Upload
vongoc
View
216
Download
1
Embed Size (px)
Citation preview
A Case Study in Predictive ProgramA Case Study in Predictive ProgramManagementManagement
John Welsh, Bipin Chadha, Zach HoriatisJohn Welsh, Bipin Chadha, Zach Horiatis
Lockheed MartinLockheed Martin
Advanced Technology LaboratoriesAdvanced Technology Laboratories
Mission CriticalMission CriticalEnterprise SystemsEnterprise SystemsSymposiumSymposium
Oct. 2004Oct. 2004
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 2
ProblemProblem Major programs oftenMajor programs often
encounter major cost andencounter major cost andschedule problems despiteschedule problems despitedetailed and comprehensivedetailed and comprehensiveplansplans
““Despite their superiority, weapon systemsDespite their superiority, weapon systemsroutinely take much longer to field, costroutinely take much longer to field, costmore to buy, and require more support thanmore to buy, and require more support thaninvestment plans provide for.investment plans provide for.””
Defense Acquisitions: Assessments of MajorDefense Acquisitions: Assessments of MajorWeapon Programs, GAO Report, GAO-03-476,Weapon Programs, GAO Report, GAO-03-476,May 2003May 2003
Complex interrelationships cause control of program cost andComplex interrelationships cause control of program cost andschedule to be difficultschedule to be difficult Program managers need a small set of key levers (vs. thousands of tasks,Program managers need a small set of key levers (vs. thousands of tasks,
schedules, budgets, risks, action items, schedules, budgets, risks, action items, ……))
““U.S. Navy and Air Force plans to stiffen penalties for cost overrunsU.S. Navy and Air Force plans to stiffen penalties for cost overruns””Bloomberg Bloomberg –– 2/5/2004 2/5/2004
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 3
Recognize complex impacts of change at the speed ofRecognize complex impacts of change at the speed ofinformation rather than speed of information rather than speed of ““status reportstatus report”” delivery delivery
What If You Could What If You Could ……
Know that a software project would be delayed beforeKnow that a software project would be delayed beforeit was reportedit was reported
See management data in real-time and play See management data in real-time and play ““what-ifswhat-ifs””
Recognize cross IPT shortfalls and impactsRecognize cross IPT shortfalls and impactsimmediatelyimmediately
See third and fourth level effects of change rather thanSee third and fourth level effects of change rather thanthose immediately apparentthose immediately apparent
Do multi-variant cost sensitivity analysis on the flyDo multi-variant cost sensitivity analysis on the fly
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 4
Levels of UnderstandingLevels of Understanding Events (Reactive)Events (Reactive)
This event happened therefore we must takeThis event happened therefore we must takethis actionthis action
Q3 earnings results announced leading toQ3 earnings results announced leading todrop in stock pricedrop in stock price
Patterns of behavior (Responsive)Patterns of behavior (Responsive) Interpretation of current events in the contextInterpretation of current events in the context
of long-term historical changeof long-term historical change Understand a pattern of behaviorUnderstand a pattern of behavior Product lifecycles follow the Product lifecycles follow the ““SS”” curve curve
Systemic structure (Generative)Systemic structure (Generative) Addresses the question - what causes theAddresses the question - what causes the
pattern of behaviorpattern of behavior Attempt to address the underlying causes ofAttempt to address the underlying causes of
behaviorbehavior-- Senge, P. , The Leader’s New Work: Building Learning Organizations, SMR, Fall, 90
Sta
tist
ical
Met
ho
ds
Sys
tem
ic M
eth
od
s
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 5
ThesisThesis Prediction on large complex programs is feasiblePrediction on large complex programs is feasible
because because ……
Most program Most program ““surprisessurprises”” result from a small number result from a small numberof key elementsof key elements DependenciesDependencies
Systemic behavior patternsSystemic behavior patterns
These are normally not considered in planningThese are normally not considered in planning Hence root causes of program surprise go undetectedHence root causes of program surprise go undetected
Information for effective analysis is straightforward /Information for effective analysis is straightforward /easy to obtaineasy to obtain
Analysis methodology tolerates parameter variationAnalysis methodology tolerates parameter variationand imprecisionand imprecision
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 6
Comprehensive Program ManagementComprehensive Program Management
22ndnd Integrated Management Integrated ManagementFrameworkFramework Linked data elementsLinked data elements Configuration managementConfiguration management WorkflowWorkflow Access controlAccess control Provides views acrossProvides views across
disciplines, supports changedisciplines, supports changemanagement and traceabilitymanagement and traceability
11stst Generation Program Management Generation Program Management IPTIPT’’s s assessments: Predictive but subjectiveassessments: Predictive but subjective Provides drilldown capabilityProvides drilldown capability
33rdrd Generation Predictive Program Management Generation Predictive Program Management System dynamic model for programSystem dynamic model for program
managementmanagement Causal relationships are modeledCausal relationships are modeled Agent assessments: Predictive and objectiveAgent assessments: Predictive and objective Provides Provides ““look ahead capabilitylook ahead capability”” and and ““what ifwhat if””
leverslevers
TechSched Risk Cost ...
Communicate
Sense
Act
Decide IMPIMS
Production
WBS
SOWCost
GFX
LogisticsXBR
Management
IPT1SE&I
Risk
Knowledge
TPMs
Requirements
Joint Analysis
IPT2
IPT3IPT4 IPT5
IPT6
Program Management
LSI
RequirementsRequirementsIMS/IMPIMS/IMP
TPMTPM RiskRisk
WBSWBS SOWSOW
CostCost
EVMSEVMS
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 7
Predictive Program ManagementPredictive Program ManagementEnvironmentEnvironment
Intelligent Mobile Agents
Collaborative Program Management Environment
Management Tools (EVMS, Schedule, etc.)
Prediction Models
StatisticalModels
Design StructureMatrices
System DynamicsModels
InformationQuality
Coupling andDynamics
Internet Backbone
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 8
Communicate
Sense
Act
Decide IMP
IMS
Production
WBS
SOWCost
GFX
LogisticsXBR
Management
IPT1
SE&I
Risk
Knowledge
TPMs
Requirements
Joint Analysis
IPT2
IPT3
IPT4 IPT5
IPT6
Program Management
LSI
System DynamicsModels
ProgramProgramPlanPlan
Program InformationProgram Information
TechSched Risk Cost TechSched Risk Cost
TodayToday
PredictedPredicted
PastPastHistoryHistory
TechSched Risk Cost TechSched Risk Cost
TechSched Risk Cost TechSched Risk Cost
TechSched Risk Cost TechSched Risk Cost
TechSched Risk Cost TechSched Risk Cost
TechSched Risk Cost TechSched Risk Cost
TechSched Risk Cost TechSched Risk Cost
TechSched Risk Cost TechSched Risk Cost
ManagementTeam
What-IfsWhat-IfsRecommenderRecommenderAgentsAgents
Data Gathering/Data Gathering/MonitoringMonitoringAgentsAgents
AlertingAlertingAgentsAgents
FeedbackFeedback
TaskTaskMonitoringMonitoringAgentsAgents
Project ControlDashboard
EVEVScheduleScheduleRisk Risk ……
Architecture ConceptArchitecture Concept
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 9
Coupling / Interdependencies
• Program Elements
— Activities
— Systems
— Organizations
• Dependency Types
— Information
— Physical
— Energy
— Resource
Key Coupling FactorsKey Coupling Factors
Program Element
Pro
gra
m E
lem
ent
Dependency Type
Design Structure Matrix
Common dependency patterns inherent across programs enablea small number of models to represent the dominant behaviors.
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 10
A B C DABCD
AB
CDx
x xx
x
Task 4Task 3Task 2Task 1
A D C BA
DC
Bx
xx
xxA
BCDTask 4
Task 1
Task 2Task 3
ModulesA affects D
DSM Dependency AnalysisDSM Dependency Analysis For each dependency type, matrices are analyzed to find task orFor each dependency type, matrices are analyzed to find task or
system groupings that are most important to watchsystem groupings that are most important to watch Supports automated analysisSupports automated analysis Easy to interpret, visualize, and analyzeEasy to interpret, visualize, and analyze
Rows and columns: Rows and columns: Program elementsProgram elements Matrix entries: Matrix entries: Coupling strengthsCoupling strengths Principle eigenvalues: Key work / design modes; determine the Principle eigenvalues: Key work / design modes; determine the ““raterate
and nature of convergenceand nature of convergence”” in the presence of change in the presence of change Principle eigenvector: Ranks tasks/subsystems by their Principle eigenvector: Ranks tasks/subsystems by their ““relativerelative
impactimpact”” on overall schedule / cost on overall schedule / cost
Dependency analysis highlights the subsets programs whichrequire closer attention / monitoring, or need to be decoupled
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 11
DSM AnalysisDSM Analysis Design mode: Group of tasks that are closely relatedDesign mode: Group of tasks that are closely related
and cause rework. Characterized by eigenvalues andand cause rework. Characterized by eigenvalues andcorresponding eigenvectorscorresponding eigenvectors
Eigenvalues and Eigenvectors determine the rate andEigenvalues and Eigenvectors determine the rate andnature of convergence of processnature of convergence of process
Magnitude of eigenvalues identifies the geometric rateMagnitude of eigenvalues identifies the geometric rateof convergenceof convergence Large positive eigenvalues have greater impactLarge positive eigenvalues have greater impact
Ignore complex eigenvalues with negative real componentIgnore complex eigenvalues with negative real component
Negative and complex eigenvalues describe dampedNegative and complex eigenvalues describe dampedoscillationsoscillations
Eigenvectors characterize relative contribution ofEigenvectors characterize relative contribution oftasks to reworktasks to rework
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 12
Selected Design Modes can beSelected Design Modes can beAnalyzed for Systemic Behavior PatternsAnalyzed for Systemic Behavior Patterns
Dynamic PatternsDynamic Patterns Balancing LoopBalancing Loop
Reinforcing LoopReinforcing Loop
Hidden ReworkHidden Rework
Fixes that BackfireFixes that Backfire
Limits to GrowthLimits to Growth
Shifting the BurdenShifting the Burden
Tragedy of CommonsTragedy of Commons
Accidental AdversariesAccidental Adversaries
Success to the SuccessfulSuccess to the Successful
EscalationEscalation
Drifting GoalsDrifting Goals
The Attractiveness PrincipleThe Attractiveness Principle
Growth and UnderinvestmentGrowth and Underinvestment(core patterns)(core patterns)
Interdependencies /Interdependencies /CouplingCoupling
Program Elements Activities
Systems
Organizations
Dependency TypesDependency Types Informational
Physical
Energy
Resource
Common systemic patterns inherent across programsenable a small number of models to represent
most of the dominant behaviors (80% solution).
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 13
SYSTEMComp1
Comp2Comp3
Comp1
Comp2 Comp3
Systemic BehaviorSystemic Behavior Behavior of a system that emerges due to dynamicBehavior of a system that emerges due to dynamic
interactions among componentsinteractions among components
System characteristicsSystem characteristics System structure drives system behavior (especially theSystem structure drives system behavior (especially the
feedback loops)feedback loops) System exhibits properties and behaviors System exhibits properties and behaviors
different from the partsdifferent from the parts Everything is connected to everythingEverything is connected to everything There are no independent variablesThere are no independent variables Many effects are indirect and delayedMany effects are indirect and delayed Many dependencies are non-linearMany dependencies are non-linear Behavior is dynamicBehavior is dynamic
Systemic Behavior Can be a Significant Source of Risk and Surprise
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 14
System Dynamics Modeling (SDM)System Dynamics Modeling (SDM) ObservationObservation
Problem arises over and over againProblem arises over and over again You apply the fix, there is some relief,You apply the fix, there is some relief,
but the problem surfaces againbut the problem surfaces again
What is going onWhat is going on You are applying a symptomaticYou are applying a symptomatic
solution that alleviates symptomssolution that alleviates symptomsimmediately but exacerbates theimmediately but exacerbates theproblem over timeproblem over time
The problem causes symptoms toThe problem causes symptoms toreappear after some delayreappear after some delay
Since the fundamental solution has aSince the fundamental solution has adelay built into it (before you seedelay built into it (before you seeresults) it forces you to take theresults) it forces you to take thesymptomatic solution in the shortsymptomatic solution in the shortterm, thus trapping you in the viciousterm, thus trapping you in the viciouscyclecycle
Source:The Fifth Discipline Fieldbook by Senge, et al
“Shifting the Burden”Pattern
ProblemSymptom
SymptomaticSolution
SideEffect
B
R
FundamentalSolution
B
ProblemSymptom
SymptomaticSolution
SideEffect
B
R
FundamentalSolution
B
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 15
Where Have These TechniquesWhere Have These TechniquesBeen applied Before?Been applied Before?
DSMDSM Developed at MITDeveloped at MIT Applied in AutomotiveApplied in Automotive
industryindustry Applied on some DoDApplied on some DoD
programsprograms MIT runs regular coursesMIT runs regular courses
on it for programon it for programmanagersmanagers
Significant body ofSignificant body ofliterature exists thatliterature exists thatsupports this techniquesupports this technique
SDMSDM Developed at MITDeveloped at MIT Applied in a wide range ofApplied in a wide range of
industries to analyzeindustries to analyzesystemic problemssystemic problems
Applied on some DoDApplied on some DoDprogramsprograms
Significant body ofSignificant body ofliterature exists thatliterature exists thatsupports this techniquesupports this technique
Several patterns haveSeveral patterns havebeen well documentedbeen well documented
Individually both techniques have been proven useful, buthaven’t been tied into a comprehensive solution
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 16
Intelligent Mobile Agents for PMIntelligent Mobile Agents for PMAgent Tasks
Information gatheringInformation gathering Data analysis resulting in R/Y/G labelingData analysis resulting in R/Y/G labeling Determine each Determine each WBSWBS’’s s associated R/Y/Gassociated R/Y/G
cost / schedule / risk / etc.cost / schedule / risk / etc. Trend analysisTrend analysis Historical data analysisHistorical data analysis Action Item monitoringAction Item monitoring Critical path monitoringCritical path monitoring Generate alertsGenerate alerts Earned value analysisEarned value analysis Predictive analysisPredictive analysis RecommendationsRecommendations
Leveraging LockheedLeveraging LockheedMartin agent technology toMartin agent technology tomanage a SDM; user seesmanage a SDM; user sees
•• InformationInformation•• Not agentsNot agents•• Not volumes of dataNot volumes of data
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 17
Case Study:Case Study:DSM Based Rework AnalysisDSM Based Rework Analysis
DSM analysis shows one dominant design mode (eigenvalue) and relativeDSM analysis shows one dominant design mode (eigenvalue) and relativeimpact of tasks (eigenvector) on the overall scheduleimpact of tasks (eigenvector) on the overall schedule
Analysis is robust in the presence of parameter variation / uncertaintyAnalysis is robust in the presence of parameter variation / uncertainty Analysis predicts approximate percent rework for the tasks due toAnalysis predicts approximate percent rework for the tasks due to
dependencies / couplingdependencies / coupling Program manager can use this analysis to verify estimated reworkProgram manager can use this analysis to verify estimated rework Analysis can be used to evaluate the impact of task restructuring on expected reworkAnalysis can be used to evaluate the impact of task restructuring on expected rework
(which is often the cause of surprise towards the end)(which is often the cause of surprise towards the end)
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 18
Dependency
DependencyWith Delay
DSM can identify asmall subset ofdependencies thatrequire detailedmodeling via SDM
Case Study: DSM used to CreateCase Study: DSM used to CreateSystem Dynamics Model (SDM)System Dynamics Model (SDM)
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 19
Case Study: Model ResultsCase Study: Model Results(Actual (Actual vsvs. Predicted). Predicted)
Model is successfully predicting the observed data.
Level 1 Problem Reports
Level 2 Problem Reports
Planned
Predicted
Actual
Linear Plan
Non-Linear Results
Crossover Point
MCES 2004 October 26 – 28, 2004 Bipin Chadha Page 20
ConclusionsConclusions Systemic model demonstrates the behavior in theSystemic model demonstrates the behavior in the
presence of rework and feedbackpresence of rework and feedback
The results of the model are broadly applicableThe results of the model are broadly applicable All iterative processes will demonstrate this behavior to aAll iterative processes will demonstrate this behavior to a
certain extent (rework pattern)certain extent (rework pattern)
Model is successfully predicting the crossover effectModel is successfully predicting the crossover effect This is where the status goes from green to red overnightThis is where the status goes from green to red overnight
Early analysis can predict this effect and program managersEarly analysis can predict this effect and program managerscan plan accordingly to prevent surprisecan plan accordingly to prevent surprise
Model can be run anytime with latest data to detect possibilityModel can be run anytime with latest data to detect possibilityof crossovers (ideal task for an agent)of crossovers (ideal task for an agent)
Planning to deploy capability on programsPlanning to deploy capability on programs