so the partners approved four hires a month. Three months later billable hours were down, two clients were complaining about missed dates, and one partner wanted to stop hiring altogether. The managing partner put the four of them in front of a whiteboard and asked only what moves what.

They drew consultants driving billable work, billable work driving revenue, and revenue paying for the next hires: a reinforcing loop, the growth they had planned for. Then the delivery lead added the arrow nobody had drawn. Every hire raised the onboarding load, and the load pulled seniors off client work, with a delay of a few weeks. That closed a balancing loop through the same revenue. The dip was the balancing loop, and it would fade as newcomers learned the work. The partners kept hiring, cut the pace to two a month, and gave each cohort one senior coach whose client hours were planned down in advance.

Why it works

People explain events in straight lines: we hired, so output should rise. Many business problems run in circles, and a straight-line explanation cannot see a circle. Drawing every variable that rises or falls, and marking each arrow same direction or opposite, makes the circles visible. A loop with an even number of opposite arrows reinforces itself and grows or collapses; a loop with an odd number balances and pulls toward a limit.

Most puzzling behavior comes from two loops of different kinds sharing variables, often with a delay on one arrow. The delay is why a sound decision looks wrong for a few months, and why a fix that works at first can backfire later. Once the loops are named, the argument moves from blaming people to choosing where to intervene.

The thinking behind it

Magoroh Maruyama Magoroh Maruyama's The Second Cybernetics: Deviation-Amplifying Mutual Causal Processes (American Scientist, 1963) described a second kind of feedback. The first cybernetics had studied feedback that corrects deviations and keeps a system stable; he described mutual causal processes that amplify a small initial deviation, and used them to explain how growth, differentiation, and divergence arise. His distinction is the one the R and B labels draw: loops that amplify and loops that correct.

Jay W. Forrester Jay W. Forrester, an MIT engineer and management theorist, founded system dynamics with Industrial Dynamics (1961), which models an organization as stocks, flows, and feedback loops with delays. The book drew on a study of an electronics manufacturer and gave the first account of what is now called the bullwhip effect. The plus and minus notation and the loop vocabulary grew up in his MIT group and its successors.

Peter M. Senge Peter M. Senge, a senior lecturer at MIT Sloan, brought system dynamics to managers in The Fifth Discipline (1990), which places systems thinking among five disciplines of a learning organization. Its appendix names the system archetypes, recurring feedback structures such as fixes that fail and shifting the burden, which are the patterns a finished causal loop diagram is checked against.

How to run it

  1. Name the variables. Write each as a quantity that can rise or fall, such as consultants, onboarding load, or revenue.
  2. Draw the arrows. Add an arrow wherever one variable directly changes another, and leave out arrows nobody can explain.
  3. Mark each arrow + or −. Use + when both move the same way and − when they move opposite ways.
  4. Mark the delays. Put two short bars across any arrow whose effect takes weeks or months to arrive.
  5. Find and label the loops. Count the − arrows in each closed loop: even is reinforcing (R), odd is balancing (B).
  6. Check against the archetypes. Compare the loops with known patterns such as limits to growth or fixes that fail.
  7. Choose the point of intervention. Pick the arrow or delay where a change would shift the balance between loops, and predict what will happen first.

Where it goes wrong

The first failure is a diagram so complete nobody can read it. Thirty variables and fifty arrows hide the two loops that matter. Spot it when nobody in the room can retell the main loop aloud, and cut back to the variables that change the behavior in question.

The second is an arrow that is really an event, such as "the merger" or "the new CEO". Events do not rise or fall, so the signs stop meaning anything. Replace each with the quantity it changed. The third is leaving out delays. The diagram then predicts a smooth response, the real system dips first, and the team abandons a sound decision during the dip.

When to use something else

  • Stock-and-Flow Models. When the loops are agreed and the team needs numbers: how big, how fast, and when the dip ends.
  • Iceberg Model. When the group is still arguing about events and needs to move to patterns before any loop can be drawn.
  • Limits to Growth. When the diagram already looks like growth running into a constraint and the question is which limit to remove.

Where it sits in SOLVE eX

Go to the source

  • Industrial Dynamics, 1961
  • The Second Cybernetics: Deviation-Amplifying Mutual Causal Processes, American Scientist 51(2), 1963
  • The Fifth Discipline, 1990
  • Thinking in Systems: A Primer, 2008

Back to the catalog