You are on page 1of 46

Applying System Dynamics to Manage Dynamic Complexity in Enterprises; by Jose J Gonzalez; professor dt.techn., dr.rer.nat.

; AUC

Agder University College

Objectives
Present System Dynamics as methodology to
identify, define, model enterprise challenges characterized by dynamic complexity

Communicate how system dynamic simulations serve to


explore scenarios, test policies, identify robust strategies, provide insights that lead to organizational learning

Exemplify by means of real-life cases


(mis)managing traffic pollution boost and bum in semiconductor industry time and cost overruns in large-size projects erosion of security and safety standards dealing with volatility and uncertainty in offshore

Presented at Complexity Seminar in Lund, November 4-5, 2002

Issues
Agder University College

1. Characteristics of Enterprise Challenges

2. Dynamic Complexity The Logic of Failure 3. System Dynamics Methods and Applications
4. Learning in Complex Domains 5. Organizational Learning
Presented at Complexity Seminar in Lund, November 4-5, 2002
2

Issues
Agder University College

1. Characteristics of Enterprise Challenges 2. Dynamic Complexity: The Logic of Failure


Examples of Enterprise Challenges Analysis Main Conclusions

3. System Dynamics Methods and Applications

4. Learning in Complex Domains


5. Organizational Learning
Presented at Complexity Seminar in Lund, November 4-5, 2002
3

Characteristic of Enterprise/Public Challenges:


Agder University College

Consider the following enterprise (or even public) challenges:


(Mis)managing traffic pollution in Mexico city Boom and bust in semiconductor industry Cost & time overruns and quality problems in large-scale projects Ubiquitous erosion of safety & security standards, making companies and nations vulnerable (organizational accidents, cyberwar, terrorism) Rig management in offshore companies (specifically, Statoil) fronting high risks (hugh investments per rig, volatile oil prices, unpredictable demand, unsafe conditions, emerging technologies)
Presented at Complexity Seminar in Lund, November 4-5, 2002
4

Characteristic of Enterprise/Public Challenges:


Managing traffic pollution
Agder University College

Traffic pollution in Mexico city:


Air pollution in Mexico City is amongst the worst in the world The authorities decided to limit vehicle use every car has a color-code, and for one workday a week is banished The expected result was a 20% reduction in car usage on weekdays there now seems more cars than ever, and they seem to be producing ever increasing pollution {Link to Causal-loop analysis explains why} Such behavior is known as policy resistance. It is a typical outcome when planning ignores the propagation of effects and the impact of (counteracting) feedback.

Presented at Complexity Seminar in Lund, November 4-5, 2002

Characteristic of Enterprise Challenges:


Boom and Bust
Agder University College

Boom and bust in semiconductor industry:


An international diversified company was forced to write down several hundred million dollars in investments in semiconductor capacity New entrants were eager to capitalize on the buoyant market, which was exaggerated by perverse buying practices by the customers In just a few years, that industry went from boom to bust from acute shortage to book-to-build ratios of only 70% at the trough {Link to Causal-loop analysis explains why} Among the crucial errors committed was failure to distinguish between perceived and real demand and to account for the impact of delays
Presented at Complexity Seminar in Lund, November 4-5, 2002
7

Characteristic of Enterprise Challenges:


Large-scale projects
Agder University College

Cost & time overruns and quality problems in large-scale projects:


Large-scale projects (e.g. design & construction of civil works & infrastructure, development of complex software or new products, military projects) are consistently mismanaged Typical for commercial projects: 140% costs & 190% time overruns for military projects: 310% costs & 460% time overruns. {Link to Famous case: Ingalls Shipbuilding, USA} Among the crucial errors committed was failure to consider the impact of propagations of delayed effects and to distinguish between perceived and real project progress

Presented at Complexity Seminar in Lund, November 4-5, 2002

10

Characteristic of Enterprise Challenges:


Erosion of standards
Agder University College

Ubiquitous erosion of safety & security standards, making companies and nations vulnerable (organizational accidents, cyberwar, terrorism) :
Human failure accounts for 70-90% of organizational accidents and security problems but human failure must be seen as interacting with technology and working environment. Rich variety of causes: priority conflicts, human behavior economics, shrinkage of viable actions as system is patched, and last not least reinforcing of wrong attitudes modulated by risk misperception {Link to Causal loop analysis shows why} Crucial causes of the erosion of standards are misperception of risk and superstitious learning apparent (but not real) empirical confirmation of misperceptions and wrong causal attributions

Presented at Complexity Seminar in Lund, November 4-5, 2002

15

Characteristic of Enterprise Challenges:


Rig management
Agder University College

Rig management in offshore companies (specifically, Statoil) :


Hugh & risky investments for rig brokers rigs costs typically 1 billion USD, take ca. 3 yr to build, financing groups demand, up-front 70% rig leasing within 5 yr to cover financial risks, emerging competing, technologies, changing safety legislation Users offshore companies risk volatile oil prices (between 10 and 30 UDS pr barrel), uncertain profitability of lots, variety of operational conditions (tasks, climate, depth), and large price differences between long-term and spot rig leasing, overruns of offshore project costs and times. Hence, unpredictable long-term demand for rig brokers and unpredictable long-term supply for offshore companies. Analysis shows that most aspects of The Logic of Failure are involved:
Complexity challenges related to big delays, propagation of effects, uncertain external conditions, long time intervals up to 30 yr , hugh financial stakes, misperception of feedback in short, most of the features identified as failure factors (Dietrich Drner: The Logic of Failure)
Presented at Complexity Seminar in Lund, November 4-5, 2002
17

Issues
Agder University College

1. Characteristics of Enterprise Challenges

2. Dynamic Complexity The Logic of About Dynamic Failure Complexity


3. System Dynamics Methods and Applications 4. Learning in Complex Domains 5. Organizational Learning
Presented at Complexity Seminar in Lund, November 4-5, 2002
18

The Logic of Failure

Dynamic Complexity The Logic of Failure


Agder University College

There are two kinds of problem complexity:


Combinatorial, a.k.a. detail complexity (many components and relationships) Dynamic complexity (complex behavior over time) The major challenge is dynamic complexity, found in nonlinear systems, because it poses tremendous challenges: The unaided mind is very poor at predicting the time development of non-linear systems, even if they only have a few components Failure to deal with future developments has crucial consequences for companies: Over one third of the Fortune 500 largest companies in 1970 had disappeared 13 years later (Arie de Geus: The Living Company)

Presented at Complexity Seminar in Lund, November 4-5, 2002

19

Dynamic Complexity The Logic of Failure


Agder University College

Research by Drner et al. about thinking, decision-making and acting in complex domains: Most people fail and the behavior patterns are (quite) universal but a few master complexity. Drner found determinants of human failure:
Linear thinking fails to account for propagation & ramification of effects Poor ability to perceive & understand feedback (misperception of feedback, wrong causal attribution), hence policy resistance Ignoring time delays, wrongly assigning causes to events close in time and space Problems to perceive nonlinear growth and decay Encapsulation falling in love with a particular aspect, ignoring other, often much more important aspects Thematic vagabonding unfocused, poorly structured thinking Etc
Presented at Complexity Seminar in Lund, November 4-5, 2002
20

Issues
Agder University College

1. Characteristics of Enterprise Challenges

2. Dynamic Complexity The Logic of Failure


3. System Dynamics Methods and Applications 4. Learning in Complex Domains 5. Organizational Learning
Presented at Complexity Seminar in Lund, November 4-5, 2002

About System Dynamics Model development Modeling perceptions & delays Structure and behavior Types of system dynamics models Integrated Solutions

21

About System Dynamics


Agder University College

System Dynamics is a discipline explicitly designed to manage systems characterized by:


nonlinear dynamics, feedback, time delays, soft factors, interdisciplinary aspects

Founded 1957 by Jay W. Forrester as extension of control theory/cybernetics to management


Later succesfully applied to all kind of complex dynamic systems, involving psychological, social, technological or even environmental aspects
Presented at Complexity Seminar in Lund, November 4-5, 2002
22

Qualitative System Dynamics


Agder University College

Qualitative System Dynamics employs causal loop diagrams to explain the likely mechanism of complex phenomena, such as attempts to manage traffic pollution in big cities or boom and bust in high-velocity industries. At this level, causal loop diagrams explain cause-effect influences by an arrow pointing from cause to effect. No indications of strength nor or type (i.e. direct impact, cumulative impact, etc.) of the effect are given.
Even at this simple level, causal-loop diagrams can qualitatively explain phenomena, or even if the causal-loop diagram is designed in advance prevent the decision-maker from costly mistakes and suggest better measures to manage the system.

To understand the relationship between (causal) structure and dynamic behavior one needs quantitative methods, i.e. System Dynamics proper.
Presented at Complexity Seminar in Lund, November 4-5, 2002
23

System Dynamics Methods


Agder University College

As methodology, System Dynamics spans from knowledge capture & problem articulation to scenario & policy analysis and improvement of organizational knowledge. System Dynamics is best understood as an eclectic methodology a joint venture of disciplines borrowing methods and tools from other disciplines and amalgamating interdisciplinary sources of knowledge, such as:
Methods: Data mining, statistical parameter estimation, econometric methods, optimization, risk assessment & management Disciplines: Nonlinear numerical methods, control theory & cybernetics, management science, economics, psychology, group dynamics, supply & value chain science, organizational learning,

System dynamics models can be stand-alone, but leading tool developers (High Performance Systems, Powersim Corporation, Ventana Systems) provide a variety of interfaces to other tools (API, OCR, ASP, etc).
Presented at Complexity Seminar in Lund, November 4-5, 2002
24

Model development
Agder University College

Model development involves the following activities (that can be iterated):


Problem definition and articulation
Who cares and why? Problem symptoms Desired behavior Policy behavior

Audience; model purpose and uses System boundary Model conceptualization


Articulating issues, identifying variables, sketching causal loop diagrams, formulating a dynamic hypothesis

Designing model with software tool, e.g. Powersim Studio Verifying and validating model Tuning model Testing model looking for policies Optimization, risk assessment, risk management last, not least, organizational learning
25

Presented at Complexity Seminar in Lund, November 4-5, 2002

System Dynamics: Stock-and-Flow Diagrams


Agder University College

System dynamic models are visualized through diagrams, the icons stocks, flows, auxiliary variables and constants having semantic content, i.e. specific topological and mathematical properties.

Constants Stock, cumulated (actually by inflows and parameters) de-cumulated by outflows Information links, Model sectoran expressing Flow, here dependencies inflow Auxiliary variables

Presented at Complexity Seminar in Lund, November 4-5, 2002

26

System Dynamics: Stock-and-Flow Diagrams


Agder University College

System dynamic models typically contain physical processes, information flow, human aspects, soft factors, formation of perceptions and expectations and delays.

Physical Information flow, processes, i.e. how e.g.Workforce how desired staff comes in and workforce affects adjustment time out of the project depends on human hiring decisions and market conditions

Presented at Complexity Seminar in Lund, November 4-5, 2002

27

System Dynamics: Stock-and-Flow Diagrams

System dynamic models typically contain physical processes, information flow, human aspects, soft factors, formation of perceptions and expectations and delays.
Show model
Presented at Complexity Seminar in Lund, November 4-5, 2002

Agder University Formation of College perception: soft factors (time to perceive productivity), soft relationships (formation of expectation)

28

System Dynamics: Modeling Perceptions and Delays


Agder University College

Human behavior and decision-making is based on perceptions of reality rather than reality itself. Examples:
Link to Boom and bust in high-velocity industries Link to Project management Link to Erosion of security standards

Presented at Complexity Seminar in Lund, November 4-5, 2002

29

Modeling Perceptions
Agder University College

How does a project manager assess the productivity of staff? In a large-scale project one has several important factors affecting productivity:
Tasks apparently completed are reported and accepted by management as being completed further down the road some of the tasks turn out to be faulty and must be reclassified as rework Existing staff experience increases, thus leading to higher productivity New hires dilute experience and require counseling from experienced staff, both aspects decreasing average productivity

All these factors generate information that changes the project managers perception of staff productivity. Perception can be seen as a smoothing of information (Change in perceived productivity) with a characteristic (individually different) time constant (Smoothing time):

Show model
Presented at Complexity Seminar in Lund, November 4-5, 2002
33

Structures and Behavior


Agder University College

Structure drives model behavior over time

Events

Issue Identification and Brainstorming


Historical Results and Patterns of Behavior Simulation

Behavior

Structure

Presented at Complexity Seminar in Lund, November 4-5, 2002

34

Feedback and Behavior


Agder University College

Feedback loops are linked to specific kinds of behavior


Basic Behavior Patterns

Diverging

Oscillations

Converging

S-Shaped

All behavior involving feedback is made up of combinations of these behavior patterns.


Presented at Complexity Seminar in Lund, November 4-5, 2002
35

Diverging behavior
Agder University College

Created by positive feedback loops

The higher the population, the more births, which in turn leads to increased population (over time) Debt with compounding interest (no installments)

Presented at Complexity Seminar in Lund, November 4-5, 2002

36

Converging behavior
Agder University College

Created by negative feedback loops

Production gradually empties reservior, causing reservior pressure to drop and production to decline
The higher the quality gets, the more difficult it gets to increase the quality further

Presented at Complexity Seminar in Lund, November 4-5, 2002

37

Oscillating behavior
Agder University College

Created by negative feedback loop involving major delay

Inventories typically fluctuate since it takes time before a decision to correct the inventory will result in new products being received (production and delivery delays).

Presented at Complexity Seminar in Lund, November 4-5, 2002

38

S-shaped behavior
Agder University College

Caused by shift in feedback loop dominance from a positive loop to a negative loop

Negative loop

Positive loop

Phase 1

Phase 2

In the first phase sales grow exponentially due to the word-ofmouth effect.
As the market gets saturated, sales decline.
Presented at Complexity Seminar in Lund, November 4-5, 2002
39

Types of System Dynamics Models:


Managerial View of the Enterprise
Strategic
Agder University College

Tactical
From 25,000 From 10,000

Operations
From 1,000

Jump to Learning in Complex Domains

From Days

Length of simulation run

To Years

Presented at Complexity Seminar in Lund, November 4-5, 2002

40

Why Business Simulation?


Agder University College

Objectives of business simulations


Simulation Purpose/Use

High

Levels of Management Planning & Decision-making

Value Communication
{Designed for Use Once or Twice}

Management Training
{Designed for Periodic Use}

Integrated Decision Support


{Designed for Continuous Use}

Decision Complexity

Strategic
(Planning) {Long-term}

$$$
Strategy Simulation

$$$$
Strategy Simulator

$$$$$$
Strategic Planning

+
Tactical
(Control) {Medium-Term}

$$
Tactical Simulation

$$$
Tactical Simulator

$$$$
Tactical Planning

+
Operational
(Execution) {Short-term}

$
Operational Simulation

$$
Operational Simulator

$$$
Operational Planning

Low

Development Complexity

High
41

Presented at Complexity Seminar in Lund, November 4-5, 2002

Varieties of business simulations


Agder University College

Different types of business simulators for use at various levels of the organizational structure.

Clevel

Value Communication Simulators

Integrated DecisionSupport Simulators

Middle managers

Department Managers

Suppliers Customers

Training Simulators
Line Supervisors & Systems Operators

Stakeholders

Presented at Complexity Seminar in Lund, November 4-5, 2002

42

Issues Domain
Agder University College

High

Levels of Management Planning & Decision-making

Issues Domain
Facility Planning, Risk Assessment E-Business Change Mgt & Growth Strategies Asset & Portfolio, Shareholder Value Mgt Enterprise security & safety HR & Knowledge Mgt Corporate Planning & Strategy Strategic Alliances Emerging Markets & Tech. Supply & Value chain Mgt

Decision Complexity & Risk Magnitude

Strategic
(Planning) {Long-term}

Tactical
(Control) {Medium-Term}

Project Mgt
Inventory Control & Mgt Production & Distribution Mgt Process Analysis

Product & Marketing Strategy


Satisfaction Measurement Capacity Adjustment Scheduling Cost/Benefit & Yield Analysis Cycle Time Analysis Market Analysis & Forecasting

Operational
(Execution) {Short-term}

Financial Analysis Performance Measurements Quality Measurement

Low

Presented at Complexity Seminar in Lund, November 4-5, 2002

43

Implementation Process
1 2 3 4 5 6 7 N (Weeks/Months) Agder

Mths/Yrs University College 1-12/1-5

Knowledge Transfer Knowledge Extraction System Diagnosis


Determine model use/ purpose Establish connection btw model & everyday realities of managerial life Determine how the model can be integrated into the political, cultural, & managerial values of the firm Specify study objectives & define system boundary Specify model assumptions & performance measures Determine alternative scenarios to investigate Search, discover, & interpret facts Describe the system to be simulated & trace effects back to causes

Knowledge Representation Model Formulation


Define system components Identify & classify system variables Specify experimental design initial system conditions, parameter values, reference modes, etc.

Knowledge Presentation Design & Build SLE


Design the look & feel of the GUI Build GUI VB, DHTML / Java Script, Active-X, etc. programming

Knowledge Delivery Business Policy Analysis


Simulate the model under different assumptions to generate the system behavior through time Compare results with available knowledge about the actual system Redesign organizational relationships and policies that can be altered in the actual system Conduct enough scenario runs to evaluate all known alternatives Recommend suitable line of action to be followed

Knowledge Maintenance

System Update Agreement to:


Conduct routine business scenario simulations Create new model structures to include business changes Extend the scope of model to address new business issues

Design & build Active Server Pages (ASP) objects Create simulation model of the decision policies, information Design & build database* sources, and interactions of Integrate the simulation the system components model with the GUI Prepare input data and Integrate audio & video parameters files Validate model structure and Test the integrated behavior simulation model for Formulate experimental consistency and validity conditions Integrate application into Conduct initial policy test runs client systems Tune and optimize the model Test application on different platforms

Conceptual Model

Simulation model

Integrated Training / Decision Support Tool

Management decisions guided by knowledge

Sustainable Knowledge
44

Presented at Complexity Seminar in Lund, November 4-5, 2002

The Decision Circle


Agder University College

Business

System

Implement Strategy

Data collection

Simulate
Model

Business Model

Compare/ Evaluate

Analyze Analyze
Presented at Complexity Seminar in Lund, November 4-5, 2002
45

Examples of Integrated Solutions:


SEM-BPS Dataset
Agder University College

SEM-BPS dataset provides realistic input data to business simulations


Industry Specific Models Company specific Business Model

Marketing & Sales

Customer Needs

Distribution

Financial

...

Build model

Tune & Optimize

Enterprise Data Warehouse

Enterprise -wide data Reporting Templates

Simulated results

29

Presented at Complexity Seminar in Lund, November 4-5, 2002

46

Examples of Integrated Solutions:


Data Manager
Agder University College

Data Manager approach lets users connect to databases and import/export Powersim variables. Simple, custom-built control panel gives capability to:
send database info to Studio at the start of a time step, advance the Powersim simulation model, and transfer data back from Studio to the database.

Connects to any SQL/ODBC database (e.g. Oracle).

Uses a mapping database (implemented with MS Access) to link database queries/fields to Powersim variables. Implemented in Visual Basic.

Database Server (Unix)

Desktop Computer Powersim Data Manager


ODBC ODBC

Mapping Database

PSAPI

Oracle Server

Oracle ODBC Client

Studio Model

Presented at Complexity Seminar in Lund, November 4-5, 2002

47

Examples of Integrated Solutions:


Web delivery
Agder University College

The interface is a mix of DHTML and JavaScript


Client

User Interface DHTML/JavaScript

All communication between client and server is HTTP


Active Server Pages (ASP) is used to control the server objects The UI Dependent Objects implement all business logic for the UI objects The PS Model objects are used to access Engine. The Data Objects are used to ensure object persistence and for historical and live data. Powersim Engine runs 1..n instances of a simulation requested by the PS Model Objects

HTTP

Presentation Tier

ASP Interface Server Side VBScript UI-centric Objects Server installed DLL PS Model Objects Server installed DLL Data-centric Objects Server installed DLL OLE DB/ADO Business Tier

COM/DCOM
Powersim Engine Server installed OCX and model file
Server

Enterprise Databases Representation Tier

Presented at Complexity Seminar in Lund, November 4-5, 2002

48

Issues
Agder University College

1. Characteristics of Enterprise Challenges

2. Dynamic Complexity The Logic of Failure 3. System Dynamics: Methods and Applications

Single-loop learning Double-loop learning Virtual worlds and doubleloop learning

4. Learning in Complex Domains 5. Organizational Learning


Presented at Complexity Seminar in Lund, November 4-5, 2002
49

Single-loop Learning
Agder University College

Reality domain

Decisions

Information feedback

Policy

Mental model of reality domain


50

Presented at Complexity Seminar in Lund, November 4-5, 2002

Double-loop Learning
Agder University College

Reality domain

Decisions

Information feedback

Policy

Mental model of reality domain


51

Presented at Complexity Seminar in Lund, November 4-5, 2002

Virtual Worlds and Double-loop Learning


Agder University College

Reality domain

Virtual world Decisions

Information feedback

Policy
Presented at Complexity Seminar in Lund, November 4-5, 2002

Mental models
52

Issues
Agder University College

1. Characteristics of Enterprise Challenges

2. Dynamic Complexity The Logic of Failure 3. System Dynamics: Methods and Applications
Fragmentation of Knowledge Group Modeling and Knowledge Capture Shared knowledge Memory of the Future Improving Mental Models

4. Learning in Complex Domains 5. Organizational Learning


Presented at Complexity Seminar in Lund, November 4-5, 2002
53

Organizational Learning: Fragmented Knowledge


Agder University College

Can anyone of you make a humble pencil? (In the sense of setting up a pencil factory from scratch in a new planet with the same resources the Earth to be colonized with an expedition on a spaceship.)

Can anybody on Earth solve that task?


No! A wonderful essay (I pencil by Leonard E Read see http://209.217.49.168/vnews.php?nid=316) convincingly shows that no one knows how to make a pencil. Rather, hundreds of thousands of different knowledge fragments have to be pulled together by all kind of mechanisms: teamwork, market mechanisms, demand & supply, etc in order to make a pencil or by that matter any product. Knowledge is fragmented. The great economist Friedrich von Hayek wrote:
Economics has long stressed the division of labour But it has laid much less stress on the fragmentation of knowledge, on the fact that each member of society can have only a small fraction of the knowledge possessed by all, and that each is therefore ignorant of most of the facts on which the working of society rests. Yet it is the utilisation of much more knowledge that anyone can possess, and therefore the fact that each moves within a coherent structure most of whose determinants are unknown to him, that constitutes the distinctive feature of all advanced civilisations.
Presented at Complexity Seminar in Lund, November 4-5, 2002
54

Organizational Learning:
Group Modeling and Knowledge Capture
Agder University College

Enterprise challenges mostly span across many fragmented knowledge domains, including knowledge found outside of the enterprise proper.
Hence, group modeling processes are necessary

In addition, much is still unknown. Hayek again:


Complete rationality of action demands complete knowledge of all the relevant facts. A designer or engineer needs all the data and full power to control or manipulate them if he is to organize the material objects to produce the intended result. But the success of any action in society depends on more particular facts than anyone can possibly know. And our whole civilization in consequence rests, and must rest, on our believing much that we cannot know to be true Implying that data mining, knowledge capturing processes, including discovey processes are needed and that a substantial proportion of assumptions (beliefs) must be made.

Presented at Complexity Seminar in Lund, November 4-5, 2002

55

Organizational Learning:
Shared Knowledge and Memory of the Future
Agder University College

The very development of a system dynamic model of an enterprise challenge leads to shared knowledge for the client. System dynamic models should not be used as predictive tools rather, they are tools to explore scenarios (answering what-if questions), thus creating Memory of the Future (term coined by the Lund neurologist, professor Dr David Ingvar, *1924, 2000). The richer such Memory of the Future (e.g. by identifying robust policies those working under a wide variety of conditions), the better. Ultimately, the objective is improving mental models:
Presented at Complexity Seminar in Lund, November 4-5, 2002
56

Organizational Learning:
Improving Mental Models
Agder University College

Models should not be used as a substitute for critical thought, but as a tool for improving judgment and intuition Improving the mental models upon which decisions are based is the proper goal of computer modeling. John D. Sterman: A Skeptics Guide to Computer Models

Presented at Complexity Seminar in Lund, November 4-5, 2002

57

You might also like