T O O L B O X REP R INTS ERI E S SYSTEMS - The ... O O L B O X REP R INTS ERI E S A User’s...

T O O L B O X R E P R I N T S E R I E S

A User’s Reference Guide

B Y DANIEL H. K IM

SYSTEMSTHINKING TOOLS

�

�

Systems Thinking Tools: A User's Reference Guideby Daniel H. Kim© 1994, 2000 by Pegasus Communications, Inc.First printing January 1994.

All rights reserved. No part of this book may be reproduced or transmitted in any form or by anymeans, electronic or mechanical, including photocopying and recording, or by any informationstorage or retrieval system, without written permission from the publisher.

Acquiring editor: Kellie Wardman O’ReillyProduction: Nancy Daugherty

ISBN 1-883823-02-1

PEGASUS COMMUNICATIONS, INC.One Moody StreetWaltham, MA 02453-5339 USAPhone 800-272-0945 / 781-398-9700Fax [email protected] / [email protected]

www.pegasuscom.com

.

T H E T O O L B O X R E P R I N T S E R I E S

Systems Archetypes I: Diagnosing Systemic Issues and Designing High-Leverage Interventions

Systems Archetypes II: Using Systems Archetypes to Take Effective Action

Systems Archetypes III: Understanding Patterns of Behavior and Delay

Systems Thinking Tools: A User’s Reference Guide

The “Thinking” in Systems Thinking: Seven Essential Skills

http://www.pegasuscom.com

mailto:[email protected]

mailto:[email protected]

T A B L E O F C O N T E N T S

Introduction 3

PART I: AN OVERVIEW 5Systems Thinking as a Language by Michael R. Goodman 6Levels of Understanding: “Fire-Fighting” at Multiple Levels 8A Palette of Systems Thinking Tools 10

PART II: DYNAMIC THINKING TOOLS 13Reinforcing and Balancing Loops: Building Blocks of Dynamic Systems14Balancing Loops with Delays: Teeter-Tottering on Seesaws 16Guidelines for Drawing Causal Loop Diagrams 18Systems Archetypes at a Glance 20

PART III: STRUCTURAL THINKING TOOLS 23From Causal Loops to Graphical Functions: Articulating Chaos 24Graphical Functions: “Seeing” the Full Story 26Structural Thinking: The World According to Accumulators and Flows 28Accumulators: Bathtubs, Bathtubs, Everywhere 30Accumulation Management: Avoiding the “Pack Rat” Syndrome 32Delays: Accumulators in Disguise 34S-Shaped Growth and the Law of Diminishing Returns 36

PART IV: COMPUTER-BASED TOOLS 39Modeling for What Purpose? by Jay W. Forrester 40Management Flight Simulators: Flight Training for Managers—Part I 42Management Flight Simulators: Flight Training for Managers—Part II 44Learning Laboratories: Practicing Between Performances 46

PART V: REFERENCE GUIDE 49The Vocabulary of Systems Thinking: A Pocket Guide 50The Do’s and Don’t’s of Systems Thinking on the Job

by Michael R. Goodman 52Further Reading on the Ten Tools of Systems Thinking 54Index to THE SYSTEMS THINKER 55

3

e are in the midst of a changing of an age—from the age of machines to the sys-tems age. Our past was defined by a view of the world as a machine that could be

understood by breaking it into smaller and smaller parts. In the machine age view of theworld, the parts are what is most important—by understanding each of the parts, webuild up our understanding of the larger wholes. In the systems view, it is the whole thatis most important—parts in isolation have no meaning in and of themselves. Systemsthinking embodies the idea that the interrelationships among parts relative to a commonpurpose of a system are what is important.

There is nothing more powerful than an idea whose time has come. But ideaswithout practical tools can take us only so far in making any meaningful changes that willhave an impact on the world. Systems thinking provides the ideas that can help us see theworld in new ways, as well as the tools that can help us take new actions that are systemicand more effective. This booklet provides a basic introduction to the various tools of sys-tems thinking that have been developed and used over the last 50 years.

ACKNOWLEDGMENTS

Much of this work has been developed over the years through the efforts of many systemdynamicists. Systems Thinking Tools: A User’s Reference Guide, part of the Toolbox ReprintSeries, was created and compiled by Kellie Wardman O’Reilly.

W

P E G A S U S C O M M U N I C A T I O N S , I N C . W W W . P E G A S U S C O M . C O M S Y S T E M S T H I N K I N G T O O L S 3

T H E L A N G U A G E O F L I N K S A N D L O O P S

I N T R O D U C T I O N

accumulator

population

deaths

flow regulator

connector to indicate causal connectionflow pipe

births

“clouds” represent the boundaries of what we want to

include in the diagram

A causal link between two variables,where a change in X causes a changein Y in the same direction, or where Xadds to Y.A causal link between two variables,where a change in X causes a changein Y in the opposite direction, or whereX subtracts from Y.A “reinforcing” feedback loop thatamplifies change.A “balancing” feedback loop that seeksequilibrium.

If there is a gap between the desired leveland the actual level, adjustments are madeuntil the actual equals the desired level. Thestarting variable is grey.

B A L A N C I N G L O O PE X A M P L E

T H E L A N G U A G E O FA C C U M U L AT O R S

s

+

o_

R

B

B

Gap

B

s

ActualLevel

Adjust-ments

ActualLevel

Adjust-ments

Gapos

–

+

+

+

Desired Level

Desired Level

s

Dela

y

P E G A S U S C O M M U N I C A T I O N S , I N C . W W W . P E G A S U S C O M . C O M S Y S T E M S T H I N K I N G T O O L S 5

ystems thinking can be thought of as a language. As a language, it is a specificway of viewing the world; it affects thought, and thought in turn affects how we

look at the world. “Systems Thinking as a Language” (p. 6) offers insight into how sys-tems thinking can be a useful framework for communicating about complex issues.

By “conversing” in the language of feedback loops, we can learn to better articulatethe complex interconnections of circular causality in which we live. Learning the lan-guage of systems thinking requires us to understand our world on at least four levels—events, pattern of events, systemic structure, and shared vision. “Levels ofUnderstanding: ‘Fire-Fighting’ at Multiple Levels” (p. 8) describes these levels and thespecific action mode associated with each one.

This section closes with “A Palette of Systems Thinking Tools” (p. 10), which out-lines ten tools of systems thinking. Seven of these tools are covered in the subsequentsections of this booklet.

S

P A R T I

A N O V E R V I E W

S Y S T E M S T H I N K I N G A S AL A N G U A G EB Y M I C H A E L R . G O O D M A N

T O O L B O X

anguage has a subtle, yet powerfuleffect on the way we view the world.

English, like most other Western languages,is linear—its basic sentence construction,noun-verb-noun, translates into a world-view of “x causes y.” This linearity predis-poses us to focus on one-way relationshipsrather than circular or mutually causativeones, where x influences y, and y in turninfluences x. Unfortunately, many of themost vexing problems confronting man-agers and corporations today are caused bya web of tightly interconnected circularrelationships. To enhance our understand-ing and communication of such problems,we need a language more naturally suitedto the task.

ELEMENTS OF THELANGUAGE

Systems thinking can be thought of as alanguage for communicating about com-plexities and interdependencies. In particu-lar, the following qualities make systemsthinking a useful framework for discussingand analyzing complex issues:• Focuses on “closed interdependencies.”The language of systems thinking is circu-lar rather than linear. It focuses on closedinterdependencies, where x influences y, yinfluences z, and z influences x.• Offers a “visual” language.Many of thesystems thinking tools—causal loop dia-grams, behavior over time diagrams, sys-tems archetypes, and structuraldiagrams—have a strong visual component.They help clarify complex issues by sum-ming up, concisely and clearly, the key ele-ments involved.

L

S Y S T E M S T H I N K I N G T O O L S6

Diagrams also facilitate learning.Studies have shown that many people learnbest through representational images, suchas pictures or stories. A systems diagram isa powerful means of communicationbecause it distills the essence of a probleminto a format that can be easily remem-bered, yet is rich in implications andinsights.

• Adds precision. The specific set of “syn-tactical” rules that govern systems diagramsgreatly reduce the ambiguities and miscom-munications that can occur when we tacklecomplex issues.

Example: In drawing out the relation-ships between key aspects of a problem,causal links are not only indicated byarrows, but are labeled “s” (same) or “o”(opposite) to specify how one variableaffects another. Such labeling makes thenature of the relationship more precise,ensuring only one possible interpretation.• Forces an “explicitness” of mental mod-els. The systems thinking language trans-lates “war stories” and individual

A systems diagram is apowerful means ofcommunication because itdistills the essence of aproblem into a format that canbe easily remembered, yet isrich in implications andinsights.

P E G A S U S C O M

perceptions of a problem into black-and-white pictures that can reveal subtle differ-ences in viewpoint.

Example: In one systems thinkingcourse, a team of managers was working onan issue they had been wrestling with formonths. One manager was explaining hisposition, tracing through the loops he haddrawn, when a team member stopped him.“Does that model represent your thinkingabout this problem?” he asked.

The presenter hesitated a bit, reviewedhis diagram, and finally answered, “Yes.”

The first man, evidently relieved,responded, “After all of these months, Ifinally really understand your thoughts onthis issue. I disagree with it, but at least nowthat we are clear on our different view-points, we can work together to clarify theproblem.”• Allows examination and inquiry.Systems diagrams can be powerful meansfor fostering a collective understanding of aproblem. Once individuals have stated theirunderstanding of the problem, they can col-laborate on addressing the challenges itposes. And by focusing the discussion onthe diagrams, systems thinking defusesmuch of the defensiveness that can arise ina high-level debate.

Example: When carrying on a systemsdiscussion, differing opinions are no longerviewed as “human resources’ view of ourproductivity problem” or “marketing’sdescription of decreasing customer satisfac-tion,” but simply different structural repre-sentations of the system. This shifts thefocus of the discussion from whetherhuman resources or marketing is right, to

M U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

constructing a diagram that best capturesthe behavior of the system.• Embodies a worldview that looks atwholes, rather than parts, and that recog-nizes the importance of understanding howthe different segments of a system are inter-connected. An inherent assumption of thesystems thinking worldview is that prob-lems are internally generated—that weoften create our own “worst nightmares.”

Example: At systems thinking coursesat Innovation Associates, participants play aboard game known as the Beer Game,where they assume the position of retailer,wholesaler, distributor, or producer. Eachplayer tries to achieve a careful balancebetween carrying too much inventory orbeing backlogged. When things go wrong,many people blame their supplier (“I keptordering more, but he didn’t respond”) orthe buyers (“fickle consumers—one daythey’re buying it by the truckload, the nextday they won’t even touch the stuff”). Inreality, neither the buyers nor the suppliersare responsible for the wide fluctuations ininventory—they are a natural consequenceof the structure of the system in which theplayers are functioning.

The systems thinking worldview dis-pels the “us versus them” mentality byexpanding the boundary of our thinking.Within the framework of systems thinking,“us” and “them” are part of the same sys-tem and thus responsible for both the prob-lems and their solutions.

LEARN ING THELANGUAGE

Learning systems thinking can be likenedto mastering a foreign language. In school,we studied a foreign language by firstmemorizing the essential vocabularywords and verb conjugations. Then webegan putting together the pieces into sim-ple sentences. In the language of systemsthinking, systems diagrams such as causalloops can be thought of as sentences con-

P E G A S U S C O M M U N I C A T I O N S , I N C .

structed by linking together key variablesand indicating the causal relationshipsbetween them. By stringing together sev-eral loops, we can create a “paragraph”that tells a coherent story about a particu-lar problem under study.

If there were a Berlitz guide to systemsthinking, archetypes such as “Fixes ThatFail” or “Shifting the Burden” would belisted as “commonly used phrases.” Theyprovide a ready-made library of commonstructures and behaviors that can apply tomany situations. Memorizing them canhelp you recognize a business situation orproblem that is exhibiting common symp-toms of a systemic breakdown.

Of course, the key to becoming moreproficient in any language is to practice—and practice often. When reading a news-paper article, for example, try to“translate” it into a systems perspective:• Take events reported in the newspaperand try to trace out an underlying patternthat is at work.• Check whether the story fits one of thesystems archetypes, or whether it is per-haps a combination of several archetypes.• Try to sketch out a causal loop or twothat captures the structure producing thatpattern.

Don’t expect to be fluent in systemsthinking right away. Remember, after yourfirst few Latin classes, you still couldn’tread The Odyssey. For that matter, you prob-ably knew only a few key phrases andvocabulary words, but you improved your

An inherent assumption of thesystems thinking worldview isthat problems are internallygenerated—we often create ourown “worst nightmares.”

W W W . P E G A S U S C O M . C O M

skills by using the language as often as pos-sible. The same holds true for systemsthinking.

When sitting in a meeting, see if youcan inform your understanding of a prob-lem by applying a systems perspective.Look for key words that suggest linearthinking is occurring—statements such as“we need more of the same” or “that solu-tion worked for us the last time this hap-pened, why not use it again?” You can alsocreate low-key practice sessions by workingwith a small team of colleagues to diagrama particular problem or issue.

BECOM ING FLUENT

We say someone is fluent when they beginto think in a particular language and nolonger have to translate. But fluency meansmore than just an ability to communicate ina language; it means understanding the sur-rounding culture of the language—theworldview. As with any foreign language,mastering systems thinking will allow us tofully engage in and absorb the worldviewthat pervades it. By learning the languageof systems thinking, we will hopefullychange not only the way we discuss com-plex issues, but the way we think aboutthem as well.

Michael Goodman is an associate director ofInnovation Associates, Inc. (Cambridge, MA).Thematerial in this article was drawn from his 20 yearsof experience in the field, as well as from businesscourses developed by Innovation Associates.

•

7S Y S T E M S T H I N K I N G T O O L S

T O O L B O X

L E V E L S O F U N D E R S T A N D I N G :“ F I R E - F I G H T I N G ” A T M U LT I P L EL E V E L S

t’s another busy night in thehospital emergency room. Several

car accident victims have been rushed intosurgery, one little boy is having a brokenarm set, a drug overdose victim is beingtreated, and numerous other people fill thechairs in the waiting room. Each night isdifferent, and yet each one is also the same.The doctors and nurses must act fast totreat the most seriously injured, while theothers wait their turn. Like an assemblyline of defective parts, patients are diag-nosed, treated, and released. Each injury isa crisis that demands immediate attention.

So what’s wrong with this picture?After all, isn’t this what emergency roomsare meant to do? The answer depends onthe level of understanding at which we arelooking at the situation.

LEVELS OFUNDERSTAND ING

There are multiple levels from which wecan view and understand the world. From asystemic perspective, we are interested in

I


Levels of Understanding

Systemic Structure

Patterns of Events

Events

Shared Vision

Action Mode Time

Generative

Creative

Adaptive

Reactive

L E V E L S O F U N D

four distinct levels—events, patterns ofevents, systemic structure, and shared vision(see “Levels of Understanding”). Events arethe things we encounter on a day-to-daybasis: a machine breaks, it rains, we eat din-ner, see a movie, or write a report. Patternsof events are the accumulated memories ofevents—when strung together in a seriesover time, they reveal recurring patterns.Systemic structure can be viewed as “eventgenerators” because they are responsible forproducing the events. Similarly, sharedvision can be viewed as “systemic structuregenerators” because they are the guidingforce behind the creation or change of allkinds of structures.

We live in an event-oriented world, andour language is rooted at the level of events.At work, we encounter a series of events,which often appear in the form of problemsthat we must “solve.” Our solutions, how-ever, may be short-lived, and the symptomscan eventually return as seemingly newproblems (see “Using ‘Fixes That Fail’ toGet off the Problem-Solving Treadmill,”

P E G A S U S C O M

Orientation

Future

Present

Typical Questions

What are the stated or unstated visions that generate the structures?

What are the mental or organizational structures that create the patterns?

What kinds of trends or patterns of events seem to be recurring?

What is the fastest way to react to this event NOW?

E R S T A N D I N G

THE SYSTEMS THINKER, V3N7). Thisis consistent with our evolutionary history,which was geared toward responding tothose things that posed an immediate dan-ger to our well-being.

Events require an immediate response.If a house is burning, we react by takingaction to put out the fire. Putting out thefire is appropriate, but if it is the only actionthat is ever taken, it is inadequate from asystemic perspective. Why? Although itsolved the immediate problem (the burninghouse), it has done nothing to alter the fun-damental structure that caused that event(e.g., inadequate building codes, lack of firedetectors, fire prevention education). The“Levels of Understanding” diagram andframework can help us go beyond typicalevent-orientation responses and begin tolook for higher leverage actions.

FROM F IRE - F I GHT ING TOF IRE PREVENT ION

At the event level, if a house is on fire, allwe can do is react as quickly as possible toput the fire out. The only mode of actionthat is appropriate and available is to bereactive. If we reacted to fires only at theevents level, we would put all of our energyinto fighting fires—and we would proba-bly have a lot more fire stations than we dotoday.

If we look at the problem of fires at thepattern of events level, we can begin toanticipate where they are more likely tooccur. We may notice that certain neigh-borhoods seem to have more fires than oth-ers. We are able to be adaptive by locatingmore fire stations in those areas, and

M U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

staffing them accordingly (based on pastpatterns of usage). Since the stations are alot closer, we can be more effective atputting out fires by getting to them sooner.Yet while being adaptive allows us to bemore effective fire-fighters, it does nothingto reduce the actual occurrence of fires.

At the systemic structure level we beginasking questions: “Are smoke detectorsbeing used? What kinds of building materi-als are less flammable? What safety featuresreduce fatalities?” Actions taken at this levelcan actually reduce the number and severityof fires. Establishing fire codes with require-ments such as automatic sprinkler systems,fireproof materials, fire walls, and fire alarmsystems saves lives by preventing or contain-ing fires. Actions taken at this level are cre-ative because they help create a differentfuture.

Systemic structure includes not onlythe organizational structures and physicalbuildings, but people’s mental models andhabits as well. Where do the systemic struc-tures come from? They are usually a reflec-tion of a shared vision of what is valued ordesired. In the case of fire-fighting, thenew structures (e.g., fire codes) are bornout of a shared value of the importance ofprotecting human lives, combined with thedesire to live and work in safe buildings.At the level of shared vision, our actionscan be generative, bringing something intobeing that did not exist before. We beginasking questions like “What’s the role ofthe fire-fighting function in this commu-nity? What are the trade-offs we are will-ing to make as a community between theamount of resources devoted to fire-fight-ing compared to other things?”

It is important to remember that theprocess of gaining deeper understanding isnot a linear one. Our understanding of asituation at one level can feed back andinform our awareness at another level.Events and patterns of events, for example,can cause us to change systemic structures


and can also challenge our shared vision.To be most effective, the full range of levelsmust be considered simultaneously. Thedanger lies in operating at any one level tothe exclusion of the others.

Our ability to influence the future doesincrease, however, as we move from thelevel of events to shared vision. Does thismean that high-leverage actions can onlybe found at higher levels? No, becauseleverage is a relative concept, not an abso-lute. When someone is bleeding, the high-est leverage action at that moment is to stop

the bleeding––any other action would beinappropriate. As we move up the levelsfrom events to shared vision, the focusmoves from being present-oriented tobeing future-oriented. Consequently, theactions we take at the higher levels havemore impact on future outcomes, not pre-sent events.

BACK AT THEEMERGENCY ROOM

While the emergency room (ER) offers agraphic example of a situation in whichpeople must be focused on the present, italso reveals the limitations of the events-oriented response. ER treatment offersmaximal leverage to affect the present situ-ation with each patient, but it provides verylittle leverage for changing the future. Ifwe go up one level and examine ER usefrom a patterns of events level, we may dis-cover that certain areas of a city seem tohave higher emergency room needs. We

The process of gaining deeperunderstanding is not a linearone. Our understanding of asituation at one level can feedback and inform our awarenessat another level.


may try an adaptive response and increaseER capacity in those regions. If diversionrates are high, we can also find out wherethe ambulances are being diverted fromand try to enhance capacity there.

At the systemic structure level, we canbegin to explore why certain regions havean increased need for ERs. We may dis-cover, for example, that 40 percent of theER admissions are children who are poi-soned, because a large percentage of thecommunity cannot read English and allwarning labels are printed in English. Byredrawing the boundary of the ER issue toinclude the community, we can take actionsthat will change the inflow of patients.Electrical utilities have been doing this forsome time. Instead of building anotherexpensive power plant to supply morepower, they are working with customers toreduce the demand for power.

At a community-wide level, we maywant to explore the question, “What is ourshared vision of the role our healthcare sys-tem plays in our lives?” Perhaps theresources that are going into ERs could bebetter utilized elsewhere, such as commu-nity education and prevention programs.The highest leverage lies in clarifying thequality of life we envision for ourselves, andthen using that as a guide for creating thesystemic structures that will help us achievethat vision.

The basic message of the “Levels ofUnderstanding” diagram is the importanceof recognizing the level at which you areoperating, and evaluating whether or not itprovides the highest leverage for that situa-tion. Each level offers different opportuni-ties for high-leverage action, but they alsohave their limits. The challenge is to choosethe appropriate response for the immediatesituation and find ways to alter the futureoccurrence of those events. •

9S Y S T E M S T H I N K I N G T O O L S

T O O L B O X

A P A L E T T E O F S Y S T E M ST H I N K I N G T O O L S

here is a full array of systems think-ing tools that you can think of in the

same way as a painter views colors—manyshades can be created out of three primarycolors, but having a full range of ready-made colors makes painting much easier.

There are at least 10 distinct types ofsystems thinking tools (an abbreviated sum-mary diagram appears on the facing page).They fall under four broad categories:brainstorming tools, dynamic thinkingtools, structural thinking tools, and com-puter-based tools. Although each of thetools is designed to stand alone, they alsobuild upon one another and can be used incombination to achieve deeper insights intodynamic behavior.

BRA INSTORM ING TOOLS

The Double-Q (QQ) Diagram is based onwhat is commonly known as a fishbone orcause-and-effect diagram. The Qs stand forqualitative and quantitative, and the tech-nique is designed to help participants beginto see the whole system. During a struc-tured brainstorming session with the QQdiagram, both sides of an issue remainequally visible and properly balanced,avoiding a “top-heavy” perspective. Thediagram also provides a visual map of thekey factors involved. Once those factors arepinpointed, Behavior Over Time Diagramsand/or Causal Loop Diagrams can be usedto explore how they interact.

A QQ diagram begins with a heavyhorizontal arrow that points to the issuebeing addressed. Major “hard” (quantita-tive) factors branch off along the top and“soft” (qualitative) factors run along thebottom. Arrows leading off of the majorfactors represent sub-factors, which can in

T

S Y S T E M S T H I N K I N G T O O L S1 0

turn have sub sub-factors. Many layers ofnesting, however, may be a sign that one ofthe sub-factors should be turned into amajor factor.

DYNAM IC TH INK INGTOOLS

Behavior Over Time (BOT) Diagrams aremore than simple line projections—theycapture the dynamic relationships amongvariables. For example, say we were tryingto project the relationship between sales,inventory, and production. If sales jump 20percent, production cannot jump instanta-neously to the new sales number. In addi-tion, inventory must drop below itsprevious level while production catches upwith sales. By sketching out the behavior ofdifferent variables on the same graph, wecan gain a more explicit understanding ofhow these variables interrelate.Causal Loop Diagrams (CLDs) provide a

useful way to represent dynamic interrela-tionships. CLDs make explicit one’s under-standing of a system’s structure, provide avisual representation to help communicatethat understanding, and capture complexsystems in a succinct form. CLDs can becombined with BOTs to form structure-behavior pairs, which provide a rich frame-work for describing complex dynamicphenomena. CLDs are the systems thinker’sequivalent of the painter’s primary colors.Systems Archetypes is the name given to

certain common dynamics that seem torecur in many different settings. Thesearchetypes, consisting of various combina-tions of balancing and reinforcing loops, arethe systems thinker’s “paint-by-numbers”set—users can take real-world examplesand fit them into the appropriate archetype.

P E G A S U S C O M

They serve as a starting point from whichone can build a clearer articulation of abusiness story or issue. Specific archetypesinclude: “Drifting Goals,” “Shifting theBurden,” “Limits to Success,” “Success tothe Successful,” “Fixes That Fail,”“Tragedy of the Commons,” “Growth andUnderinvestment,” and “Escalation” (see“Systems Archetypes at a Glance,” p. 20).

STRUCTURAL TH INK INGTOOLS

Graphical Function Diagrams, Structure-Behavior Pairs, and Policy StructureDiagrams can be viewed as the buildingblocks for computer models. GraphicalFunctions are useful for clarifying nonlin-ear relationships between variables. Theyare particularly helpful for quantifying theeffects of variables that are difficult to mea-sure, such as morale or time pressure.Structure-Behavior Pairs link a specificstructure with its corresponding behavior.Policy Structure Diagrams represent theprocesses that drive policies. In a sense,when we use these tools we are movingfrom painting on canvas to sculpting three-dimensional figures.

COMPUTER - BASED TOOLS

This class of tools, including computermodels, management flight simulators, andlearning laboratories, demands the highestlevel of technical proficiency to create.On the other hand, very little advancetraining is required to use them once theyare developed. •

M U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

1 1P E G A S U S C O M M U N I C A T I O N S , I N C . W W W . P E G A S U S C O M . C O M S Y S T E M S T H I N K I N G T O O L S

D Y N A M I C T H I N K I N G T O O L S S T R U C T U R A L T H I N K I N G T O O L S C O M P U T E R - B A S E D T O O L S

Can be used to graph the behavior ofvariables over time and gain insights intoany interrelationships between them.(BOT diagrams are also known asreference mode diagrams.)

Time

BA

C

Behavior Over Time Diagram

Captures the way in which one variableaffects another, by plotting the relation-ship between the two over the full rangeof relevant values.

x

f(x)

Graphical Function Diagram

Lets you translate all relationshipsidentified as relevant into mathematicalequations. You can then run policyanalyses through multiple simulations.

Computer Model

Used in conjunction with behavior overtime diagrams, can help you identifyreinforcing (R) and balancing (B)processes.

C

B

A

s o

s

s

R B

Causal Loop Diagram

Consists of the basic dynamic structuresthat can serve as building blocks fordeveloping computer models (for exam-ple, exponential growth, delays, smooths,S-shaped growth, oscillations, and so on).

Time

Structure-Behavior Pair

Provides “flight training” for managersthrough the use of interactive computergames based on a computer model. Userscan recognize long-term consequences ofdecisions by formulating strategies andmaking decisions based on those strategies.

STOCK

HIRING

STOCKHIRINGDECISION INFOCOCKPIT

Management Flight Simulator

Helps you recognize common systembehavior patterns such as “DriftingGoals,” “Shifting the Burden,” “Limits toGrowth,” “Fixes That Fail,” and so on—all the compelling, recurring “stories” oforganizational dynamics.

Systems Archetype

A conceptual map of the decision-makingprocess embedded in the organization.Focuses on the factors that are weighedfor each decision, and can be used tobuild a library of generic structures.

Policy Structure Diagram

A manager’s practice field. Is equivalentto a sports team’s experience, whichblends active experimentation withreflection and discussion. Uses all thesystems thinking tools, from behaviorover time diagrams to MFSs.

Learning Laboratory

Experimentation

Reflection

1 3P E G A S U S C O M M U N I C A T I O N S , I N C . W W W . P E G A S U S C O M . C O M S Y S T E M S T H I N K I N G T O O L S

BEHAV IOR OVER T IME D IAGRAM

CAUSAL LOOP D IAGRAM

SYSTEMS ARCHETYPES

s discussed in “Levels of Understanding: ‘Fire-fighting’ at Multiple Levels” on p.8, we need to develop our capability to see beyond the event-to-event view of the

world. The dynamic thinking tools provide the means to represent the patterns ofevents that occur over time and also map the structures that are producing thosedynamics. This section begins by discussing reinforcing and balancing loops, which arethe fundamental building blocks that help us represent the feedback loop structuresresponsible for generating the dynamic patterns that we observe.

Although the basic concept of reinforcing and balancing loops is simple, actuallymapping out one’s own issues in a free-form causal loop diagramming session requiresa fair amount of skill. “Guidelines for Drawing Causal Loop Diagrams” (p. 18) cangive you some heuristics to follow in trying to construct your own diagrams.

Even with experience, it can be rather daunting to stare at a blank page and try toconstruct a systemic picture of your issue from scratch. This is where systemsarchetypes can be very helpful in providing the initial story line from which to elicitunderstanding of an issue. The archetypes represent generic story lines and structuresthat have been found to be prevalent in our systems. “Systems Archetypes at a Glance”(p. 20) offers descriptions and guidelines for each of the archetypes (for a more com-plete coverage of all the archetypes, see Systems Archetypes I: Diagnosing Systemic Issuesand Designing High-Leverage Interventions, also published by PegasusCommunications). By developing an understanding of each of the archetypes, we canbegin to enrich our intuitive sense of how complex structures work.

A

P A R T I I

D Y N A M I C T H I N K I N G T O O L S

R E I N F O R C I N G A N D B A L A N C I N GL O O P S : B U I L D I N G B L O C K S O FD Y N A M I C S Y S T E M S

T O O L B O X

n the book The Double Helix,James Watson describes the process

through which he and Robert Crick“cracked” the DNA code. While otherresearchers were searching for complexstructures to explain the diversity of life

I


Reinforcing loops compound change in one direcencouragement can enhance an employeeʼs perfior can lead to poor employee performance over

E M P L O Y E E - S U P E R V I S O R

EmployeePerformance

Supervisor’sSupportiveBehavior

s

s

R1

PerfLeve

Structure

Balancing loops try to bring a system to a desiredtrol system, the desired inventory is maintained bthere is too much or too little.

I N V E N T O R Y C O N T R O

Discrepancy

ActualInventory

s

sB2

Structure

InventoryAdjustments

so

DesiredInventory

forms, Watson and Crick explored moresimple geometrical designs. They eventu-ally received the Nobel Prize for revealingthe double helix structure that is thegenetic basis for all life. Through theirresearch, Watson and Crick proved that

P E G A S U S C O M

tion with even more change. For example,ormance, while critical or unsupportive behav-time.

R E I N F O R C I N G L O O P

Time

.l

Behavior Over Time

UnsupportiveBehavior

SupportiveBehavior

state and keep it there. In an inventory con-y adjusting the actual inventory whenever

L B A L A N C I N G L O O P

Time

Perf.Level

Behavior Over Time

Desired Inventory

ActualInventory

ActualInventory

the infinite variations we see in nature canall be produced by one simple, elegantstructure.

Similarly, two basic loops—reinforcingand balancing—can be seen as the equiva-lent building blocks of complex social andeconomic systems. These simple structurescombine in an infinite variety of ways toproduce the complex systems that we asmanagers are expected to control.

RE INFORC ING LOOPS :ENG INES OF GROWTHAND DECAY

Reinforcing loops produce both growthand decay. That is, they compound changein one direction with even more change.For example, in the “Employee-SupervisorReinforcing Loop” diagram, encourage-ment from the supervisor is capable ofproducing good employee performance—that is, as the supervisor demonstratessupportive behavior, the employee’s perfor-mance will improve, which will lead thesupervisor to be even more supportive. Atthe same time, unsupportive behavior canproduce poor employee performance overtime—if the supervisor is not supportive,performance will likely decrease, leadingthe supervisor to be even less supportive.The same loop can create either kind ofreinforcing cycle.

BALANC ING LOOPS :GOAL - SEEK INGPROCESSES

Of course, most things in life cannot con-tinue growing forever. There are otherforces—balancing loops—that resist fur-

M U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

Reinforcing and balancing loops can be combined to describe more complex behavior. Forexample, encouragement by the supervisor could lead the employee to work longer and longerhours in order to continue impressing the supervisor, eventually leading to burnout and adecrease in performance.

R E I N F O R C I N G L O O P C O U P L E D W I T HA B A L A N C I N G L O O P

EnergyLevel

HoursWorked

s

s

B2

Time

Perf.Level

Behavior Over TimeStructure

Diminishing Returns

“Burnout”

Supervisor’sSupportiveBehavior

o

PositiveReinforcement

s

EmployeePerformances

R1R

B

ther increases in a given direction.Balancing loops try to bring things to adesired state and keep them there, muchlike a thermostat regulates the temperaturein a house.

An equivalent example in manufactur-ing involves maintaining buffer inventorylevels between production stages. In thissituation, there is a desired inventory levelthat is maintained by adjusting the actualinventory whenever there is too much ortoo little (see “Inventory Control BalancingLoop”).

US ING THE BU I LD INGBLOCKS

To see how these two basic loops can com-bine to form more complex structure-behavior pairs, let’s revisit theemployee-supervisor feedback loop.Clearly the employee’s performance willnot improve indefinitely just because thesupervisor is supportive. The employeemay have been putting in longer hours inorder to continue impressing the supervi-sor. Over a period of time, the increasedwork hours may begin to wear down theemployee’s energy level (see “ReinforcingLoop Coupled with a Balancing Loop”


diagram). If this continues, at some pointthe supervisor’s supportive behavior will beeclipsed by the sheer energy drain ofworking long hours. Improved perfor-mance will gradually be offset by theeffects of burnout, until finally the balanc-ing loop connecting energy level and hoursworked becomes dominant. At this pointthe employee’s performance will eitherplateau or decline.


SUMMARY PO INTS

All complex dynamic behavior is producedby two loops: reinforcing and balancing.Behind every growth or decay is at leastone reinforcing loop. For every goal-seeking behavior, there is a balancing loop.

A period of growth followed by aslowdown in growth is usually caused by ashift in dominance from a reinforcing to abalancing loop. •

S Y S T E M S T H I N K I N G T O O L S 1 5

B A L A N C I N G L O O P S W I T H D E L A Y S :T E E T E R - T O T T E R I N G O N S E E S AW S

T O O L B O X

ost of us have played on a seesaw atone time or another and can recall

the up and down motion as the momen-tum shifted from one end to the other. Themore equal the weights of both people, thesmoother the ride. At a very basic level, afree market economy is a lot like a seesawwith supply at one end and demand on theother end. Prices indicate the imbalancebetween the two, like a needle positionedat the pivot point of the seesaw.

M


A free market economy is a lot like a seesaw withThe dynamics that result from trying to balance suing loops that try to stabilize on a particular price.cycle of overshoot and collapse occurs.

S U P P L Y A N

Supply

Lo

Price

s

Supply

s

o

B1 Price

Delay

The goal of a seesaw ride is to alwayskeep things in a state of imbalance (itwould be pretty boring to sit on a perfectlybalanced one). But the goal in the market-place is exactly the opposite—to bringsupply in balance with demand. Unfortu-nately, the supply and demand balancingprocess feels a lot more like a seesaw ridethan a smooth adjustment to a stable equi-librium. As shown in a causal loop dia-gram, the dynamics of this adjustment

P E G A S U S C O M

supply at one end and demand on the other.pply and demand are produced by two balanc-Due to the presence of significant delays, a

D D E M A N D

DemandHi

o

B2 Demand

Delay

Demand

Supply

Time

process are produced by two balancingloops that try to stabilize on a particularprice. But the process is complicated bythe presence of significant delays (see“Supply and Demand”).

BALANC ING SUPPLY ANDDEMAND

Tracing through the loops you can see thatif demand rises, price tends to go up (allelse remaining the same), and as price goesup, demand tends to go down (BeanieBabies notwithstanding). If there is enoughinventory or capacity in the system toabsorb the increased demand, prices maynot go up immediately. As demand out-strips supply, however, price will rise.

On the supply side of the seesaw, anincrease in price provides a profit incentivefor firms to produce more. Of course, ittakes time for firms to expand. The lengthof the delay depends on how close theyalready are to full capacity and howquickly they can add capacity to producemore. Hiring new workers may take onlya few days, while obtaining additional cap-ital equipment or factory floor space maytake months or even years. While firms aremaking supply adjustments, the gapbetween supply and demand widens andprice goes even higher. The higher pricespurs companies to increase their produc-tion plans even more.

As supply eventually expands andcatches up with demand, price begins tofall. By this time, firms have overexpandedtheir production capacity and supply over-shoots demand, causing price to fall. Whenthe price falls low enough, the productbecomes more attractive again and

M U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

Air Traffic Growth

AirplaneLease Rate

s

Demand forAirplanesB3

Profits

Airplanesfor Lease

o

o

B4

s

s

s

Delay

Delay

AirplaneOrders

s

A I R P L A N E L E A S I N G I N D U S T R Y

A causal loop diagram of the airplane leasing industry shows the same seesaw structure atwork.

demand picks up, starting the cycle allover again.

A IRPLANES ON SEESAWS

The supply and demand seesaw is playedout in all but the most tightly regulatedmarkets. A good example of this balancingact is described in a Forbes article entitled“Fasten Seat Belts, Please” (April 2, 1990),about airplane leasing companies.

Leasing companies, which account forroughly 20 percent of all commercial jetaircraft currently on order, have enjoyedenormous profits during booms in airtravel. At one time, one carrier alone putin an order to lease 500 planes. Based onleasing and buying rates in the industry,the total number of airplanes was expectedto increase by 50 percent between 1990 and1995. But in the meantime, air trafficgrowth had slowed in the late 1980s. Theleasing companies, however, did not seemtoo worried.

According to the article, “eight years ofunbroken prosperity have created the illu-sion that many cyclical businesses aren’tcyclical any longer.” But, as one airlineexecutive warned, “This is a cyclical busi-ness. Always has been, always will be.With a small change in load factor, the air-lines can go from spilling cash to bleedingred ink like the Mississippi River goingthrough the delta.”

If you draw out a causal loop diagramof this industry operating in this way, yousee the same supply and demand structureat work. An increase in air traffic growth


fueled a strong demand for airplanes. Thatin turn sparked an increase in airplanelease rates as airlines scrambled for addi-tional airplanes. The high lease rates led toincreased profits and a surge in airplaneorders. Since airplanes take many monthsto build, the supply of leasable airplanesdid not adjust right away, making leaserates go even higher. This led to higherprofits, which attracted more capital,which was then plowed into even moreorders for airplanes.

As the supply catches up to demand,however, the airplane lease rates will fall(the slowing of air traffic growth will accel-erate this process). With so many airplanesin the pipeline, the supply will likely beginto outstrip demand and drive lease ratesdown even further. This puts a squeeze onprofits and force marginal firms out of busi-ness. Some orders will be canceled; otherswill be renegotiated.


This example makes it clear that piecesof the airline leasing industry have oper-ated within a seesaw structure. Althoughthe extended period of air traffic growthkept demand ahead of supply for severalyears, it did not change the nature of thedelays in the supply line. Whenever thesupply adjustments bring the seesaw backdown, airplane leasing companies will bein for a bumpy landing.

SUMMARY

The balancing loop with delay structure isat once simple and complex: simple, becauseit seems to be an innocuous single loopstructure that is easy to comprehend; com-plex because the resulting behavior is nei-ther simple nor easily predictable. Thedelays in a typical system are rarely consis-tent or well known in advance, and thecumulative effects are usually beyond thecontrol of any one person or firm. •


G U I D E L I N E S F O R D R AW I N GC A U S A L L O O P D I A G R A M S

T O O L B O X

he old adage “if the only tool youhave is a hammer, everything begins

to look like a nail” can also apply to lan-guage. If our language is linear and static,we will tend to view and interact with ourworld as if it were linear and static. Takinga complex, dynamic, and circular worldand linearizing it into a set of snapshotsmay make things seem simpler, but we maytotally misread the very reality we wereseeking to understand. Making such inap-propriate simplifications “is like putting onyour brakes and then looking at yourspeedometer to see how fast you weregoing,” says Bill Isaacs of DIA•logos.

ART ICULAT ING REAL I TY

Causal loop diagrams provide a languagefor articulating our understanding of thedynamic, interconnected nature of ourworld. We can think of them as sentencesthat are constructed by linking together keyvariables and indicating the causal relation-ships between them. By stringing togetherseveral loops, we can create a coherent storyabout a particular problem or issue.

Following are some more generalguidelines that should help lead youthrough the process:• Theme selection. Creating causal loopdiagrams is not an end unto itself, but partof a process of articulating and communi-cating deeper insights about complex issues.It is pointless to begin creating a causal loopdiagram without having selected a theme orissue that you wish to understand better.“To understand the implications of chang-ing from a technology-driven to a market-ing-oriented strategy,” for example, is abetter theme than “To better understandour strategic planning process.”

T


• Time horizon. It is also helpful to deter-mine an appropriate time horizon for theissue—one long enough to see the dynamicsplay out. For a change in corporate strategy,the time horizon may span several years,while a change in advertising campaignsmay be on the order of months.

Time itself should not be included as acausal agent, however. After a heavy rain-fall, a river level steadily rises over time, butwe would not attribute it to the passage oftime. You need to identify what is actuallydriving the change. In computer chips,$/MIPS (million instructions per second)decreased in a straight line in the 1990s. Itwould be incorrect, however, to draw acausal connection between time and$/MIPS. Instead, increasing investments andlearning curve effects were likely causalforces.• Behavior over time charts. Identifyingand drawing out the behavior over time ofkey variables is an important first steptoward articulating the current understand-ing of the system. Drawing out futurebehavior means taking a risk—the risk ofbeing wrong. The fact is, any projection ofthe future will be wrong, but by making itexplicit, we can test our assumptions anduncover inconsistencies that may otherwisenever get surfaced. For example, drawingprojections of steady productivity growthwhile training dollars are shrinking raisesthe question, “If training is not driving ourgrowth, what will?” The behavior overtime diagram also points out key variablesthat should be included in the diagram,such as Training Budget and Productivity.Your diagram should try to capture thestructure that will produce the projectedbehavior.

P E G A S U S C O M

• Boundary issue.How do you knowwhen to stop adding to your diagram? Ifyou don’t stay focused on the issue, youmay quickly find yourself overwhelmed bythe number of connections possible.Remember, you are not trying to draw outthe whole system—only what is critical tothe theme being addressed. When indoubt, ask, “If I were to double or halvethis variable, would it have a significanteffect on the issue I am mapping?” If not,it probably can be omitted.• Level of aggregation.How detailedshould the diagram be? Again, the levelshould be determined by the issue itself.The time horizon also can help determinehow detailed the variables need to be. If thetime horizon is on the order of weeks (fluc-tuations on the production line), variablesthat change slowly over a period of manyyears may be assumed to be constant (suchas building new factories). As a rule ofthumb, the variables should not describespecific events (a broken pump); theyshould represent patterns of behavior(pump breakdowns throughout the plant).• Significant delays.Make sure to identifywhich (if any) links have significant delaysrelative to the rest of the diagram. Delaysare important because they are often thesource of imbalances that accumulate inthe system. It may help to visualize pres-sures building up in the system by viewingthe delay connection as a relief valve thateither opens slowly as pressure builds oropens abruptly when the pressure hits acritical value. An example of this might bea delay between long work hours andburnout: After sustained periods of work-ing 60+ hours per week, a sudden collapsemight occur in the form of burnout. •

M U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

P E G A S U S C O M M U N I C A T I O N S , I N C . W W W . P E G A S U S C O M . C O M S Y S T E M S T H I N K I N G T O O L S 1 9

1 Use nouns when choosing a variable name. Avoid verbs and action phrases, because theaction is conveyed in the loop’s arrows. For example, “Costs” is better than “Increasing

Costs,” because a decrease in Increasing Costs is confusing. The sign of the arrow (“s” for sameor “o” for opposite) indicates whether Costs increase or decrease relative to the other variable.

G U I D E L I N E E X A M P L E

Litigation

Increasing Costs

Costs

Rewards

State of Mind

Happiness

Demand

Contraction

Growth

Production Output

Production Pressure

s

s

oStress

Quality, etc.

Quality

Actions to Improve Quality

B1

Quality

B2 Gap in QualityActions to

Improve Quality

ss

s

o

os

Desired Quality

s

s

s

s

s

Perceived Quality

Desired Quality

R1

Actual Quality

Gap inQuality

Actions to ImproveQuality o

B2

Delay

o

sStress Coping

StrategiesB

oo

o

s

s

Stress AlcoholUse

Productivity Health

B1

R2

Demand Quality

Demand QualityProduction Pressure

o

s o

BankFailures

oo

o

o

Withdrawals from Banks

Solvency Depositors’ ConfidenceR

GE

NE

RA

LT

IPS

LOO

PC

ON

ST

RU

CT

ION

SE

LEC

TIN

GV

AR

IAB

LEN

AM

ES

10A shortcut to determining whether a loop is balancing or reinforcing is to count thenumber of “o’s” in the loop. An odd number of “o’s” indicates a balancing loop (i.e., an

odd number of U-turns keeps you headed in the opposite direction); an even number or no“o’s” means it is a reinforcing loop. CAUTION: After labeling the loop, you should alwaysread through it to make sure the story agrees with your R or B label.

9 If a link between two terms requires a lot of explanation to be clear, redefine the vari-ables or insert an intermediate term. Thus, the relationship between “Demand” and

“Quality” may be more obvious when “Production Pressure” is inserted between them.

8 Actions almost always have different long-term and short-term consequences. Drawlarger loops as they progress from short- to long-term processes. Loop B1 shows the

short-term behavior of using alcohol to combat stress. Loop R2, however, draws out the long-term consequences of this behavior, showing that it actually increases stress.

7 If a variable has multiple consequences, start by lumping them into one term whilecompleting the rest of the loop. For example, “Coping Strategies” can represent many

different ways we respond to stress (exercise, meditation, alcohol use, etc.).

6 Distinguishing between perceived and actual states, such as “Perceived Quality”versus “Actual Quality,” is important. Perceptions often change slower than reality does,

and mistaking the perceived status for current reality can be misleading and create undesir-able results.

5 All balancing loops are goal-seeking processes. Try to make explicit the goals drivingthe loop. For example, Loop B1 may raise questions as to why increasing “Quality”

would lead to a decrease in “Actions to Improve Quality.” By explicitly identifying “DesiredQuality” as the goal in Loop B2, we see that the “Gap in Quality” is really driving improve-ment actions.

4 Think of the possible unintended consequences as well as the expected outcomes forevery course of action included in the diagram. For example, an increase in “Production

Pressure” may increase “Production Output,” but it may also increase “Stress” and decrease“Quality.”

3 Whenever possible, choose the more “positive” sense of a variable name. For example,the concept of “Growth” increasing or decreasing is clearer than an increase or decrease

in “Contraction.”

2 Use variables that represent quantities that can vary over time. It does not make sense tosay that “State of Mind” increases or decreases. A term like “Happiness,” on the other

hand, can vary.

S Y S T E M S T H I N K I N G T O O L S P E G A S U S C O M M U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 02 0

S Y S T E M S A R C H E T Y P E S AT A G L A N C E

T O O L B O X

In a “Drifting Goals” archetype, a gapbetween the goal and current reality canbe resolved by taking corrective action(B1) or lowering the goal (B2). Thecritical difference is that lowering thegoal immediately closes the gap, whereascorrective actions usually take time. (SeeThe Systems Thinker, October 1990.)

• Drifting performance figures are usu-ally indicators that the “Drifting Goals”archetype is at work and that real cor-rective actions are not being taken.

• A critical aspect of avoiding a potential“Drifting Goals” scenario is to determinewhat drives the setting of the goals.

• Goals located outside the system will beless susceptible to drifting goals pressures.

In the “Escalation” archetype, oneparty (A) takes actions that are per-ceived by the other as a threat. Theother party (B) responds in a similarmanner, increasing the threat to A andresulting in more threatening actionsby A. The reinforcing loop is tracedout by following the outline of the fig-ure-8 produced by the two balancingloops. (See The Systems Thinker,November 1991.)

To break an escalation structure, ask thefollowing questions:• What is the relative measure that pitsone party against the other and canyou change it?

• What are the significant delays in thesystem that may distort the true natureof the threat?

• What are the deep-rooted assumptionsthat lie beneath the actions taken inresponse to the threat?

In a “Fixes That Fail” situation, aproblem symptom cries out for resolu-tion. A solution is quickly imple-mented that alleviates the symptom(B1), but the unintended consequencesof the “fix” exacerbate the problem(R2). Over time, the problem symptomreturns to its previous level or becomesworse. (See The Systems Thinker,November 1990.)

• Breaking a “Fixes that Fail” cycle usu-ally requires acknowledging that thefix is merely alleviating a symptom,and making a commitment to solvethe real problem now.

• A two-pronged attack of applying thefix and planning out the solution willhelp ensure that you don’t get caughtin a perpetual cycle of solving yester-days “solutions.”

In a “Growth and Underinvestment”archetype, growth approaches a limitthat can be eliminated or pushed intothe future if capacity investments aremade. Instead, performance standardsare lowered to justify underinvestment,leading to lower performance whichfurther justifies underinvestment.(See The Systems Thinker, June/July1992.)

• Dig into the assumptions which drivecapacity investment decisions. If pastperformance dominates as a consider-ation, try to balance that perspectivewith a fresh look at demand and thefactors that drive its growth.

• If there is potential for growth, buildcapacity in anticipation of futuredemand.

A R C H E T Y P E D E S C R I P T I O N G U I D E L I N E S

Drifting Goals

Escalation

Fixes That Fail

Growth and Underinvestment

Actual CorrectiveAction

Pressure toLower Goal

Gap

B2

B1

s

s

s

s

o

o

Goal

Delay

s

o

B2

s

Activityby A

Threatto A

A’s Result

ss

ss o

B1 Quality of A’s PositionRelative to B’s

B’s Result

Activity by B

Threatto B

FixProblemSymptom

UnintendedConsequence s

s

s

oB1

R2Delay

Capacity Perceived Needto Invest

Performance

B2

B3

s

s

s

s

os

Demand

Delay Delay

Investmentin Capacity

R1

PerformanceStandard

o

GrowthEffort

s

s


In a “Limits of Success” scenario, con-tinued efforts initially lead toimproved performance. Over time,however, the system encounters alimit which causes the performance toslow down or even decline (B2), evenas efforts continue to rise. (See TheSystems Thinker, December1990/January 1991.)

• The archetype is most helpful when it isused well in advance of any problems,to see how the cumulative effects ofcontinued success might lead to futureproblems.

• Use the archetype to explore questionssuch as What kinds of pressures arebuilding up in the organization as aresult of the growth?

• Look for ways to relieve pressures orremove limits before an organizationalgasket blows.

In a “Shifting the Burden,” a problem is“solved” by applying a symptomaticsolution (B1), which diverts attentionaway from more fundamental solutions(R3). (See The Systems Thinker,September 1990.) In an “Addiction”structure, a “Shifting the Burden”degrades into an addictive pattern inwhich the side-effect gets so entrenchedthat it overwhelms the original problemsymptom. (See The Systems Thinker,April 1992.)

• Problem symptoms are usually easier torecognize than the other elements of thestructure.

• If the side-effect has become the problem,you may be dealing with an “Addiction”structure.

• Whether a solution is “symptomatic” or“fundamental” often depends on one‘s per-spective. Explore the problem from a dif-fering perspective in order to come to amore comprehensive understanding ofwhat the fundamental solution may be.

In a “Success to the Successful”archetype, if one person or group (A)is given more resources, it has ahigher likelihood of succeeding thanB (assuming they are equally capable).The initial success justifies devotingmore resources to A, and B’s successdiminishes, further justifying moreresource allocations to A (R2). (SeeThe Systems Thinker, March 1992.)

• Look for reasons why the system was setup to create just one “winner.”

• Chop off one half of the archetype byfocusing efforts and resources on onegroup, rather than creating a “winner-take-all” competition.

• Find ways to make teams collaboratorsrather than competitors.

• Identify goals or objectives that definesuccess at a level higher than the individ-ual players A and B.

In a “Tragedy of the Commons” struc-ture, each person pursues actionswhich are individually beneficial (R1and R2). If the amount of activitygrows too large for the system to sup-port, however, the “commons”becomes experiences diminishing ben-efits (B5 and B6). (See The SystemsThinker, August 1991.)

• Effective solutions for“Tragedy of theCommons” scenario never lie at theindividual level.

• Ask questions such as: “What are theincentives for individuals to persist intheir actions?” “Can the long-term col-lective loss be made more real andimmediate to the individual actors?”

• Find ways to reconcile short-termcumulative consequences. A governingbody that is chartered with the sustain-ability of the resources limit can help.

A R C H E T Y P E D E S C R I P T I O N G U I D E L I N E S

Limits to Success

Shifting the Burden/Addiction

Success to the Successful

Tragedy of the Commons

s

B2Efforts

o

R1

s

Performance LimitingAction

s

s

Constraint

Side-EffectProblemSymptom

B1

B2

s

s

o

SymptomaticSolution

FundamentalSolution

Delay

s

o

oR3

Resourcesto B

Allocation to AInstead of B

s Successof B

R2

Resourcesto A

Successof A

o

o

R1

s

s s

R4

R3

R2

R1 B5

B6

Total Activity

A’s Activity

B’s Activity

Net Gains for B

R2

s

s

s

s

s

s

s

s o

o

o sDelay

Net Gains for A

Resource Limit

Gain per Individual Activity


GRAPH ICAL FUNCT ION D IAGRAM

STRUCTURE - BEHAV IOR PA IR

tructural thinking tools can help us become even more explicit about the struc-tures that create the dynamic behaviors we are trying to understand. “From

Causal Loops to Graphical Functions: Articulating Chaos” (p. 24) and “GraphicalFunctions: ‘Seeing’ the Full Story” (p. 26) describe how graphical function diagramscan easily represent nonlinear relationships. These relationships characterize thenature of most interconnections in complex systems (as opposed to the simple, linearrelationships that are often assumed).

The second half of this section focuses on structure-behavior pairs. Accumulatorsand flows provide a rigorous framework for representing systemic structures in amore precise way than through causal loop diagrams alone. They can better representa system’s nonlinearity, as well as distinguish between things that accumulate (water ina bathtub) and things that flow (water flowing through a faucet). The articles on accu-mulators and flows (p. 28–37) show how these concepts add further precision to ourthinking and understanding about the link between structure and behavior.

S

P A R T I I I

S T R U C T U R A L T H I N K I N G T O O L S

F R O M C A U S A L L O O P S T OG R A P H I C A L F U N C T I O N S :A R T I C U L A T I N G C H A O S

T O O L B O X

S A V I N G S A N DS P E N D I N G L O O P S

n Chaos: Making a New Science(Penguin Books, New York), James

Gleick describes a relatively new branch ofscience that has profound implications forhow we view our world. Chaos, simplyput, is the science of seeing order and pat-tern where formerly only the random,erratic, and unpredictable had beenobserved. In a way, systems thinking alsodeals in the science of chaos. Diagramssuch as causal loops, accumulators andflows, and graphical functions are all waysof extracting the underlying structurefrom the “noise” of everyday life.

RELAT ING BEHAV IOR TOSTRUCTURE

Both systems thinking and chaos insist thatreal-world phenomena need to bedescribed in terms that reflect our intu-ition. Writing partial differential equationsto describe clouds, for example, misses thepoint, because we don’t perceive clouds in

I


Savings

InterestEarned

s

s

R1

$

If we begin to explore our savings account “systea causal loop diagram, we see that an increase inlead to more interest earned, which increases ouance still further (left). The graph of this behaviorlook something like the exponential growth curve

S A V I N G S L O O P

InterestEarned

so

s

R1

B2

Spending s

Savings

A balancing loop explains the slowing growthof the savings account: As our savingsincreases, we are more likely to increasespending, which will reduce our savings.

that way. It is not enough to have a modelthat reproduces some real-world phenom-ena if we cannot identify the structuresthat produce the behavior of the actualsystem. That is why systems thinking dia-grams focus on capturing reality in a for-mat that taps into our intuitiveunderstanding of the systems in which wemanage and live.

FROM CAUSAL LOOPS TOGRAPH ICAL FUNCT IOND IAGRAMS

To see how a range of systems thinkingtools can help capture the structure of asystem at increasing levels of detail, let’slook at a system we are all familiar with—a savings account. If we plot out the struc-ture of a savings account using a causalloop diagram (see “Savings Loop”), we seethat an increase in savings will lead tomore interest earned, which increases oursavings balance still further. The graph of

the behavior over timewould look somethinglike the exponentialgrowth curve shown onthe right of the diagram.

“Wait a minute,” youmay protest, “I don’tknow whose bankaccount that is, but it cer-tainly doesn’t look likemine!” That’s true—rarely is a system so sim-ple in real life; nor arebank accounts that well-behaved. There are usu-ally many other factors

Time

m” by drawingsavings will

r savings bal-over time will(right).

P E G A S U S C O M

involved. The question of how many fac-tors to include always depends on the pur-pose of examining the system. Since thedetails of any system are infinitely complex,it is futile to strive to “model the system.”In our sample case, the purpose is to repre-sent as concisely as possible the importantfactors that affect the balance of a typicalsavings account, so we want to look at sav-ings, income, interest earned, and spending(see “Savings and Spending Loops”). If wewere only interested in capturing the factthat there is a balancing loop that explainsthe slowdown in the growth of our savingsaccount, we could stop at this point. On theother hand, if we want to be more explicitabout the structure behind the behavior, weneed to translate our diagram into accumu-lators and flows.

M U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

terestarned

Interest Rate

Fixed Expense

ncome

Savings

Spending

ss

s

R3

This Graphical Function Diagram capturesour savings policy in an intuitive way by map-ping out the relationship between savingsand discretionary expenses.

S A S A N A C C U M U L AT O R

Savings

Spending

s

500

200000Savings

Disc

r.Ex

pens

es

d flows add more detail and understanding todiagram by differentiating between those vari-ram that “accumulate” (savings) and those thate system (income and spending).

S AV I N G S P O L I C YG R A P H I C A L F U N C T I O N

D I A G R A M

ACCUMULATORS ANDFLOWS

When we translate CLDs into accumula-tors and flows, we are becoming even moreprecise about the structures producing thedynamics. The bathtub as a metaphor foraccumulations helps us visualize how con-cepts as diverse as savings, pollution, cus-tomers, and corporate reputation share asimilar underlying structure (see“Accumulators: Bathtubs, BathtubsEverywhere,” p. 30).

Accumulators and flows add moredetail and understanding to our causal loopdiagram by differentiating between thosevariables in the diagram that “accumulate”(our savings balance) and those that just“flow” through the system (income andspending). In the “Savings as anAccumulator” diagram, we can visually seemoney flowing into and out of savings inthe form of income and spending. Moreimportantly, we can relate to this structureintuitively because we experience money interms of flows and accumulators (or lackthereof!).

GRAPH ICAL FUNCT IONS :MAPP ING POL IC I ES

So now we have a pretty good idea of boththe basic dynamic behavior of the savingsaccount, and a feel for the importantinflows and outflows. But our model is stillpretty elementary. Suppose now youwanted to go a little further and use a sys-tems diagram for describing your family’spolicy for managing your savings. “Ourdiscretionary spending depends on howmuch savings we have,” you explain. “Ifthe balance in our savings account is below$5000, we don’t spend a dime. As our sav-ings rise above $5000, we may increase dis-cretionary spending by, say, $15-20 permonth. If our savings tops $10,000, thenwe’re likely to spend several hundred dol-lars a month. But in any case, we don’t seeourselves spending more than $500 per


month on discretionary expenses.”Graphical functions allow us to expand

our exploration of a system to include poli-cies and interrelationships between vari-ables. If we tried to capture the savingsplan we described above in an analyticalform, we would have to do quite a bit ofwork in order to come up with a suitableequation. And when we were done, itwould be hard to tell if the equation repre-sented our savings account or the numberof widgets on sale at Wal-Mart. The truthis, most of us don’t think in abstract mathe-matical concepts, but in images and struc-tures grounded in our everyday experience.That is why graphical functions are useful.They capture policies in an intuitive waythrough a simple graph that maps out therelationship between one variable in rela-tion to another (see “Savings PolicyGraphical Function Diagram”). In our sav-ings policy plan, for example, we see at aglance that savings has no impact on ourdiscretionary expenses until savings hits$5000. After that, discretionary expensesrise until savings reaches $20,000, at whichpoint they level out at $500.

ART I ST I C MANAGERS

Physicist Mitchell Feigenbaum suggeststhat art is a theory about the way theworld looks to humanbeings. “It’s abundantlyobvious that one doesn’tknow the world aroundus in detail. What artistshave accomplished is real-izing that there’s only asmall amount of stuffthat’s important, and see-ing what it is.”

Whether we recognizeit or not, we are artists aswell, selectively pickingout details of the worldthat we choose to focuson. Those details appear

InE

ISalary

ss

S AV I N G

Accumulators anour causal loopables in the diag“flow” through th


as items on our production reports, finan-cial statements, and customer surveys. Tothe extent that those details do not capturethe core structures that are important, wemay be the unwitting producers of ourown chaos. As one systems thinkingmaxim warns, “It ain’t what you don’tknow that hurts you, it’s what you DOknow that ain’t so.” •


G R A P H I C A L F U N C T I O N S :“ S E E I N G ” T H E F U L L S T O R Y

T O O L B O X

n executive of a large automotivecompany tells the story of two engi-

neers who were arguing about the correctangle of an engine mount. The two hadbeen at it for more than half an hour—oneengineer swearing that the angle was 40degrees while the other fumed that it was50 degrees. After several civil attempts tocorrect each other’s viewpoint, they had juststarted attacking each other’s intelligence,ability, and character when the executivehappened to walk by.

“What axis of reference are youusing?” he asked.

“The vertical, of course!” exclaimedone engineer.

“The horizontal!” said the other.Both stopped in amazement as they

realized they had been saying the same thing!Because they had not established a common

A


Sales($)

1000 K

900 K

800 K

0 1 2 3 4Delivery Delay (weeks)

Sales = 1000 - 25 * Delivery Delay

S A L E S V S . D E L I V E R YD E L AY

Scatter charts plot one variableʼs dataagainst another and answer the question,“What happened historically?” The view isretrospective.

frame of reference for their discussion, eachhad assumed the other’s viewpoint was wrong.

I ND I V I DUAL WORLDS

The story of the two engineers points outan age-old communication problem. Eachof us carries our own set of assumptionsabout reality—our own individual pictureof the world. Oftentimes, we mistakenlyassume that our viewpoint is the only wayof looking at a situation. Both engineers,for example, believed the other person’sposition was based on the same axis astheir own—they never even questioned it.If we don’t acknowledge our assumptionsat the outset of a discussion, we risk expe-riencing the same frustrations as the twoengineers.

In many instances, spoken languagecan be a hindrance rather than a help incommunicating our mental pictures ofreality because words, unlike pictures, donot force us to be explicit when explainingour reasoning. Graphics, because they canrepresent ideas more clearly, can be amuch more powerful and effective meansof communication (see “Systems Thinkingas a Language,” p. 6). Trite as it maysound, the saying “a picture is worth athousand words” still holds true. Had thetwo engineers simply drawn two axes anda line, they would have saved a great manyangry words. When the issue is more com-plex than a single angle, the use of graph-ics can become even more important forreaching a shared understanding.

Using graphical function diagrams(GFDs), it is much easier to capture howtwo variables relate in a format that is con-cise and invites others to share their own

P E G A S U S C O M

perspectives. Graphical functions can helpus go beyond merely observing correla-tional relationships (when X happens, Yhappens) to exploring our understandingof the causal connection between two vari-ables (X causes Y). In constructing GFDs,one should follow the 60 percent rule—it’sbetter to get it 60 percent right veryquickly and spend time modifying it thanspend a great deal of effort trying to get it100 percent right the first time.

GRAPH ICAL FUNCT IONSVS . SCATTER CHARTS

Graphical functions are best described byfirst establishing what they are not.Although they may look similar, graphicalfunctions are not the same as scatter charts,which plot one variable’s data againstanother’s. If we were to look at the rela-tionship between sales and delivery delayusing a scatter chart, we would plot somedata points and then draw a regression linethrough them (see “Sales vs. DeliveryDelay”).

From the scatter chart, we can see thatin weeks one through four, sales fall by$25K for each one-week delay. We can thenextrapolate beyond the historical data topredict that a five-week delay will result inan additional $25K drop in sales. In general,scatter diagrams answer the question,“What happened historically, was there acorrelation, and based on that informationwhat can I expect to happen in the future?”They tend to be retrospective.

A graphical function, on the otherhand, is very much prospective in nature.By including the full spectrum of possiblevalues, GFDs can help you see beyond the

M U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

20Delivery Delay (weeks)

nction diagram conveys a much richer descrip-relationship between delivery delay and sales

catter chart based on historical data. Looking atelivery delay on sales (bottom) allows you torelative effects of different variables, to see whichnt driver.

D E L I V E R Y D E L AYI C A L F U N C T I O N D I A G R A M

20Delivery Delay (weeks)

4 10

aximum Advantage

Maximum Disadvantage

Neutral Zone

C T O F D E L I V E R Y D E L AYO N S A L E S

historical range of operating values andask, “Given my understanding of the sys-tem, what do I think will happen at eachpossible point?”

CREAT ING A GFD

A graphical function diagram can helpexplicate your (or a team’s) mental modelof the relationship between two criticalvariables. Unlike behavior charts, GFDsdo not show how variables change overtime, but how two variables interrelate. Tocreate a GFD, it is best to begin by answer-ing the following questions:• What do we know from the outsetabout the causal relationship between thesetwo variables?• Are there any “neutral zones” wherethe variable on the y-axis is not affected bychanges in the x-variable?• What are the extreme values that bothvariables can assume?If we looked at the sales and delivery delayexample using a GFD, we would start byasking what we think the general natureof the relationship is between the two vari-ables—is it flat, is it upward sloping, or isit downward sloping? With most products,longer delivery delays mean lower sales, sowe can assume that the relationship wouldslope downward.

Using available historical data and pastexperience, we can then take a first cut atidentifying a neutral zone where sales maybe insensitive to differences in the length ofthe delay (see “Delivery Delay GraphicalFunction Diagram”). Past experience maysuggest that sales will increase steadily asthe delay falls below four weeks. A sam-pling of customer contacts may tell us thatthere is not a whole lot of differencebetween 3.5 and 4.5 weeks. On the otherhand, past market research also tells us thatif the delay grows greater than five weeks,sales will fall dramatically. Looking fur-ther, we realize that even in the extremecase of a 20-week delay, $200K of sales will


still come from captive customers who havenowhere else to turn in the short term.

BU I LD ING SHAREDUNDERSTAND ING

The resulting diagram is a concise causalhypothesis that states that customers willreward shorter delays with slightly higherorders, but will severely penalize delaysthat extend beyond an acceptable range.The GFD conveys a muchricher description about therelationship between deliv-ery delay and sales than ascatter chart based on histor-ical data. The diagram helpsvisualize the full range ofimplications and minimizesthe danger of remainingmyopically focused on a nar-row band of possible out-comes. Developing thediagram as a group can alsohelp surface differing mentalassumptions about the poten-tial impact of deterioratingor improving delivery per-formance (remember theengineers!).

Sometimes it is helpful toconvert the relationship into amore general form where they-variable is converted to an“effect-of” variable. Insteadof “Sales” on the y-axis, forexample, we would have“Effect of Delivery Delay onSales” (see graph), whichshows that a 3.5 to 4.5 weekdelay has no effect, shorten-ing the delay nets us a maxi-mum gain of 5 percent (1.05times the sales number wewould have obtained if wewere in the neutral zone),and lengthening the delay to20 weeks can choke sales by

Sales ($)1000 K

900 K

800 K

0

700 K

600 K

500 K

400 K

300 K

200 K

A graphical fution about the(top) than a sthe effect of dcompare theis the domina

G R A P H

Effe

ctof

Deliv

ery

Dela

yon

Sale

s 1.4

1.2

0

1.0

.8

.6

.4

.2

0

M

E F F E


as much as 80 percent.The “Effect-of” version of the GFD

focuses attention on the relative impact ofthe delivery delay on sales instead of on thespecific numbers themselves. In this way, wecan compare across different variables, suchas the relative effects of quality on sales vs.marketing spending on sales, and makeexplicit our understanding of which factor isthe dominant driver. •


S T R U C T U R A L T H I N K I N G :T H E W O R L D A C C O R D I N G T OA C C U M U L A T O R S A N D F L O W S

T O O L B O X

vice president of a major U.S. man-ufacturer once questioned whether

today’s rapid pace of change means that allour old tools and ways of managing are nowinadequate. “Are we doomed to keep onthrowing out our current tools and practicesas soon as the next wave of innovationscomes along?” he asked.

The answer is . . .“it depends.” Itdepends on the underlying theory on whichthe current tools and methods are based. Ifone’s management practices are based ontransient or situation-specific phenomena,they are likely to require revision wheneverthe circumstances change. If, on the otherhand, they are based on a structural under-standing, the situation may change, but thetools will still apply.

WHERE ARE THE COWS?

Barry Richmond of High Performance

A


Maturation_T

Annual

Births

Calves

Maturatio

If we wanted to create a structural representation oftral accumulator “milk cows.” Milk production is deteamount of milk per cow. To create our hypotheticalenter zero for the number of milk cows. The resultin

M I L K P R O D U C T

Systems tells this story: “While perusing awell-known economic journal, I cameacross an article which described a modelthat had been constructed to forecast U.S.milk production. The model was of theY=f(Xi) form [Y = Y0 + a1X1 + a2X2 +...+anXn], where the Xi’s included such thingsas: last year’s milk production, interest rates,spending on cattle feed, GNP growth, andother macroeconomic factors. As the articledetailed, the model performed quite well asa predictive device—at least in terms of itsability to ‘track history.’ The obvious thingabout this model, that would bother bothdairy farmers and people who were partialto operational specifications, is: ‘where’s thecows?!’ Simply stated, if you’ve got nocows, you’ve got no milk! Crude, but true.”

How does all this talk about cows relateto our vice president’s question? Well,imagine that an epidemic swept over the

P E G A S U S C O M

Milk_per_Cow

ime

_Milk_Prodn.

n

Milk_Cows

MC_Deaths

milk production, we would begin with the cen-rmined by the number of cows and the

scenario of an epidemic, we would simplyg annual milk production would also be zero.

I O N M O D E L

country and killed all the cows. Whatwould the above model predict for nextyear’s milk production? The answer wouldmost likely look a lot like the number forlast year’s milk production, which is clearlyincorrect. The model must be abandoned.

“Unfair,” you might say. “It’s not thatthe model is wrong. It’s just that the worldhas changed dramatically since the modelwas originally built and the changes mustnow be added.” But what has reallychanged? Yes, the cows are now dead, butthe basic fact that milk comes from cows,and that without cows there can be no milk,is as true now as it was before the mass dec-imation. From a structural perspective, thenature of the world has not changed at all.The model was inadequate because it wasbased on situation-specific data that hasnow changed.

STRUCTURAL TH INK ING

When we look at the world through astructural lens, we are interested in under-standing how things actually work. We areless interested in correlational relationshipsand more interested in the causal structuresthat produce the observed behavior. This isnot to say that nonstructural models aren’tvaluable. Regression models, for example,have many applications and are useful foridentifying correlation, explaining sourcesof variance, and extrapolating from histori-cal data. Those models are inadequate,however, for gaining insight into how a sys-tem actually operates.

If we were to look at the milk produc-tion model from a structural viewpoint, wewould start with the basic fact that milk

M U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

Annual_Milk_Prodn.Annual_Sales

Training TimeMaturation_Time

BirthsHires

Calves Trainees

MaturationTraining_Rate

Milk_Cows Sales_Managers

MC_Deaths SM_Quits

Milk_per_Cow Sales_per_SM

If we replace the names of the variables in the “Milk Production Model” with those listed above,we can create a model that explores sales growth. The same generic resource developmentstructure can be used to describe both processes.

S A L E S G R OW T H M O D E L

comes from cows. Therefore, cows are thecentral accumulator in the model—thenumber of cows accumulates over time, ascows are born, mature, and become milkcows (see “Milk Production Model”).

Depending on the scope of our study,we may be interested in representing thelifecycle of all cows, or just milk cows. Inthis case, we will focus our attention on theflow of cows from birth through maturityinto the milk cow accumulator. The annualmilk production is then determined by thenumber of milk cows at any one time andthe amount of milk per cow. Of course,there are many other factors that affectmilk production, such as food supplies,milk demand, and dairy farmers. Thesefactors could also be added to our diagramin the form of additional accumulators andflows.

The resulting model can then be simu-lated on a computer to see how annualmilk production behaves over time. To cre-ate our hypothetical epidemic scenario, forexample, we would simply put zero for thestock of cows. In that event, the annualmilk production would also equal zero.Because this model is tied to the structureof the system, not just historical data, itwould not have to be thrown out even if allof the cows suddenly died.

LEVELS OF EXPLANAT ION

We live in the world of events. As a result,we encounter and navigate through therapids of life on an event-by-event basis.But this does not mean that we must act oneach event as if it were an isolated occur-rence. We can look at patterns of behaviorover time and try to glean lessons from thepast that will improve our ability to handlepresent situations. That is the purpose offorecasting models.

Forecasting models, like the economist’smilk production model described above,attempt to provide information about thefuture by looking at the past. But in many


ways, managing on the basis of forecasts is alot like trying to drive a car by lookingthrough the rearview mirror. When does itwork best? When the road is straight andthere are no obstacles in the way. Whendoes it fail? The rest of the time! Whenusing a forecasting model, you only realizeyou have missed a turn once you see thecliff’s edge behind you and feel the sensa-tion of free fall hit your stomach.

Forecasting provides very little insightinto what actually produces the observedbehavior. Consequently, it allows us toanticipate and react to changes only if theydo not deviate too much from past behav-ior. Models, on the other hand, capture thestructural forces at work and are thereforeless situation-dependent. To come back tothe vice president’s question, structuralthinking provides a more stable basis ofunderstanding that will last even throughtimes of turbulent change.

GENER IC TH INK INGSK I L LS

If we begin to view the world through astructural perspective, another benefit


emerges—the ability to transfer insight.This ability to see similar structures occur-ring in diverse settings is referred to as“generic thinking,” and the structuresthemselves are referred to as “genericstructures.”

For example, if we take the “MilkProduction Model” and substitute “hires”for “births,” “trainees” for “calves,” and“sales managers” for “milk cows,” we cantransform the milk cow model into a modelthat can be used to explore the structuralforces that influence annual sales (see “SalesGrowth Model”). The same genericresource development structure underliesboth models. Although we may debatewhether it takes longer to produce a milkcow or a sales manager, we can both agreethat the structure of both processes is fun-damentally the same.

For further reading about structural thinking and theother critical thinking skills included under the systemsthinking umbrella, see Barry Richmond’s TheThinking in Systems Thinking: Seven EssentialSkills (Pegasus Communications, 2000).

•


A C C U M U L A T O R S : B A T H T U B S ,B A T H T U B S E V E R Y W H E R E

T O O L B O X

hen’s the last time you actually tooka real, honest-to-goodness bath? If

you are like most people, it has probablybeen quite a while. We live in the world ofquick showers and instant breakfasts. Yet,it wasn’t too long ago when taking bathswas part of our normal daily routine. Theshift from baths to showers marked a farmore deeper change in our thinking thanmerely a change in personal hygienehabits.

When we run the bathwater, we canvisually see the water accumulating in thetub (see “Bathtubs and Accumulators”). Weknow we have to keep an eye on the waterlevel so it won’t overflow. When we takeshowers, however, the accumulation processis virtually eliminated. Water flows out ofthe showerhead, over our bodies, and outthe drain. Where does the water go? Wehardly give it any thought.

W


Bathtub

Faucet

The core building blocks of dynamic thinking toolsgous elements in the structural thinking set of tools(or stock) is represented by a rectangular box, anda directional arrow, a valve, and a circle. The circlea visual reminder that the dynamics of the two arestant flow from the circle to the box as indicated inrise in the water level. No other behavior is possibl

B AT H T U B S A N D A

SHOWERHEAD VS .BATHTUB TH INK ING

Taking showers disconnects us from experi-encing one of nature’s most basic struc-tures—accumulators. Lakes and ponds areaccumulators of various water flows. Globalwarming has been attributed to the cumula-tive effects of burning fossil fuels. Plants areaccumulators of energy and nutrition.Displacement, velocity, and acceleration canbe represented in terms of accumulators.That is, displacement represents the accu-mulation of past velocity, and velocity is anaccumulation of past acceleration.

If we use showerhead thinking, we areless conscious of accumulations. Flows ofmaterials such as water, fuel, or energysimply “go away” somewhere. But from abathtub—or systems—perspective, there isno “away.” Everything accumulates some-where. Forgetting about that “somewhere”

P E G A S U S C O M M

Flow(faucet flow)

Time

Accumulator(bathtub)

Time

are reinforcing and balancing loops. The analo-are accumulators and flows. An accumulatorthe flow (or rate) is represented by a pipe withand the box each contain a timeline graph asintimately connected. For example, the con-the diagram must produce the straight lineare for that structure as it is drawn.

C C U M U L AT O R S

can lead to disastrous results.When Just-in-Time (JIT) manufactur-

ing first hit the U.S., for example, manycompanies implemented it using a shower-head perspective. The basic concept of JITis to manage a steady flow of materialsthrough a factory with minimal accumula-tions of inventory at each step. Many com-panies that instituted JIT tried tominimize their own accumulations bydemanding that their suppliers providethem with materials just when theyneeded them and not any sooner.

The problem with the above approach,of course, is that the flow of materials hasto accumulate somewhere, and it was accu-mulating in the suppliers’ warehouses. TheJIT flow was accomplished by shifting theaccumulations to suppliers, severely strain-ing the relationship between suppliers andmanufacturers. Bathtub thinking wouldhave highlighted the fact that unless theentire flow from raw materials to finalcustomer worked together, there would beundesirable accumulations for somebodyin the system.

I NV I S I B LE BATHTUBS

When’s the last time you actually let abathtub overflow? Probably not in a longtime. Of course, we all know not to let thewater run indefinitely, because the tub hasa limited capacity. The tub’s dimensionsare obvious and so is the rising water line.But suppose the bathtub is invisible, and sois the water once it leaves the faucet. Andsuppose you are not in the bathroom tokeep an eye on the tub—you are offanswering phone calls and dealing withthe latest crisis at the office. How will you

U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

Stressful Events

Time

Stress Level

Time

Work PressureCapacity to Handle Workload

Workload s

oR

Productivity

os

s

Interest

Time

Savings Balance

Time

InterestRate

s

R

s

Savings

InterestEarned

s

s

R

The reinforcing loop of savings and interest can be represented as a causal loop diagram (left)or as an accumulator and flow diagram (right), where you can visualize the flow of interest intosavings.

S T R E S S A C C U M U L AT O R

Increasing work pressure can lead to anincreased number of stressful events, whichadds to the accumulation of stress.

F R O M L O O P D I A G R A M S T OA C C U M U L AT O R S A N D F L O W S

know when the bathtub is getting full oralready overflowing?

Flows are easy to keep track of becausethey involve action, and actions are easy tomeasure—how many products to ship,how many people to hire, for example.Some accumulations are also very visible,such as order backlogs or bulging invento-ries. There are, however, many accumula-tions that are not tangible but nonethelessvery real. These possess the same behav-ioral characteristics as physical accumula-tors and flows, but they are like invisiblebathtubs—we can never tell for surewhether they are overflowing or not.

I D ENT I FY INGACCUMULAT IONS

So how can you locate the “invisible bath-tubs” lurking in your company? For everyflow (action, decision, policy), try to figureout what, if anything, is accumulatingand what are the implications of thoseaccumulations.

For example, as workload outstripscapacity and work pressures become exces-sively high (see “Stress Accumulator”), youshould question whether those pressuressimply come and go or whether theireffects are accumulating somehow. Forexample, extra pressure may generatemore stressful events, which will accumu-late into increasing levels of stress. Highstress levels will then lead to lower produc-tivity, which further reduces work capacityand leads to more stressful events. Thisreinforcing loop of accumulating stress isintangible, yet all too real for many people.

If you look at the situation from theaccumulator viewpoint and trace out thereinforcing loop, it becomes clear why typ-ical stress reduction efforts do not workvery well. Each round of stressful eventsproduces more stress, like compoundinterest in a savings account. And copingmechanisms are like savings with-


drawals—unless you withdraw more thanyou are earning in interest, the accountbalance never goes down. Likewise, if thestress “withdrawal” rate (coping mecha-nisms) are not exceeding the stress “inter-est” rate (stressful events), then the bestyou can do is learn to live with the higherstress level. From the accumulator perspec-tive, the high-leverage action would be to“close the account” by reducing or elimi-nating the real source of stress.

LOOP D IAGRAMS VS .ACCUMULATORS ANDFLOWS

If causal loop diagrams and systemsarchetypes are such powerful tools, why dowe need to bother with accumulators andflows? Both tools have their uniquestrengths. Tools like systems archetypescapture and communicate dynamic issuesin a concise way, but they do not provide adetailed representation of the structureproducing the dynamics.

There are cases when tracing througha loop diagram can be confusing. Forexample: “Savings and interest form areinforcing loop where higher savings bal-ance leads to higher interest payments,


which leads to still higher savings (see“From Loop Diagrams to Accumulatorsand Flows”). If we start making with-drawals, the balance goes down and inter-est payments decrease, but savings does notdecrease. It still increases but at adecreased rate.” Sound confusing? That’swhere the accumulator and flow diagramcan help you actually visualize how thatloop works in terms of the flow of moneyinto and out of the account. •


A C C U M U L A T I O N M A N A G E M E N T :A V O I D I N G T H E “ P A C K R A T ”S Y N D R O M E

T O O L B O X

here is a story about a trivia “packrat,” a man who had spent his

entire life memorizing trivia. He knewbaseball statistics of every player in the his-tory of the major league. He had memo-rized the titles, directors, and actors ofhundreds of movies. He knew the name ofevery television show that had ever aired.

But one day he found himself in anawkward predicament—no matter howhard he tried, he could not memorize anymore trivia. He had finally taxed the limitsof his rote memorization capacity. Althoughhe had worked hard at acquiring his stockof trivia throughout his life, he had neverconsidered how he might go about depletingit. He had not learned the fundamentals of

T


Acquisitions Depletions

Gap

sCorrectiveAction

Accumulationo

B

s

Delay

DesiredAccumulation

s

Delay

Accumulation

CorrectiveAction

GapDesiredAccumulations

o

s

s

B

Accumulator management can be viewed sim-ply as a balancing loop with delay (top). Astructural diagram (bottom) reveals that theflows controlling the accumulation are acquisi-tions and depletions.

A C C U M U L AT O RM A N A G E M E N T

accumulator management.

PACK RATS AND NOMADS

Life can in some ways be viewed as anever-ending task of managing variousaccumulators. Our pantries, refrigerators,checking accounts, and closets are amongthe many accumulations we manage daily.

On one end of the accumulation man-agement spectrum is the pack rat whothrows nothing away. On the other end isthe “nomad” who makes a virtue of owningno more than what can be packed into onesuitcase. In between these two extremes liesthe majority of the population who is con-stantly struggling to maintain the right bal-ance between acquisitions and depletions.


New Policies Loss Events

Claims PendPolicyholders

Policies LostInvestmen

Cash In

Retentio

The insurance business can be mapped into a relaaccumulators and flows. If we assign numbers nexpercentage of organizational resources devoted tozationʼs current emphasis.

I N S U R A N C E BA C C U M U L AT I O N

ANATOMY OFACCUMULATORMANAGEMENT

A typical accumulator management struc-ture (AMS) has the following elements: theAccumulation, Acquisitions, Depletions,Desired Accumulation, and a CorrectiveAction (see “Accumulator Management”).In addition, there is almost always somedelay between the Corrective Action and theAcquisition, because it takes time to actuallymemorize data or clear out the closet oncewe have decided to do so.

The accumulator management struc-ture is a generic structure that can repre-sent a wide range of business settingswhere accumulation management is

U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

ing

Claims Adjusted

Claims Settled

Policies Lost

ts

Cash Out

n

tively simple diagram that highlights the majort to each accumulator or flow indicating theit, the diagram can help reevaluate the organi-

U S I N E S S A SM A N A G E M E N T

ntory management system in the Beer Game isam. Understanding the nature and source ofch as the supply line delay above—often playsg accumulations without overcorrecting.

Y L I N E A N D D E L AYT H E B E E R G A M E

Shipments

ply Line

Gap Desired Inventorys

o

s DeliveryDelay

Inventory

B

R

Deliveries

o

important. For example, the insurancebusiness can be mapped into a relativelysimple diagram by focusing on the basicaccumulators and flows (see “InsuranceBusiness as Accumulation Management”).Insurance revolves around managing twomain accumulators—policyholders andinvestments.

If managers assign a number next toeach accumulator and flow in the diagramto represent the percentage of organiza-tional resources devoted to each, the dia-gram can highlight which areas receive thelargest organizational focus. This exercisecan point out any weaknesses in the currentorganizational emphasis—for example,spending too little time trying to retaincurrent policyholders—and reveal ways inwhich the company can better serve itscustomers.

SUPPLY L INES ANDDELAYS

If we had direct and immediate controlover all the elements in the AMS diagram,managing accumulations would be simple:We would calculate the depletion rate, setour desired accumulations accordingly,and implement actions that will immedi-ately result in acquisitions. In our homelife we already pretty much follow thispattern. For example, we plan our meals,decide on an appropriate amount of foodto have on hand, figure out how long itwill be before we run out of certain staples,and go to the grocery store as needed.Unfortunately, things are not that straight-forward when we move into the organiza-tional context.

One of the most challenging aspects ofmanaging accumulations within organiza-tions is captured in one word—delays.Identifying and characterizing the natureand source of delays often plays a critical rolein managing accumulations effectively. A bigpart of the problem is that we usually havevery little control over the supply line delay.


MANAG ING THE “BEERGAME ”

In a production distribution system gamefondly known as the Beer Game, partici-pants are given the task of managing theirown inventory (accumulation) of beer. Eachteam is composed of four players linkedtogether in a structure similar to that repre-sented in the AMS diagram (see “SupplyLine and Delay in the Beer Game”). Withinthat team, each participant must makeordering decisions in order to maintain hisdesired level of inventory.

According to MIT professor JohnSterman, when participants try to manageaccumulations in the Beer Game they usu-ally run into three common problems. First,they typically underestimate the true lengthof the delay from the time they order towhen they receive the beer and then overad-just their orders—even when they are givenfull information about the supply line delays.They do not appear to recognize that theirordering decisions affect the length of thesupply line delay—that is, the more theyorder, the longer it takes to receive the beer.

In addition, he found that when peoplefind it difficult to determine their optimalinventory level, they simply anchor theirdesired inventory on the initial inventoryand adjust from there.This finding high-lights the more gen-eral tendency peoplehave to anchor onpast goals or stan-dards rather thansearch for better ones.

The third obser-vation is that peoplegenerally point to fac-tors outside the systemas being responsiblefor the instabilitiesthey observe in thegame. That is, peopleoffer open loop expla-

The structure of the invesimilar to the AMS diagrdelays in a systems—sua critical role in managin

S U P P LI N

Orders

Sup

Corrective Action

s


nations rather than connecting the dynamicsback to their own decision-making. In fact,the wide oscillations in inventory are actuallygenerated by the decisions they make.

AVO ID ING THE “PACKRAT ” SYNDROME

If you want to avoid the “pack rat” syn-drome, you need to manage the wholeaccumulator management structure andnot just focus on one piece of it. The obser-vations about the difficulties of managingthe Beer Game suggests that you shouldthink through the following questionswhen confronting a typical accumulatormanagement situation: (1) Where are thesupply line delays and how are they chang-ing? (2) What factors are determiningwhat Desired Accumulation should be? (3)How do current policies and decisions feedback into this system to produce the resultswe have observed? The accumulationmanagement structure diagram is a usefulstarting point to begin addressing thesequestions.

Further Reading: “Modeling Managerial Behavior:Misperceptions of Feedback in a Dynamic DecisionMaking Experiment,” by John D. Sterman,Management Science, Vol. 35, No. 3, March 1989.

•


D E L A Y S : A C C U M U L A T O R SI N D I S G U I S E

T O O L B O X

dv

S

“aesun

magine a new manager at a beefpackaging plant who knows nothing

about the birthing process of calves. On hisfirst day, his workers show him a newborncalf. The dollar signs go off in his head ashe calculates: More calves mean more beef;more beef means more sales; more salesmean more profits. He points to the mothercow and barks, “I want you to get two morecalves out of that sucker by Monday morn-ing, and that’s an order!”

Of course, the workers will find a wayto fulfill his request, either by bringing twocalves from another part of the plant or per-haps slaughtering the mother cow to pro-duce the extra pounds of beef. Theworkers will have successfully executedtheir task, and the manager will continueto believe that his orders control the pro-duction cycle of calves.

The story is obviously far-fetched. Noone would expect a cow to produce a calf inone weekend. But how do we know thatequally ridiculous demands are not beingmade every day on processes where wehave less understanding of the time dynam-ics? Are there such “calving” equivalents inmanufacturing, for example, where arbi-trary quarterly sales targets given to invest-ment analysts translate into marchingorders for the production line?

AG ING CHA INSTRUCTURE

Why is it that whenever we want some-thing right away, it seems to take forever?Then when we do get it, it often is morethan we ever wanted? Chances are, delaysplayed a large part in our mis-timing. Werealize that “things take time,” but unlesswe have a clear understanding of the struc-

I


ture that produces the delays, it is difficultto know how long to wait before we shouldtake further action.

Whenever there are significant delaysin a system, you can bet that there are accu-mulators involved. In some cases the accu-mulators are less obvious than in others,but they almost always play a role. In thecase of the packaging plant, the accumula-tor is the cow (or stock of cows). When youtake a shower, the delay in getting hotwater is due to accumulation in the lengthof pipe from the hot water heater to theshower. Even though the water “flows”

through thepipe, the volume ofwater in the pipes can beviewed as an accumulation of coldwater being pushed through the pipes bythe subsequent hot water.

The “aging chain” structure is thename given to a whole class of processesthat includes the birthing cycle of cows,production-distribution systems, and thespread of infectious diseases (see “GenericAging Chain Structure”). The key princi-ple in the aging chain is that there aredelays in the system that depend on theinherent nature of that system, and thosedelays cannot be shortened except withinsome narrow bounds. For calf productionthat delay is the gestation period. For newideas it may be an incubation period, fornew products it may be the development

Inflow

Stage 1

Advance Rate 1

Stage 2

A

GENERIC AGING CHAIN

Theproc


time, and for production it is the manufac-turing cycle time. Trying to shorten theinherent time delay by pushing thingsthrough the accumulators faster can wreakhavoc on the system.

The aging chain represents a multi-stage process where “stuff” (ideas, prod-ucts, calves) moving through the systemundergoes various stages of development.Each stage can be represented by an accu-mulator, where the stuff “ages” beforemoving on to the next stage. The “aging”time at any stage can vary.

PRODUCT ION SYSTEMS

One way to get a better idea of the delaysinvolved in a system is to create a struc-tural map. A typical production system isshown in the “Production Chain” diagram.In the diagram, different stages of produc-tion process are represented by accumula-tors. (For dimensional consistency, thisdiagram represents accounting numbersand not the actual physical stuff movingthrough the system.) The accumulator andflow diagram is very much like a processflow chart, showing how production canbe mapped out as a series of stages. Each

Outflow

Stage n

ance Rate 2

TRUCTURE

ging chain” structure represents a multi-stages where “stuff” moving through the systemdergoes various stages of development (indi-

cated by accumulators).

U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

roduction Requests

ProductionRequestBacklog

Shipments

Schedule Rate

ProductionQueue

FinishedGoods

Production Starts

Work inProcess

ProductionSchedulingLoad

FullUtilization

DownTime

R1

R2

s

s

s

o

o

Shipment GapExpediting Customer

Orders

s

s

s

s

o

R4

R3

o expedite orders can backfire by actually lengthening the average production delay and reinforcingd to expedite (loops R3 and R4).

E X P E D I T I N G L O O P S

P

DT

accumulator, in effect, adds a delay to thesystem.

If we wanted to add greater detail, eachaccumulator could be further broken outinto smaller stages. For example, theWork-in-Process accumulator may bebroken into various production stagessuch as assembly, paint, bake, test, andinspection. Each stage has its owntime delay or cycle time associated with it.

These time delays have an importantaspect: They usually do not stay con-stant—a point that is not captured in atypical process flow chart. In the firststage, for example, a rising backlog of pro-duction requests increases the schedulingload, which decreases the rate at which therequests can be scheduled. This keeps thebacklog high, further exacerbating theschedule load problem (R1). Once a sched-ule gets behind, the system can actuallyreinforce the tendency to fall furtherbehind.

Similarly, at the work-in-process (WIP)stage, if we load up on production starts, theWIP inventory rises. As theWIP rises, thesheer amount of extra “stuff” in the workscan slow things down. Another consequenceof pumping up the system with additionalWIP is that it creates pressure torun equipment at full capacity,which can lead to increased downtime (no time for maintenancework) and lower yields (higherscrap rates). All this will ultimatelylead to lower production and anaccumulation of moreWIP (R2).

DON ’ T JUST DOSOMETH ING , STANDTHERE !

The “Production Chain” diagramdoes not show the complete pictureof what goes on in a typical manu-facturing setting (see “ExpeditingLoops”). When shipments fall shortof customer orders, customers are

P

Efforts tthe nee


left waiting for their goods. One response tosuch a gap is to go into the production sys-tem and expedite some of the more “impor-tant” orders and/or push more production

requests into the sys-tem. The intent is to getmore products through thechain, but these actions arelikely to produce the oppositeeffect.

Given that the system was already run-ning at full tilt, the expediting actions arelikely to create additional reinforcingsources of delays. Pushing on productionrequests will kick in the scheduling loadloop (R1), which will further delay ship-ments and intensify pressure to expedite(R3). Similarly, rearranging the production

roduction Requests

ProductionRequestBacklog

Schedule Rate

ProductionQueue

Production

SchedulingLoad

FullUtilization

R1

R2

s

s

s

o

PRODUCTION CHAIN


starts and pushing orders into productionwill exacerbate the full utilization loop(R2), leading to further delays and morepressure to expedite (R4).

In general, the aging chain structuretells us that there are structural limits tohow fast you can force a system to respond.Unless you can somehow change theinherent delays built into the various stagesof the system, the best expediting actionone can take may be to simply do noth-ing—and wait. •

Shipments

FinishedGoods

Starts

Work inProcess

Production

ownime

o

Each stage of the production chain flow dia-gram has its own time delay or cycle timeassociated with it. In many cases, the lengthof time changes as a function of the system—

a point that is usually not captured in atypical process flow diagram.


S - S H A P E D G R O W T H A N D T H E L AWO F D I M I N I S H I N G R E T U R N S

T O O L B O X

ost of us are familiar with the storyof Sir Isaac Newton sitting under-

neath an apple tree and “discovering” thelaw of gravity when he saw an apple fall(some say on his head). Had Newton beenan entrepreneur, he might have discoveredanother law. With apple sales—and prof-its—in mind, he might have shaken thetree vigorously, causing more apples todrop to the ground. The harder he shook,the more apples would fall, meaning moresales and more profits. After a while, how-ever, each shaking would produce fewerand fewer apples.

We can almost picture the scene: SirIsaac wipes the sweat from his brow, thenclimbs the tree to knock down the tena-cious few apples left. Precariously perched,he strains to reach one of the last remain-ing apples. The limb gives way, and hefalls. As he lands on the ground, another

M


The Bass Diffusion Model is an example of S-shapFrom a structural perspective, it can be representeple from a pool of Potential Adopters to Adopters. IfPotential Pet Rock Owners for Potential Adopters,Adoption Rate, and Pet Rock Owners for Adopters,tural diagram describes the dynamics of the Pet Ro

B A S S D I F F U S I O N M OA S T R U C T U R A L V I E W

PotentialAdopters

Adoption Rate

B

Time

Time

Advertising

R

discovery hits him—the Law ofDiminishing Returns.

THE LAW OFD IM IN I SH ING RETURNS

The phenomena of diminishing returns—when more effort yields fewer results—isubiquitous. Oil recovery and mining opera-tions exhibit this behavior. Companies expe-rience rapid new product sales followed bydecreasing demand. At a personal level, wesee that working longer hours, joggingmore miles, and eating less food lead todiminishing returns in productivity gains,health benefits, and weight loss.

The Law of Diminishing Returns canbe considered the law of gravity for thebusiness world. Launching a marketingcampaign, for example, is like the trajectoryof a cannonball—the returns climb higherand higher, until the “force” of diminishing

returns kicks in andpulls the rate of returndown. When traced outover time, the cumula-tive returns of the mar-keting effort produce anS-shaped curve.

The Law ofDiminishing Returns isessentially about satura-tion effects—reachingthe limits of a particularsystem. The characteris-tic S-shaped behavior ofthis process can be pro-duced by two differentstructures: the well-known Bass DiffusionModel and the moregeneral “Limits to

ed dynamics.d by a flow of peo-we substitute

Purchase Rate forthe same struc-ck fad.

D E L —P O I N T

Time

Adopters

Interaction


Success” archetype. At the heart of bothstructures is a pair of reinforcing and bal-ancing loops that interact to produce the S-shaped pattern.

D I F FUS ION DYNAM ICS

The basic Bass Diffusion Model is usuallygiven as a set of equations:

f(t) = dF(t)/dt = [p+qF(t)] [1– F(t)]p, q ≥ 0p = coefficient of advertising,q = coefficient of interaction.

Integration yields an S-shapedgrowth curve of diffusion.

Although the equations may offer anelegant way to represent such dynamics,most of us don’t view the world as a set ofequations. From an accumulator and flowsperspective, we see diffusion dynamics as aflow of people from a pool of potentialadopters to adopters (see “Bass DiffusionModel—A Structural Viewpoint”).

Instead of p’s and q’s, we talk about anadvertising effect and a word-of-moutheffect. This structural view makes thedynamics more explicit, closer to the way weactually think about and experience theworld.

Faddish products, such as hula hoopsand Cabbage Patch dolls, usually exhibitclassic S-shaped growth. Remember thesudden popularity of “pet rocks” in the late1970s? They were just plain old rocksrepackaged in a pet-carrier style box. Atfirst, driven by strong advertising, they wereviewed as a popular novelty item. Salesbegan to grow, increasing the demand forthe rocks and spreading word-of-mouthendorsements. Soon sales—and the rocks’

U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

The “Limits to Success” archetype (top) is one way of describing the capacity limitation dynamicsthat produce S-shaped behavior. Looking at the archetype from an accumulator and flow per-spective (bottom), we can see more clearly the structures producing the unwanted behavior.

“ L I M I T S T O S U C C E S S ” — F R O M A R C H E T Y P ET O A C C U M U L A T O R S

Revenues

Quality

Marketing

s

R3

s

s

B2Efforts

o

R1

s

Performance LimitingAction

s

s

Constraint

New Customers

Time

Customers Lost

Customers

Service Load Service Capacity

Time Time

CapacityAdequacy

B4B5

s

s

s

s

o

o

o

popularity—began to skyrocket. But eventu-ally the pool of potential adopters (or poten-tial pet rock owners, in this case) wasdrained, and there was no one left to buythem.

CAPAC I TY L IM I TS

In general, diminishing returns occur when-ever we hit a capacity limit. In the BassModel, capacity is the number of people whocan ultimately become adopters of a particu-lar product, technology, or idea. The adop-tion rate falls to zero when the potentialadopters accumulator is depleted, or, in thepet rock case, when everyone comes to theirsenses (whichever comes first!).

The “Limits to Success” archetype (THESYSTEMSTHINKER, December1990/January 1991) is another way of describ-ing the capacity limit dynamics that produceS-shaped behavior. In a “Limits to Success”structure, a system’s performance improvesowing to certain efforts. Better performanceresults in more efforts, leading to furtherimprovement (loop R1 in “‘Limits toSuccess’—From Archetype toAccumulators”). Over time, however, perfor-mance begins to plateau despite increasedefforts—the system has reached some limitor resistance that is preventing furtherimprovements (loop B2).

If we look at the “Limits to Success”archetype from an accumulator and flowperspective, we can see more clearly thestructures producing the unwanted behavior.In a service organization, for example, servicecapacity may become the limiting factor if itdoes not keep up with increasing demand.In the beginning, growth in customers willlead to higher revenues, increased marketing,and further growth in customers (R3). Thisreinforcing loop drives the initial growth ofthe customers accumulator.

As the number of customers grows,however, so does the service load on thecompany. If the service capacity does notgrow at least as fast as the service load,


capacity adequacy decreases. This leads tolower quality and produces a downwardpressure on customer growth. That is,capacity constraints eventually diminish theeffectiveness of the marketing efforts.

In both the Bass Diffusion Model andthe “Limits to Success” archetype, the S-shaped curve is produced by a reinforcingloop coupled with a balancing loop. Thereinforcing loop creates the initial growth indemand, while the balancing loop is gener-ally responsible for the diminishing returns.The balancing loops do not suddenly“appear.” They are almost always presentfrom the very start. When the dynamicchanges from one of rising growth to a slow-ing pace, the force driving the system hassimply shifted from a reinforcing to a balanc-ing loop. “Saturation” occurs in both cases—whether it is the saturation of a given marketor the full utilization of a specific capacity.


BREAK ING THE LAW

If you find yourself “caught” by the Law ofDiminishing Returns, using a structural dia-gram may help you identify the critical fac-tors and find a way to break out of it. In adiffusion dynamics case, for example, quan-tifying and measuring each of the differentpieces of the diagram may help decidewhether you should try to expand the poolof potential adopters, segment adopters intodifferent categories, beef up the advertisingbudget, or push on direct sales efforts.

In the more general case of capacity lim-its, breaking out of the diminishing returnsphenomena requires identifying the accu-mulator(s) that are operating at or near fullcapacity and calculating the true workloaddemand. Eliminating any gaps betweendemand and capacity is likely to producemore results than simply pushing harder onthe reinforcing loops in the system. •



C O M P U T E R M O D E L

M A N A G E M E N T F L I G H T S I M U L AT O R

L E A R N I N G L A B O R AT O RY

any of the systems we are charged with managing are so dynamically complexthat they are almost incomprehensible. Complex social systems frequently exhibit

counterintuitive behavior, where actions that provide short-term relief often result ingreater long-term pain. When actions and consequences are greatly separated in spaceand time, making effective decisions for the long-term well-being of a system becomesextremely difficult.

Causal loop diagrams, archetypes, and structure-behavior pairs can help us build abetter conceptual understanding of the key relevant structures of a system and perhapseven predict the general behavior of the system over time. “Modeling for WhatPurpose” (p. 40), however, describes times when we need even greater precision aboutthe ramifications of certain actions at specific points in time.

The rest of this section describes how we can translate our pen-and-paper represen-tations into computer-based models that can be simulated, converted into interactivedecision-making games (management flight simulators), and embedded in a rich learn-ing environment (learning laboratories).

M

P A R T I V

C O M P U T E R - B A S E D T O O L S

M O D E L I N G F O R W H AT P U R P O S E ?B Y J A Y W . F O R R E S T E R

T O O L B O X

ystem dynamics does not imposemodels on people for the first time—

models are already present in everything wedo. One does not have a family or corpora-tion or city or country in one’s head. Instead,one has observations and assumptions aboutthose systems. Such observations andassumptions constitute mental models,which we then use as a basis for action.

The ultimate success of a systemdynamics investigation depends on a clearinitial identification of an important pur-pose and objective. Presumably a systemdynamics model will organize, clarify, andunify knowledge. The model should givepeople a more effective understandingabout an important system that has previ-ously exhibited puzzling or controversial

S


ExpectationsAbout Behavior

ObservedStructure

and Policies

ActualBehavior

aa

bb

T H R E E C A T E G O R I E SO F I N F O R M A T I O N

There are three categories of informationabout a system: knowledge about structureand policies; assumptions about how the sys-tem will behave based on the observed struc-ture and policies; and the actual systembehavior as it is observed in real life. Theusual discrepancy is across the boundary a-a:expected behavior is not consistent with theknown structure and policies in the system.

behavior. In general, influential systemdynamics projects are those that changethe way people think about a system. Mereconfirmation that current beliefs and poli-cies are correct may be satisfying buthardly necessary, unless there are differ-ences of opinion to be resolved. Changingand unifying viewpoints means that therelevant mental models are being altered.

UN I FY ING KNOWLEDGE

Complex systems defy intuitive solutions.Even a third-order, linear differentialequation is unsolvable by inspection. Yet,important situations in management, eco-nomics, medicine, and social behavior usu-ally lose reality if simplified to less thanfifth-order nonlinear dynamic systems.

Attempts to deal with nonlineardynamic systems using ordinary processes ofdescription and debate lead to internalinconsistencies. Underlying assumptionsmay have been left unclear and contradic-tory, and mental models are often logicallyincomplete. Resulting behavior is likely tobe contrary to that implied by the assump-tions being made about underlying systemstructure and governing policies.

System dynamics modeling can beeffective because it builds on the reliablepart of our understanding of systems whilecompensating for the unreliable part. Thesystem dynamics procedure untangles sev-eral threads that cause confusion in ordi-nary debate: underlying assumptions(structure, policies, and parameters), andimplied behavior. By considering assump-tions independently from resulting behav-ior, there is less inclination for people todiffer on assumptions (on which they actu-


ally can agree) merely because they initiallydisagree with the dynamic conclusions thatmight follow.

If we divide knowledge of systems intothree categories, we can illustrate whereinlie the strengths and weaknesses of mentalmodels and simulation models (see “ThreeCategories of Information”). The top ofthe figure represents knowledge aboutstructure and policies; that is, about theelementary parts of a system. This is localnon-dynamic knowledge. It describesinformation available at each decision-making point. It identifies who controlseach part of a system. It reveals how pres-sures and crises influence decisions. Ingeneral, information about structure andpolicies is far more reliable, and is moreoften seen in the same way by differentpeople, than is generally assumed. It is onlynecessary to dig out the information byusing system dynamics insights about howto organize structural information toaddress a particular set of dynamic issues.

The middle of the figure representsassumptions about how the system willbehave, based on the observed structure andpolicies in the top section. These beliefs are,in effect, the assumed intuitive solutions tothe dynamic equations described by thestructure and policies in the top section ofthe diagram. They represent the solutions,arrived at by introspection and debate andcompromise, to the high-order nonlinearsystem described in the top part of the fig-ure. In the middle lie the presumptions thatlead managers to change policies or leadgovernments to change laws. Based onassumptions about how behavior isexpected to change, policies and laws in the

U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

D E R I N V E S T M E N TI N C A P A C I T Y

ens customer orders because of increasingHowever, if management is reluctant to invest inuntil the backlog reaches a certain level

acklog), orders will be driven down until demand. The awaited signal to expand capacity neveracity is controlling sales rather than potentialapacity (B2). If management tries lowering price(B4), the resulting lower profit margins will fur-capacity investment (R5).

o

B2

s

o

o

Delay

Backlogs

Perceived Need to Invest

Capacity

Customer Orders

o

Standard“Buffer” Backlog

B1

s

s

R3

CapacityInvestments

Delay

s

Delivery Delay

s

s

top section are altered in an effort toachieve assumed improved behavior in themiddle section.

The bottom of the figure representsthe actual system behavior as it is observedin real life. Very often, actual behavior dif-fers substantially from expected behavior.In other words, discrepancies exist acrossthe boundary b-b. The surprise thatobserved structure and policies do not leadto the expected behavior is usuallyexplained by assuming that informationabout structure and policies must havebeen incorrect. Unjustifiably blaminginadequate knowledge about parts of thesystem has resulted in devoting uncountedmillions of hours to data gathering, ques-tionnaires, and interviews that have failedto significantly improve the understandingof systems.

A system dynamics investigation usu-ally shows that the important discrepancyis not across the boundary b-b, but acrossthe boundary a-a. When a model is builtfrom the observed and agreed-upon struc-ture and policies, the model usuallyexhibits the actual behavior of the real sys-tem. The existing knowledge about theparts of the system is shown to explain theactual behavior. The dissidence in the dia-gram arises because the intuitivelyexpected behavior in the middle section isinconsistent with the known structure andpolicies in the top section.

These discrepancies can be foundrepeatedly in the corporate world. A fre-quently recurring example in whichknown corporate policies cause a loss ofmarket share and instability of employ-ment arises from the way delivery delayaffects sales and expansion of capacity (see“Underinvestment in Capacity”). Risingbacklog (and the accompanying increase indelivery delay) discourage incoming ordersfor a product (B1) even while managementfavors larger backlogs as a safety bufferagainst business downturns. As manage-


ment waits for still higher backlogs beforeexpanding capacity, orders are drivendown by unfavorable delivery delay untilorders equal capacity (R3). The awaitedsignal for expansion of capacity nevercomes because capacity is controlling sales,rather than potential demand controllingcapacity (B2).

When sales fail to rise because of longdelivery delays, management may thenlower price in an attempt to stimulate moresales (B4). Sales increase briefly but onlylong enough to build up sufficient addi-tional backlog and delivery delay to com-pensate for the lower prices. In addition,price reductions lower profit margins untilthere is no longer economic justification forexpansion (R5). In such a situation, ade-quate information about individual rela-tionships in the system is always availablefor successful modeling, but managers arenot aware of how the different activities ofthe company are influencing one another.

Lack of capacitymay exist in manufac-turing, product service,skilled sales people, oreven in promptanswering of tele-phones. For example,airlines cut fares toattract passengers. Buthow often, because ofinadequate telephonecapacity, are potentialcustomers put on“hold” until they hangup in favor of anotherairline?

System dynamicsmodels have littleimpact unless theychange the way peopleperceive a situation. Amodel must help toorganize informationin a more understand-

U N

Rising backlog dampdelivery delay (B1).capacity expansions(Standard “Buffer” Bequals capacity (R3)comes, because capdemand controlling cto stimulate demandther justify a delay in

B4

s

Price

R5

ProfitMargin

s


able way. A model should link the past tothe present by showing how present condi-tions arose, and extend the present into per-suasive alternative futures under a varietyof scenarios determined by policy alterna-tives. In other words, a system dynamicsmodel, if it is to be effective, must commu-nicate with and modify the prior mentalmodels. Only people’s beliefs—that is, theirmental models—will determine action.Computer models must relate to andimprove mental models if the computermodels are to fill an effective role.

JayW. Forrester, professor emeritus at theMassachusetts Institute of Technology and formerdirector of the MIT System Dynamics Group, is thefounder of the field of system dynamics. Since hisretirement in 1989, he has been working towardbringing system dynamics into K through 12th gradeschools as the basis for a new kind of education.

This article is a selection from “System Dynamics andthe Lessons of 35 Years,” in Kenyon B. De Greene(ed.) Systems-Based Approach to Policymaking,(Kluwer Academic Publishers, 1993).

•


M A N A G E M E N T F L I G H TS I M U L A T O R S : F L I G H T T R A I N I N GF O R M A N A G E R S ( P A R T I )

T O O L B O X

magine you’re leaving on a six-hourflight from Boston to Los Angeles.

As the plane pulls away from the gate, thepilot comes on over the loudspeaker: “Hi,I’m Captain Bob, and I want to thank youfor choosing to fly with us today. . . . Justwanted to let you know I’ve recently com-pleted ground school training, and I haveread all the manuals, but this is my firsttime in the cockpit. So sit back, relax, andenjoy the flight, as we learn together. . . . ”

Of course this scenario is ludicrous—apilot is allowed into a cockpit only afterhundreds of hours of experience in a flightsimulator. Then he or she spends manyadditional hours as a co-pilot, assisting in theoperation of an aircraft. The result of thiscareful system of education and training isan industry with the highest safety record ofany mode of transportation.

FL IGHT TRA IN ING FORMANAGERS

Imagine if we trained pilots like we domanagers; how many people would bewilling to take a flight? Managerial train-

I


TimePressure

CoSelect Issue Focus

Service Quality• capacity• quality• productivity• time pressure• etc.

GatherData

GatheDat

M A N A G E M E N T F L I

ing, in the traditional business-schoolsense, is the equivalent of ground-schoolfor pilots. Managers-to-be read textbooksand solve already formulated problems,but they don’t get much real experiencebefore they have to perform on-line.

MANAGEMENT FL IGHTS IMULATORS

Management Flight Simulators (MFSs) pro-vide a simulated environment in whichmanagers can “learn from experience” in acontrolled setting. The simulator capturesthe interconnections between the differentparts of the system under study and pro-vides a computer interface that allowsmanagers to interact with the modelthrough a familiar “lens” (reports, graphs,and spreadsheets).

Similar to a pilot’s flight simulator cock-pit, an MFS puts managers in control of arealistic environment where they are incharge of making key decisions similar tothe ones they face in their actual work set-tings. MFS’s are particularly useful for get-ting away from the details of day-to-day


WorkDone

ProductivityB

Morale

R

Computer Modelnceptual Model

Hires T/O

Personnel

TimePressure

Productivity

ra

GatherData

Test tModel

Test theModel

Revise

G H T S I M U L A T O R D E V E L

operations and focusing on the long-termdynamics of managerial decisions.

CREAT ING A FL IGHTS IMULATOR

There are four stages involved in creatinga management flight simulator: (1) select-ing an issue focus, (2) developing a concep-tual model, (3) constructing a computermodel, and (4) translating the computermodel into an interactive simulator (see“Management Flight SimulatorDevelopment Stages”). These four stagesinvolve integrating many of the tools ofsystems thinking into a single, powerfullearning tool (see “A Palette of SystemsThinking Tools,” p. 10, for a description ofeach of the tools).

1. Select issue focus. The first step indesigning a flight simulator is to choose anissue to explore. To select a topic, look fora problem symptom that has been aroundfor a long time or a puzzling dynamic youwant to investigate (see “The Do’s andDon’t’s of Systems Thinking on the Job,”p. 52, for guidelines on identifying good

U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

STOCK

HIRING

STOCKHIRINGDECISION INFOCOCKPIT

Flight Simulator

Test the Model

he

O P M E N T S T A G E S

WorkDone

TimePressure

sProductivity

B2

o

o

o

ss

s

MoraleDelay R3

P R E S S U R E L O O P S

puts pressure on service personnel. Initially,rder, thus increasing productivity and reducingver time, however, morale can suffer, hurting

reasing the time pressure (R3).

systems problems). The goal at this stage isto gather relevant data through interviews,company records, and the experience baseof those involved in developing the MFS.

For example, let’s say we are puzzledby a pattern of oscillating quality levels inthe customer service department.Interviews with people in the departmentreveal a pattern of tremendous time pres-sure that repeats in a regular cycle.Company data provide a record of steadilyrising sales and irregular levels of customersatisfaction. This process grounds the pro-ject in real data from which to build acausal theory.

2. Build conceptual model. After select-ing an issue focus, you can begin to build aconceptual model that organizes the datainto a coherent dynamic theory. Systemsarchetypes and causal loop diagrams (CLDs)can be very helpful for trying to understandwhat is going on (see “Systems Archetypes ata Glance,” p. 20, and “Guidelines forDrawing Causal Loop Diagrams,” p. 18).

In the customer service quality exam-ple, we can start building causal structuresthat provide plausible explanations for theobserved data. When customer demandincreases, we know our service people feeladded pressure. If the increase in timepressure is not addressed, quality tends todrift downward and eventually dampensdemand (B1 in “Time Pressure Loops”).

Our service people tell us people ini-tially respond to the time pressure byworking harder, thereby increasing pro-ductivity and getting more work done(B2). If the time pressure persists, however,morale declines and begins to hurt produc-tivity even though people continue to workharder. As time pressure escalates, moralespirals downward (R3). By adding loops,we can continue building a dynamic theoryabout our customer service setting.

3. Construct computer model. Thedynamic theory developed in the concep-tual model helps guide the construction of


a computer model. It provides a frame-work for people to distill their experienceinto explicit statements that can be repre-sented in a computer model. Just as thepilot’s flight simulator is created based onthe laws of physics and aerodynamics, thecomputer modeling process uses a set offundamental building blocks (e.g., accumu-lators and flows) to represent a coherent setof theories about the interconnections in anorganization.

In the customer service quality example,we could model the number of “personnel”as an accumulator and “hires” and“turnover” as inflows and outflows, respec-tively. The effect of time pressure onturnover may be modeled using a graphicalfunction diagram representing a nonlinearlink between the two variables. That is,there may be little or no negative effects atlow levels of pressure, but beyond a certainthreshold, there may be a sudden dramaticincrease in turnover.

4. Translate to flight simulator. Pilotsfirst learn about the principles and con-cepts of aviation in school and then use thesimulator to gain a better understanding ofhow those principles actually play out inreal life. Likewise, a management flightsimulator is created by translating the“principles” captured in the computermodel into a form that allows managers tointeract with it in a real-istic way.

A good simulatorinterface should providemanagers with a set ofdecisions that either theycontrol, or that directlyaffect them. The maincriteria should be that thedecisions are directly rel-evant or easily transfer-able from the simulatorto the workplace.

In the service qualityexample, the simulator

Quality

CustomerDemand

B1s

T I M E

Customer demandpeople may work hathe pressure (B2). Oproductivity and inc


could require managers to make decisionsabout hiring/firing, monthly productionnumbers, and quality standards. By imple-menting a “Quality First” policy, for exam-ple, we may discover that if we raise qualitystandards but don’t adjust capacity, we actu-ally end up with lower quality in the longrun. Quality improves in the short run, butas time pressure persists, morale decreases,turnover increases, which in turn increasestime pressure, resulting in more turnover.The overall dynamic is a vicious reinforcingcycle in which capacity continually erodesand quality suffers.

The simulator should also providemanagers with the same types of reports,spreadsheets, and graphs they use to makedecisions. There are many issues involvingthe design of the simulator’s managementinformation system that are entwined withthe intended use of the MFS as a whole.These and other issues are covered in PartII (p. 44).

For help on converting conceptual maps to computermodels, see “Accumulators: Bathtubs, BathtubsEverywhere,” p. 30; “Structural Thinking: The WorldAccording to Accumulators and Flows,” p. 28; and“Graphical Functions: ‘Seeing’ the Full Story,” p. 26.To learn more about constructing computer models,see Introduction to System Dynamics Modeling(Pegasus Communications), and Academic User’sGuide to STELLA (Hanover, NH: HighPerformance Systems).

•


M A N A G E M E N T F L I G H TS I M U L A T O R S : F L I G H T T R A I N I N GF O R M A N A G E R S ( P A R T I I )

T O O L B O X

management flight simulator,along with causal loop diagrams

and systems archetypes, allows you to seemore clearly the connections between yourdecisions and future consequences. As sim-ulated months pass in a matter of minutesand the consequences of your actionsunfold, an MFS provides a means formaking sense of the short-term and long-term effects of your decisions.

Management flight simulators can bemost useful for understanding situations inwhich causality is distant in time andspace. When the inherent time lag is par-ticularly long (on the order of months oryears) and organizational complexity ishigh (see “Organizational Complexity”),learning from experience can be fraughtwith pitfalls. An MFS allows you to lever-age your ability to learn from experience ina complex environment.

A


ORG

AN

IZA

TIO

NA

LCO

MPL

EXIT

YO

FA

“MES

S”

TIME LAG

Hours

Cross-Industry

Cross-Organizational

Cross-Functional

Multiple-Intrafunctional

Unifunctional

Days

PrDe

Scrap andRepair Costs

ManufacturingCycle Time

O R G A N I Z A T I O N A

The greater the complexity or the longer the timeing people learn from experience.

MANAG ING VS . L EARN ING

There are two fundamentally differentuses for a computer model and simula-tor—managing and learning. Simulatorsand models designed to support decision-making in a real operational setting mustfocus on capturing the operational realityprecisely because operational or strategicdecisions will be based on those numbers.

Simulators that are designed for learn-ing, on the other hand, are much moreconcerned with surfacing the tacit mentalmodels that drive managers’ decision-making. Accuracy of specific numbers isnot as important as the relevancy of theissues and concepts captured in the simula-tor; in other words, simulators for learningare idea-rich versus data-rich.

There are several different simulatordesign criteria to keep in mind whendesigning an MFS:


S OF PROCESS (on the order of)

Increasing Dynamic Complexity

Months Years Decades

oductvelopment Cycle

National Standard of Living

Supplier-Distribution Networks

L C O M P L E X I T Y

lags, the more an MFS is beneficial for help-

• A clear, real-world context provides areal operational focus that engages linemanagers in learning more about their ownissues.

• Face validity: Make the MFS realenough so the simulator grounds people intheir own real-life experiences.

• A strong conceptual framework helpsmake systemic sense out of the complexdynamics (e.g., systems archetypes).

• Conventional and unconventionalinformation systems provide a familiarinformation environment, as well as anopportunity to explore and experiment withnew ones.

• Surface and challenge mental modelsto break through individual mental straight-jackets and corporate sacred cows andadvance team learning.

DES IGN ING MFS ’ S ASTRANS I T IONAL OBJECTS

Designing an MFS for learning requires aninterface that maintains a careful balancebetween realism and comprehensibility. Itneeds to be real enough to serve as a transi-tional object, which Seymour Papert saysallows managers to “play out” a scenario,learn about the system, and explore how theyinteract with that system. However, it alsoneeds to be manageable—if the model triesto capture every little detail of reality, it canbecome just as complex and incomprehensi-ble as the real system.

In a learning setting, it is also importantnot to position the model as an answer-generator, but rather as an exploratory toolfor gaining a better understanding aboutone’s environment. The MFS acts like a

U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

mirror that reflects mental models in a waythat helps us understand current reality better.

There are three major elements of anMFS: decisions, reports, and a managementinformation system (see below for a sampleinterface produced withMicroWorldCreator).

Decisions. The kinds of decisions madein the simulator should be those the partici-pants would either make themselves in reallife or those someone else in the organiza-tion would make that affect the partici-pants. The decisions should be directlyrelevant or easily transferrable from thesimulator to the workplace. If the decisionsare too far removed, the simulator becomesmore of an academic exercise or a game,even when a meaningful context is builtaround it.

Although the participant might not bethe one who makes hiring and staffing deci-sions at his/her level, for example, thesedecisions can be included in the MFSbecause they are still part of the real envi-ronment in which the participants manage.In fact, putting that manager or supervisorinto the decision-making role can be an illu-minating experience: He or she will learnwhat role they play in the system and real-ize the challenge of managing from thelevel above.

Reports. As far as the actual physicaldesign of the simulator interface, there aresome general guidelines. The reportsshould look similar to what people typicallyreceive—they should not provide additionaldata that is normally not accessible, such astime pressure, or perceived quality by thecustomer. If these variables need to beincluded, they should not be as prominentas more typical day-to-day data.

Information systems. Designing theinformation system provides a lot moreflexibility in reporting variables that arenormally inaccessible. For example, a criti-cal variable like “time pressure” can helpyou experiment to see how people may


manage differently when provided withsuch information.

OUTCOMES FROM THES IMULATOR

Once participants work with the simulatorand understand the theory behind it, theycan make connections between the simula-tor and their real work situation more eas-ily. Participants can also explore whatinterventions they might make in order tobetter manage the process: What kinds ofadjustments need to be made? What con-trols do participants need to monitor?

Theories-in-use. The simulator can bea powerful tool for surfacing tacit assump-tions, for it reflects participants’ under-standing of the system. When someonemakes a decision and then explains it withdata or information that are not in themodel, they explicate their own theoriesand understanding of what’s going on. Forexample, in a management flight simulatorcreated for insurance claims managers, par-ticipants assumed that settlement dollarswere rising because of inflation. However,when they discovered that inflation was notincluded as part of the simulator, they hadto rethink their own understanding of whatcauses settlement dollars to rise.

The simulator can also reveal the gapthat often exists between espoused theories(what we conceptually believe is the rightcourse of action) and our theories-in-use(what we choose to do, given the surround-ing circumstances). Forexample, in a session with aproduct development flightsimulator, the participantsall agreed that investing incoordination and commu-nication between upstreamand downstream activitiesis important. But whenthey were placed in thesimulator and given theobjective of meeting tim-


ing, quality and cost, most of them actuallychose to invest very little in coordination.Instead, they focused on trying to get thetasks done in each area.

Team learning. A simulator can be evenmore useful if used in groups. The interplaybetween the participants, as they proposenew strategies and explain their reasoning,helps them to surface and clarify theirassumptions. The simulator can be struc-tured to require participation and coordina-tion among a group of people to encourageteam learning. For example, in a productengineering case, the team could be madeup of a product and a process engineer.Each one would be responsible for staffingand workweek decisions for their particu-lar function, but together they woulddecide on a schedule completion date andmanage the coordination between the twofunctions. The use of the simulator cantherefore be designed to provide a richerpractice field for a team to manage.

SUMMARY

MFS’s provide managers with a simulatedexperience of working through issues orimplementing a strategy. The practice fieldelement also enables the simulator to providean experiential “feel” for the dynamics ofdecision-making. Participants gain practicein the art of decision-making: reflecting onthe consequences, exploring the causal con-nections, and understanding the underlyingstructure producing the behaviors. •


L E A R N I N G L A B O R A T O R I E S :P R A C T I C I N G B E T W E E NP E R F O R M A N C E S

T O O L B O X

magine you are walking across atightrope stretched between the

world trade towers in New York City. Thewind is blowing and the rope is shaking asyou inch your way forward. One of yourteammates is sitting in the wheelbarrow youare balancing in front of you, while anotherperches on your shoulders. There are nosafety nets, no harnesses. You are thinking toyourself, “One false move and the three of uswill be taking an express elevator straightdown to the street.” Suddenly your traineryells from the other side, “Try a newmove!Experiment! Take some risks! Remember,you are a learning team!”

Sound ludicrous? No one would becrazy enough to try something new in a sit-uation like that. And yet that is preciselywhat many companies expect managementteams to do—experiment and learn in anenvironment that is risky, turbulent, andunpredictable. Unlike a high-wire act orsports team, however, management teamsdo not have a practice field in which tolearn; they are always on the performance field.

DES IGN ING MEAN INGFULPRACT ICE F I E LDS

A learning lab can be viewed as a manager’sequivalent of a sports team’s practice field.The goal of a learning lab design is to pro-vide a “real” enough practice field so thatthe lessons are meaningful, but safe enoughto encourage experimentation and learning.In the tightrope example, a practice fieldcould be a rope stretched across two pillarssix feet off the ground. There may be matsbelow to cushion the fall, but also a large fanto simulate the kinds of winds you would

I


encounter in the actual setting.A managerial practice field should also

have its own sets of equipment and tools formaking the practice sessions meaningful.The purpose of a “learning laboratory” is toprovide an environment in which managerscan experiment with alternative policies, testassumptions, and practice working throughcomplex issues productively. It should allowmanagers to practice working together as ateam on issues of real significance to them.To be effective, the learning lab must pro-vide (1) an environment conducive to learn-ing, (2) a way of surfacing deep-rootedassumptions that affect the way we thinkand act, (3) tools for understanding our real-ity in a way that highlights the interconnec-tions and the systemic consequences of ouractions, and (4) a management flight simula-tor that allows us to speed up or slow downtime, experiment with different strategies,and see the long-term consequences of ouractions (see “A Sample Learning LaboratoryDesign”).

CREAT ING A SAFELEARN ING SPACE

Learning usually involves making mistakesbecause we are trying things we have neverdone before. It requires us to approachthings from a place of “not knowing.” Itinvolves risk. How, then, can we create asafe space where people feel free to learn?

There are some ground rules that canhelp create such safe spaces. One groundrule is to hold each person’s viewpoint asvalid. That requires taking the position that“If I could stand in the other person’s shoes,I too could see what the other person sees.”


It does not mean you agree or disagree withthat person’s view; you simply acknowledgethe right of that person to hold that view. Asecond rule is to suspend one’s own assump-tions and the other person’s and hold themequally in our minds, without judging oursto be superior or “right.” Creating such alearning space also means engaging in dia-logue rather than discussion—operating in aspirit of inquiry rather than advocacy.

MAPP ING MENTALMODELS

Along with the proper environment, weneed tools for helping people surface andshare their assumptions. For example, the“Ladder of Inference,” developed by ChrisArgyris, distinguishes between directlyobservable data, shared cultural meanings,judgments, conclusions, and values andassumptions. Argyris uses the ladder toillustrate the “leaps of inference” that occurwhen people take a little bit of observed data(a person walks into a 2:00 meeting at 2:15)and go straight up the ladder to the level ofvalues and assumptions (“He’s late anddoesn’t care about the project or the otherplayers”) without even being conscious of it.The ladder provides a useful framework forhelping people “walk back down the lad-der” to understand what is really happeningand begin managing by facts, not opinions.

Systems archetypes also provide a pow-erful set of tools for mapping out a person’sunderstanding of a problem or issue in aform that invites others to inquire and clar-ify the picture together. Having one’sassumptions captured in terms of archetypesand causal loop diagrams helps depersonal-

U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

ize the issue and focuses everyone’s energyon the diagram, not the person. These dia-grams also explicate the assumptions behindthe connections, and clarify the points ofagreement or contention.

MANAGEMENT FL IGHTS IMULATORS

When practicing a concerto, an orchestra hasthe ability to slow down or speed up time topractice certain sections. Through manage-ment flight simulators (computer modelsthat have been turned into interactive deci-sion-making games), managers can alsoaccelerate time to see the long-term conse-quences of decisions, or slow down the flow


A S A M P L E

of time at each decision point. With a simu-lator, a manager can test out new strategiesand policies, reflect on the outcomes, anddiscuss pertinent issues with others in theteam.

By providing quick and accurate feed-back, the computer simulator can facilitatelearning by shortening the delay betweenaction and outcome. Managers can chart astrategy and implement it over a simulatednumber of years in a matter of minutes.They can try scenarios that bankrupt thecompany or lose market share without risk-ing a single dollar or job. As they explore thesystemic reasons for their results, managerscan begin to understand the underlying


L E A R N I N G L A B O R A T O R

forces that produce a given set of outcomes.

PERFORM ING ON THET IGHTROPE

In order to foster learning among teams ofmanagers, we must look for alternate ways tohelp them deal with the increasing complex-ity of today’s business world. Providing safe,yet meaningful learning environments wherethey can continually alternate between prac-tice and performance is one approach.Whether we are walking on a tightropestretched across two buildings or across twocompeting product strategies, practice isbound to improve our performance. •

Y D E S I G N

In designing a Learning Lab (LL), the goal is to create anenvironment that is of operational relevance. The labshould help managers step out of day-to-day demands to:• reflect on their decision-making• develop a common language• learn new tools for thinking systemically• discuss operational objectives and strategies in an

open forum• test operating assumptions• experiment with new policies and strategies for

managing• have fun.1. The First Crucial Hour—Buy-InExplaining the context of the LL to participants (the historyof its development, the original intent or purpose) is criticalfor establishing a common understanding: The LL is notmeant to provide “the answers,” but to serve as a usefulvehicle for illuminating and communicating issues ofimportance. The facilitators are positioned as enablers,not authority figures, and the participants are encouragedto question the assumptions behind the LL design.2. Current Reality—Where Are We?This exercise helps the participants construct a group pic-ture of current problems and issues they face in their jobs.Working in small groups, they are then asked to brain-storm and come up with a list of operational objectives,strategies required to achieve them, and obstacles thatneed to be overcome in order to reach their goals (e.g.,reduce settlement expenses by 10 percent). The idea is toget everyone thinking in terms of their own operationalreality.

3. Introducing the ToolsCausal loop diagrams (CLDs) and systems archetypesare introduced in a “storytelling” format in which partici-pants begin to tell systemic stories about their issues.Facilitators then describe a small portion of the CLDs thatwere in the game model to connect the tool to the issuesat hand. The underlying purpose is to get people to imme-diately begin to connect each structure to correspondingpatterns of behavior over time.4. Using the Tools—ConceptualizingIn small groups, the participants are asked to focus on aparticular issue, such as one of the decision variables inthe management flight simulator, and (1) determine thekey factors that affect that variable, (2) sketch patterns ofbehavior, (3) provide structural explanation (using CLDs),and (4) identify intervention points. By having the groupexplore these variables, the participants can replicate partof the model-building process and accept the predevel-oped model. The overall objective in this section is tohave the group cover all the major issues contained in themodel and have a chance to challenge and test the inter-relations that different people within the group may pro-pose.5. Introducing the Computer SimulatorThe facilitators begin by showing a simplified CLD thatcontains all the major variables in the model. They tracethrough the major loops and explain the dynamic conse-quences of a particular action or incident, and then sketcha corresponding pattern of behavior and connect it back tothe structure. This is followed by a hands-on introductionto the computer and simulator.


Continued on the next page

4

A S A M P L E L E A R N I N G L A B O R A T O R Y D E S I G N

Continued from the previous page

6. Planned Scenarios—Holding the ReinsIn this section, it is best if people work in groups of two ateach computer. The teams are instructed to pursue asingle-minded strategy where they are accountable formeeting one particular goal (e.g., hiring freeze). Each two-person team is responsible for doing the following:(1) plan a strategy and commit to it on paper, (2) predictthe consequences of executing the strategy by sketchingin behavior over time of some key variables, (3) play thegame, and (4) debrief game results and explain to the restof the group. This stage allows participants to begin toaddress particular organizational issues within a carefullycontrolled setting.


7. Free Plays—Cutting the ReinsThis time, the participants are free to choose their ownobjectives and time tables. Again, each team strategizes,and then explains to the rest of the group how they plan toachieve their goals. For the designers of the LL, this sec-tion provides the opportunity to challenge deep-rootednorms and assumptions, address specific “hot topics,” orre-create various historical behavior modes for furtherexploration.

P E G A S U S C O M M U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0


P A R T V

R E F E R E N C E G U I D E

T H E V O C A B U L A R Y O F S Y S T E M ST H I N K I N G : A P O C K E T G U I D E

T O O L B O X

ystems thinking can serve as a lan-guage for communicating about com-

plexity and interdependencies. To be fullyconversant in any language, you must gainsome mastery of the vocabulary, especiallythe phrases and idioms unique to that lan-guage. This glossary lists many terms thatmay come in handy when you’re faced with asystems problem.

Accumulator—Anything that builds up ordwindles; for example, water in a bath-tub, savings in a bank account, inventoryin a warehouse. In modeling software, astock is often used as a generic symbol foraccumulators. Also known as Stock orLevel.

Balancing Process/Loop—Combined withreinforcing loops, balancing processesform the building blocks of dynamic sys-tems. Balancing processes seek equilib-rium: They try to bring things to adesired state and keep them there. Theyalso limit and constrain change generatedby reinforcing processes. A balancing loopin a causal loop diagram depicts a balanc-ing process.

Balancing Process with Delay—A commonlyoccurring structure. When a balancingprocess has a long delay, the usualresponse is to overcorrect. Overcorrectionleads to wild swings in behavior. Example:real estate cycles.

Behavior Over Time (BOT) Diagram—Oneof the 10 tools of systems thinking. BOTdiagrams capture the history or trend ofone or more variables over time. Bysketching several variables on one graph,you can gain an explicit understanding ofhow they interact over time. Also calledReference Mode.

Causal Loop Diagram (CLD)—One of the10 tools of systems thinking. Causal loopdiagrams capture how variables in a sys-tem are interrelated. A CLD takes the

S


form of a closed loop that depicts cause-and-effect linkages.

Drifting Goals—A systems archetype. In a“Drifting Goals” scenario, a gradualdownward slide in performance goalsgoes unnoticed, threatening the long-termfuture of the system or organization.Example: lengthening delivery delays.

Escalation—A systems archetype. In the“Escalation” archetype, two parties com-pete for superiority in an arena. As oneparty’s actions put it ahead, the otherparty “retaliates” by increasing its actions.The result is a continual ratcheting up ofactivity on both sides. Examples: price bat-tles, the Cold War.

Feedback—The return of information aboutthe status of a process.Example: annual performance reviewsreturn information to an employee aboutthe quality of his or her work.

Fixes That Fail—A systems archetype. In a“Fixes That Fail” situation, a fix isapplied to a problem and has immediatepositive results. However, the fix also hasunforeseen long-term consequences thateventually worsen the problem. Alsoknown as “Fixes That Backfire.”

Flow—The amount of change somethingundergoes during a particular unit oftime.Example: the amount of water thatflows out of a bathtub each minute, or theamount of interest earned in a savingsaccount each month. Also called a Rate.

Generic Structures—Structures that can begeneralized across many different settingsbecause the underlying relationships arefundamentally the same. Systemsarchetypes are a class of generic struc-tures.

Graphical Function Diagram (GFD)—Oneof the 10 tools of systems thinking. GFDsshow how one variable, such as deliverydelays, interacts with another, such assales, by plotting the relationship between


the two over the entire range of relevantvalues. The resulting diagram is a concisehypothesis of how the two variables inter-relate. Also called Table Function.

Growth and Underinvestment—A systemsarchetype. In this situation, resourceinvestments in a growing area are notmade, owing to short-term pressures. Asgrowth begins to stall because of lack ofresources, there is less incentive foradding capacity, and growth slows evenfurther.

Learning Laboratory—One of the 10 tools ofsystems thinking. A learning laboratoryembeds a management flight simulator ina learning environment. Groups of man-agers use a combination of systems think-ing tools to explore the dynamics of aparticular system and inquire into theirown understanding of that system.Learning labs serve as a manager’s prac-tice field.

Level—See Accumulator.Leverage Point—An area where small

change can yield large improvements in asystem.

Limits to Success—A systems archetype. In a“Limits to Success” scenario, a companyor product line grows rapidly at first, buteventually begins to slow or even decline.The reason is that the system has hit somelimit—capacity constraints, resource lim-its, market saturation, etc.—that isinhibiting further growth. Also called“Limits to Growth.”

Management Flight Simulator (MFS)—Oneof the 10 tools of systems thinking.Similar to a pilot’s flight simulator, anMFS allows managers to test the outcomeof different policies and decisions without“crashing and burning” real companies.An MFS is based on a system dynamicscomputer model that has been changedinto an interactive decision-making simu-lator through the use of a user interface.

U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

Policy Structure Diagram—One of the 10tools of systems thinking. Policy structurediagrams are used to create a conceptual“map” of the decision-making process thatis embedded in an organization. It high-lights the factors that are weighed at eachdecision point.

Rate—See Flow.Reference Mode—See Behavior Over Time

Diagram.Reinforcing Process/Loop—Along with bal-

ancing loops, reinforcing loops form thebuilding blocks of dynamic systems.Reinforcing processes compound changein one direction with even more change inthat same direction. As such, they gener-ate both growth and collapse. A reinforc-ing loop in a causal loop diagram depictsa reinforcing process. Also known asvicious cycles or virtuous cycles.

Shifting the Burden—A systems archetype.In a “Shifting the Burden” situation, ashort-term solution is tried that success-fully solves an ongoing problem. As thesolution is used over and over again, ittakes attention away from more funda-mental, enduring solutions. Over time,the ability to apply a fundamental solu-tion may decrease, resulting in more andmore reliance on the symptomatic solu-tion. Examples: drug and alcohol depen-dency.

Shifting the Burden to the Intervener—Aspecial case of the “Shifting the Burden”systems archetype that occurs when anintervener is brought in to help solve anongoing problem. Over time, as the inter-vener successfully handles the problem,the people within the system become lesscapable of solving the problem them-selves. They become even more depen-dent on the intervener. Example: ongoinguse of outside consultants.

Simulation Model—One of the 10 tools ofsystems thinking. A computer model thatlets you map the relationships that are


important to a problem or an issue andthen simulate the interaction of those vari-ables over time.

Stock—See Accumulator.Structural Diagram—Draws out the accumu-

lators and flows in a system, giving anoverview of the major structural elementsthat produce the system’s behavior. Alsocalled flow diagram or accumulator/flowdiagram.

Structure-Behavior Pair—One of the 10 toolsof systems thinking. A structure-behaviorpair consists of a structural representationof a business issue, using accumulatorsand flows, and the corresponding behav-ior over time (BOT) diagram for the issuebeing studied.

Structure—The manner in which a system’selements are organized or interrelated.The structure of an organization, forexample, could include not only theorganizational chart but also incentivesystems, information flows, and inter-personal interactions.

Success to the Successful—A systemsarchetype. In a “Success to the Successful”situation, two activities compete for a com-mon but limited resource. The activity thatis initially more successful is consistentlygiven more resources, allowing it to suc-ceed even more. At the same time, theactivity that is initially less successfulbecomes starved for resources and eventu-ally dies out. Example: the QWERTY lay-out of typewriter keyboards.

System Dynamics—A field of study thatincludes a methodology for constructingcomputer simulation models to achievebetter understanding of social and corpo-rate systems. It draws on organizationalstudies, behavioral decision theory, andengineering to provide a theoretical andempirical base for structuring the rela-tionships in complex systems.

System—A group of interacting, interrelated,or interdependent elements forming a


complex whole. Almost always definedwith respect to a specific purpose within alarger system. Example: An R&D depart-ment is a system that has a purpose in thecontext of the larger organization.

Systems Archetypes—One of the 10 tools ofsystems thinking. Systems archetypes arethe “classic stories” in systems thinking—common patterns and structures thatoccur repeatedly in different settings.

Systems Thinking—A school of thought thatfocuses on recognizing the interconnec-tions between the parts of a system andsynthesizing them into a unified view ofthe whole.

Table Function—See Graphical FunctionDiagram.

Template—A tool used to identify systemsarchetypes. To use a template, you fill inthe blank variables in causal loop dia-grams.

Tragedy of the Commons—A systemsarchetype. In a “Tragedy of the Commons”scenario, a shared resource becomes over-burdened as each person in the system usesmore and more of the resource for individ-ual gain. Eventually, the resource dwindlesor is wiped out, resulting in lower gains foreveryone involved. Example: theGreenhouse Effect.

The above glossary is a compilation of definitions frommany sources, including:

• Innovation Associates’ and GKA’s Introduction toSystems Thinking coursebooks

• The Fifth Discipline: The Art and Practice of theLearning Organization, by Peter Senge

•High Performance Systems’ Academic User’sGuide to STELLA

• The American Heritage Dictionary and TheRandom House Dictionary.

•


T H E D O ’ S A N D D O N ’ T ’ S O FS Y S T E M S T H I N K I N G O N T H E J O BB Y M I C H A E L G O O D M A N

T O O L B O X

o, you’ve taken a systems thinkingcourse—or maybe you’ve read a few

issues of The Systems Thinker—and now youwant to start using systems thinking on thejob. How do you begin? Your best bet is toapproach this endeavor in the spirit of“learning to walk before you run.” Here aresome suggestions:

OVERALL GU IDEL INE

The tools of systems thinking are best usedas vehicles to promote team learning in theorganization. Whether you are doing “paperand pencil” models or creating full-fledgedmicroworlds, the process of constructingand using models is primarily about explor-ing and examining our “mental models”—the deeply held assumptions that influencethe way we think and act.

S


The problem should have ALL of the fcharacteristics:

1. The issue is important to me and mness.2. The problem is chronic, rather thantime event.3. The problem has a known history thdescribe.

Example: Profits were steady for 2but have been declining for the last 6 mOr: Productivity rose rapidly until abouago, when it leveled off.4. People have tried to solve this probbefore, with little or no success.

If your problem does not have all of thacteristics (especially the first three), itbe appropriate for a systems thinkingTry redefining it for a different approac

I D E N T I F Y I N G A S Y S TP R O B L E M

GENERAL GU IDEL INES

DON’T use systems thinking to furtheryour own agenda. Systems thinking is mosteffective when it is used to look at a problemin a new way, not to advocate a predeter-mined solution. Strong advocacy will createresistance—both to your ideas, and to sys-tems thinking. It should be used in the spiritof inquiry, not inquisition.

DO use systems thinking to sift outmajor issues and factors.Benefit: Systems thinking can help you

break through the clutter of everydayevents to recognize general patterns ofbehavior and the structures that are produc-ing them. It also helps in separating solu-tions from underlying problems. Too oftenwe identify problems in terms of their solu-tion; for example, “the problem is that we

have too many ________(fill in the blank: people,initiatives, steps in our pro-cess),” or “the problem isthat we have too little__________ (resources,information, budget . . .).”

DON’T use systemsthinking to blame individu-als. Chronic, unresolvedproblems are more oftenthe result of systemic break-downs than individual mis-takes. Solutions to theseproblems lie at the systemic,not the individual, level.

DO use systems think-ing to promote inquiry andchallenge preconceivedideas.

ollowing

y busi-

a one-

at I can

years,onths.

t a year

lem

ese char-may notanalysis.h.

E M S


Cues that non-systemic thinking is goingon: Phrases such as “We need to have imme-diate results,” “We just have to do more ofwhat we did last time,” or “It’s just a matterof trying harder.”

GETT ING STARTED

DON’T attempt to solve a problem imme-diately. Don’t expect to represent, much lessunderstand, persistent and complex systemicproblems overnight. The time and concen-tration required should be proportional tothe difficulty and scope of the issuesinvolved.More realistic goal: to achieve a fuller and

wider understanding of the problem.DO start with smaller-scale problems.DON’T attempt to diagram the whole

system—otherwise you’ll quickly becomeoverwhelmed.Better: Try to focus on a problem issue

and draw the minimum variables and loopsyou’ll need to capture the problem.

DON’T work with systems thinkingtechniques “on line” under pressure, or infront of a group that is unprepared for orintolerant of the learning process.Additional danger: If the audience is not

familiar with the concepts and methods ofsystems thinking, they might not under-stand that the process reveals mental mod-els, can be controversial, and is highlyiterative in nature. It is far more beneficialto have the group engage in their own loopbuilding after appropriate instruction andfoundation have been given.

DO develop your diagrams graduallyand informally, in order to build confidencein using systems thinking.

U N I C A T I O N S , I N C . 7 8 1 . 3 9 8 . 9 7 0 0

effective, an intervention must be self-, self-correcting and long-lasting. Ite long-term changes in the perfor-nd.of interventions in a causal loop

a link.k a link.ten a delay.

a goal explicit.down a growth process; relieve ag process.st intervention is likely to be a combi-

interventions applied gently and

pushing on a structure from the out-

or variance between long- and short-cts, to anticipate unexpected effects.

G U I D E L I N E S F O RI N T E R V E N T I O N S

Good practice: Look at newspaper arti-cles and try to draw a few loops that capturethe dynamics of a problem being described.Even better: Try matching a template to thearticle.

DON’Tworry about drawing loopsright away. One of the strongest benefits ofthe systems thinking perspective is that it canhelp you learn to ask the right questions.This is an important first step toward under-standing a problem.

DRAW ING D IAGRAMS

DO start with the process of defining vari-ables.DO encourage airing of assumptions.Benefit: better shared understanding of a

problem. Diagramming is a very effectivetool for promoting group inquiry into aproblem or issue.

DO start with a central loop or process.Then add loops to “fill in” detail.Example: The central loop may show

how the system is supposed to work, and theadditional loops can explore what is pushingit out of whack.

DON’T get bogged down in details.Start simply, at a high level of generaliza-tion, but with enough detail to sum up theobserved behavior.Example: If you are exploring the causes

of missed delivery dates in a factory, lumptogether the types of products that are expe-riencing similar delays.

DO begin by looking for templates orgeneral structures that might clarify theproblem.Advantage: Systems archetypes provide a

focal point or a storyline to begin the pro-cess of understanding a problem.

DOwork with one or more partners.Advantage: Multiple viewpoints add

richness and detail to the understanding of aproblem.

DO check with others to see if they canadd some insight or improve upon yourdiagram—especially people in other func-tional areas who might have a different


perspective on the problem.Example: With a manufacturing delay

problem, you might check with finance tosee if there are any dynamics in the financearena that are affecting the manufacturingdelays (capital investments and purchases,etc.). The same can be done for marketing,sales, etc.

DO work iteratively. There is no “final”model (set of loops). Looping is a learningprocess that should continue to evolve withnew data and perspectives.

DON’T present “final” loop diagramsas finished products.Better: Present as a tentative and evolv-

ing picture of how you are seeing things. Toget buy-in and maximize learning, the audi-ence needs to participate in the modelingprocess.

DO learn from history. When possible,check data to see if your diagram correctlydescribes past behavior.

I N TERVENT IONS

DO get all stakeholders involved in theprocess. This will help ensure that allviewpoints have been con-sidered, and will improvethe acceptance rate for theintervention.

DON’T go for vague,general, or open-endedsolutions such as “Improvecommunications.”Better: “Reduce the

information delay betweensales and manufacturing bycreating a new informationsystem.”

DO make an interven-tion specific, measurable,and verifiable.Example: “Cut the infor-

mation delay between salesand manufacturing down to24 hours.”

DO look for potential

1. To besustainingmust makmance tre2. Typesdiagram:

• Add• Brea• Shor• Make• Slow• limitin

3. The benation ofpatiently.4. Avoidside.5. Look fterm impa


unintended side-effects of an intervention.General principle: “Today’s problems

often come from yesterday’s solutions.”Any solution is bound to have trade-offs,so use systems thinking to explore theimplications of any proposed solutionbefore trying to implement it.

DON’T be surprised if some situationsdefy solution, especially if they are chronicproblems. Rushing to action can thwartlearning and ultimately undermine effortsto identify higher leverage interventions.Resist the tendency to “solve” the issue andfocus on gaining a deeper understanding ofthe structures producing the problem. Bewary of a symptomatic fix disguised as along-term, high-leverage intervention.

Michael Goodman is a principal of InnovationAssociates, Framingham, MA, an Arthur D. Littlecompany.The material in this article was drawn fromhis 20 years of experience in the field, as well asbusiness courses developed by InnovationAssociates.

•



1. Double-Q DiagramBased on TQC tool “Cause-and-Effect Diagram.” See

Ishikawa, Kaoru (1982) Guide to Quality Control, Ann Arbor,MI: UNIPUB.

2. Behavior Over Time DiagramBased on diagrams referred to as “reference modes” in sys-

tem dynamics literature. See Richardson, George and AlexanderPugh (1981) Introduction to System Dynamics Modeling,Waltham, MA: Pegasus Communications.

3. Causal Loop DiagramSee Richardson, George and Alexander Pugh (1981)

Introduction to System Dynamics Modeling, Waltham, MA:Pegasus Communications.

4. Systems ArchetypesSee Senge, Peter (1990) The Fifth Discipline, New York:

Doubleday. Also covered regularly with current business appli-cations in a management newsletter, THE SYSTEMSTHINKER, published by Pegasus Communications, Inc.,Waltham, MA.

5. Graphical Function DiagramBased on diagrams referred to as “table functions” in system

dynamics literature. See Richardson, George and AlexanderPugh (1981) Introduction to System Dynamics Modeling,Waltham, MA: Pegasus Communications.

6. Structure-Behavior PairsReferred to as “Atoms of Structure” in Academic User’s Guide

to STELLA by Barry Richmond, published (as part of softwaredocumentation) by High Performance Systems, Hanover, NH.Also, see Goodman, Michael (1974) Study Notes in SystemDynamics, Waltham, MA: Pegasus Communications.

7. Policy Structure DiagramContact Professor John Morecroft at the London Business

School.

8. Computer ModelOne of the best software for building system dynamics com-

puter models (Macintosh) is ithink™ and STELLA™ by HighPerformance Systems, Hanover, NH. For IBM-compatibles,there is Vensim by Ventana Systems, and PowerSim StudioEnterprise 2000 by PowerSim Corp.

9. Management Flight SimulatorsContact Professor John Sterman at the M.I.T. Sloan School

of Management (617-253-1951) for copies of computer simula-tors on People Express, managing product lifecycles, real-estatemanagement, and super tanker management.

10. Learning LaboratoryKim, Daniel (1989) “Learning Laboratories: Designing a

Reflective Learning Environment,” Proceedings of the 1989International System Dynamics Conference, Stuttgart,Germany: Springer-Verlag.

F U R T H E R R E A D I N G O N T H E 1 0 T O O L S


I N D E X T O T H E S Y S T E M S T H I N K E R

PART I: AN OVERVIEW

6 Systems Thinking as a Language by Michael R. Goodman V2N3, April 19918 Levels of Understanding: “Fire-Fighting” at Multiple Levels V4N5, June/July 199310 A Palette of Systems Thinking Tools V1N3, August 1990

13 PART II: DYNAMIC THINKING TOOLS14 Reinforcing and Balancing Loops: Building Blocks of Dynamic Systems V1N1, April/May 199016 Balancing Loops with Delays: Teeter-Tottering on Seesaws V1N2, June/July 199018 Guidelines for Drawing Causal Loop Diagrams V3N1, February 199220 Systems Archetypes at a Glance V3N4, May 1992

23 PART III: STRUCTURAL THINKING TOOLS24 From Causal Loops to Graphical Functions: Articulating Chaos V2N8, October 199126 Graphical Functions: “Seeing” the Full Story V2N7, September 199128 Structural Thinking: The World According to Accumulators and Flows V2N2, March 199130 Accumulators: Bathtubs, Bathtubs, Everywhere V2N1, February 199132 Accumulation Management: Avoiding the “Pack Rat” Syndrome V2N4, May 199134 Delays: Accumulators in Disguise V2N5, June/July 199136 S-Shaped Growth and the Law of Diminishing Returns V2N3, April 1991

39 PART IV: COMPUTER-BASED TOOLS40 Modeling for What Purpose? by Jay W. Forrester V4N4, May 199342 Management Flight Simulators: V3N9, November 1992

Flight Training for Managers—Part I44 Management Flight Simulators: V3N10, Dec 1992/Jan 1993

Flight Training for Managers—Part II46 Learning Laboratories: Practicing Between Performances V3N8, October 1992

49 PART V: REFERENCE GUIDE50 The Vocabulary of Systems Thinking: A Pocket Guide V2N10, Dec 1991/Jan 199252 The Do’s and Don’t’s of Systems Thinking on the Job V3N6, August 1992

by Michael R. Goodman


A B O U T T H E T O O L B O X R E P R I N T S E R I E S

Systems Thinking Tools: A User’s Reference Guide is a volume in the Toolbox Reprint Series. Other volumesinclude Systems Archetypes I: Diagnosing Systemic Issues and Designing High-Leverage Interventions, SystemsArchetypes II: Using Systems Archetypes to Take Effective Action, Systems Archetypes III: Understanding Patternsof Behavior and Delay, and The “Thinking” in Systems Thinking: Seven Essential Skills. All volumes are avail-able for $16.95 each. As these booklets are often used in training and introductory courses, volumediscounts are available. See the copyright page or call 1-800-272-0945 for details.

The Toolbox Reprint Series has been compiled from The Systems Thinker® Newsletter, which presents asystems perspective on current issues and provides systems tools for framing problems in new and insightfulways. The Systems Thinker includes articles by leading systems thinkers, case studies of systems thinkingimplementation, software and book reviews, and numerous other columns geared to different levels ofsystems thinking ability. Individual subscription rates to The Systems Thinker are as follows: A special rate of20 issues for $169, or 10 issues (one year) for $109. Library subscriptions are available for $189 for 10 issues(one year). Visit www.pegasuscom.com for more information.



P E G A S U S P U B L I C A T I O N S

AnthologiesManaging the Rapids: Stories from the Forefront of the Learning OrganizationReflections on Creating Learning OrganizationsThe New Workplace: Transforming the Character and Culture of Our OrganizationsOrganizational Learning at Work: Embracing the Challenges of the New WorkplaceMaking It Happen: Stories from Inside the New Workplace

The Pegasus Workbook SeriesSystems Archetype Basics: From Story to Structure Systems Thinking Basics: From Concepts to Causal LoopsThe Learner’s Path: Practices for Recovering Knowers

The “Billibonk” SeriesBillibonk & the Thorn Patch Frankl’s “Thorn Patch” FieldbookBillibonk & the Big Itch Frankl’s “Big Itch” Fieldbook

Learning FablesOutlearning the Wolves: Surviving and Thriving in a Learning OrganizationShadows of the Neanderthal: Illuminating the Beliefs That Limit Our OrganizationsThe Lemming Dilemma: Living with Purpose, Leading with VisionThe Tip of the Iceberg: Managing the Hidden Forces That Can Make or Break Your OrganizationListening to the Volcano: Conversations That Open Our Minds to New Possibilities

Other TitlesHuman Dynamics: A New Framework for Understanding People and Realizing the Potential in Our OrganizationsWhen a Butterfly Sneezes: A Guide for Helping Kids Explore Interconnections in Our World Through Favorite Stories

The Innovations in Management SeriesFrom Mechanistic to Social Systemic Thinking: A Digest of a Talk by Russell L. AckoffApplying Systems ArchetypesToward Learning Organizations: Integrating Total Quality Control and Systems ThinkingDesigning a Systems Thinking Intervention: A Strategy for Leveraging ChangeThe Natural Step: A Framework for Achieving Sustainability in Our OrganizationsAnxiety in the Workplace: Using Systems Thinking to Deepen UnderstandingThe Soul of Corporate Leadership: Guidelines for Values-Centered GovernanceCreating Sustainable Organizations: Meeting the Economic, Ecological, and Social Challenges of the 21st CenturyCreating Value: Linking the Interests of Customers, Employees, and InvestorsRelinking Life and Work: Toward a Better FutureFacing the Competition: An Organization Mobilizes for Large-Scale ChangeOrganizational Change at Philips Display Components: Reflections on a Learning JourneyIntroduction to Systems ThinkingRebounding, Rebuilding, Renewing at Shell Oil: A Former CEO Reflects on Large-Scale ChangeReinventing Human Resources at L.L. Bean: Lessons for Learning and ChangeThe Essentials of Servant-Leadership: Principles in PracticeDialogue at Work: Skills for Leveraging Collective Intelligence

The Toolbox Reprint SeriesSystems Archetypes I: Diagnosing Systemic Issues and Designing High-Leverage InterventionsSystems Archetypes II: Using Systems Archetypes to Take Effective ActionSystems Archetypes III: Understanding Patterns of Behavior and DelaySystems Thinking Tools: A User’s Reference GuideThe “Thinking” in Systems Thinking: Seven Essential Skills

E-NewslettersTHE SYSTEMS THINKER® LEVERAGE POINTS® for a New Workplace, New World



Pegasus Communications, Inc. is dedicated to providing resources that help people explore, understand, articulate, andaddress the challenges they face in managing the complexities of a changing world. Since 1989, Pegasus has worked tobuild a community of practitioners through newsletters, books, audio and video tapes, and its annual Systems Thinkingin Action® Conference and other events. For more information, contact us at:Pegasus Communications, Inc. • One Moody Street • Waltham, MA 02453-5339 USA • www.pegasuscom.comOrder Phone: 800-272-0945 / 781-398-9700 • Fax: 781-894-7175



T O O L B O X REP R INTS ERI E S SYSTEMS - The ... O O L B O X REP R INTS ERI E S A User’s...

Documents

Transcript of T O O L B O X REP R INTS ERI E S SYSTEMS - The ... O O L B O X REP R INTS ERI E S A User’s...