Driving Behavioral Change for Information Management through Data-Driven Gree...
William.jarvis
1. Cost and Schedule Integration
A Practical Perspective
Will Jarvis
NASA PA&E/IPAO
Presented at the PM Challenge 2010 Conference, February 9-10, 2010,
Galvelston, TX
2. Cost-Schedule Integration
A common wisdom in cost estimating is that “one needs
a good schedule in order to do a good cost estimate”.
Good cost and schedule estimates are, in turn,
conditional upon the baseline technical definition of the
Program or Project for which the estimates are being
performed.
Cost = F (Technical, Schedule)
4. Cost-Schedule Integration
Variables cost and schedule are dependent.
Independent
Dependent
Correlated Uncorrelated
… and they are usually correlated.
5. 70% JCL Frontier
0% Correlation
GPM Core Observatory Total Cost With - RY$ vs Launch Date
Costs in RY $M, 5000 Iterations
1400.000
GPM Core Observatory Total Cost With - RY$
1200.000
1000.000
JCL=70%
800.000
SRB ICE Conditional
Prob = 60.9%
GPM Project (With
600.000 Reserves)
400.000
26-Jan-13 17-Mar-13 6-May-13 25-Jun-13 14-Aug-13 3-Oct-13 22-Nov-13 11-Jan-14 2-Mar-14 21-Apr-14
Launch Date
6. 70% JCL Frontier
80% Correlation
GPM Core Observatory Total Cost - RY$ vs Launch Date
Costs in RY $M, 5000 Iterations
1400.000
GPM Core Observatory Total Cost - RY$
1200.000
1000.000
JCL=70%
800.000
SRB ICE Conditional
Prob = 69.2%
GPM Project (With
600.000 Reserves)
400.000
26-Jan-13 17-Mar-13 6-May-13 25-Jun-13 14-Aug-13 3-Oct-13 22-Nov-13 11-Jan-14 2-Mar-14 21-Apr-14
Launch Date
7. Cost-Schedule Integration
By treating cost and schedule as joint random variables, we can reduce
overall estimating risk by leveraging the dependency between them.
8. Cost-Schedule Integration
Targets c and s are chosen to achieve a certain confidence
level. For example 70 percent where the cumulative joint
probability,
P(C c and S s) 0.7
We know,
P(C c and S s) P(C c|S s) P(S s) Why?
Resource-loaded
Schedule Model
9. Cost-Schedule Integration
P(Cost<c) P(Schedule<s)
P(C c and S s)
P(C c|S s)
P( S s )
The cost estimate is improved by making it conditional on the
schedule estimate.
10. Cost-Schedule Integration
If cost and schedule are treated as independent then,
P(C c and S s) P(C c) P(S s)
In other words, Carl and Sally work independently then combine
their results at the end.
For example,
Carl finds c so that P(C c) 0.7
Sally finds s so that P( S s) 0.7
Then,
P(C c and S s) 0.49
11. Cost-Schedule Integration
If cost and schedule are not assumed independent, then
P(C c and S s) P(C c | S s) P(S s)
In other words, Carl and Sally work together on a integrated
cost-loaded schedule model.
For example,
Together, Carl and Sally find that
P(C c | S s) P(C c)
For the same values of c and s, the joint probability is
increased (i.e., estimating confidence is increased).
12. GPM Conditional Cost
Distribution
• Two plots of P (Cost < x | Schedule < y) the symbol, | , means “given”
• The blue curve is the original Cost S-curve, i.e., P (Cost < x | Schedule < )
• The pink curve is the modified Cost S-curve, given that we know that the
launch will occur before 01 Oct 13.
P
TY$M
• Cost S-curve becomes steeper with increased certainty of the project’s
duration.
13. Case Studies
• Constellation Program (CxP) Ground Operations
Project (GOP)
• Global Precipitation Measurement (GPM)
Analytic Method
Simulation Method
Cost Loaded Schedule Method
• Radiation Belt Storm Probe (RBSP)
15. GOP Case Study
• Independent Cost Estimate (ICE) & subsequent analysis
(2007-2008)
• Initial Attempt to Integrated Cost and Schedule Risk
Analysis
• Analysis focused on the Ground Systems Development
for the IOC phase due to the lack of available detailed
schedules for future phases
• Tools
Cost Model developed in ACEIT
Schedule Model developed in GOLDPAN
Cost Schedule Interactions Implement in ACEIT
• Method
Cost Risk Analysis adjusted for impact of Schedule Uncertainty
No Inefficiency Penalty for schedule slips
16. GOP Case Study Process
• Cost Estimate
Facilities Hardware Estimate
Fixed Price Construction Contracts and GSE Acquisition/Installation
A Category of Cost Now Referred to as “Time Independent” Costs
Government Labor Estimate
FTE and WYE Labor and Related Costs
Project management, system engineering, acceptance, and activation activities
A Category of Cost Now Referred to as “Time Dependent” Costs
• Schedule Estimate
Durations for Completion of Major Facilities
Baseline Durations (B) From Deterministic Schedule used for Critical Path Analysis
CDF for Days of Deviation (D) from Baseline
Transfer Schedule CDF to ACEIT
• Cost Schedule Adjustment Factor CSAF=(B+D)/B
Factor applied to Labor or “Time Dependent” Costs
Calculated on each iteration of Simulation
Adjustment
For D>0 CSAF>1 increases costs
For D<0 CSAF<1 decreases costs
Straight Line Adjustment – No penalties for inefficiencies caused by schedule slips
17. GOP Case Study Results
Initial Operational Capability (IOC)
Impact of Schedule Risk
100%
90%
80%
Confidence Level (CDF)
70%
60%
50%
40%
30%
20%
10%
0%
4,000 4,500 5,000 5,500 6,000 6,500 7,000 7,500 8,000
TY $M
Discrete Risk Case (cdf) Point Estimate Discrete Risk No Schedule Risk Case (cdf)
18. GOP Case Study Evaluation
• Strengths
Incorporates schedule uncertainty into the cost risk analysis
Can be implemented at detailed levels of WBS
Implemented at the Major Facility Level (Pad, MLP, VAB, etc.)
• Weaknesses
Does not display Joint Cost/Schedule results
Cost S-curve impacted by schedule
No visibility into schedule
Limited Schedule Scope – Did not include the complete program
Limited WBS Implementation – Schedule Impacts Only included for Facilities
20. GPM Case Study
• Analysis focused on the Core Observatory Satellite due to the lack
of available detailed schedules for the Low Inclination Satellite
• Tools
Cost Model developed in ACEIT
Schedule Model developed in MS Project and Pertmaster
Cost Schedule Interactions Implement in EXCEL
NASA Cost-Schedule Integration Spreadsheet (MCR, Inc.)
• Method
Analytical Calculation of Bivariate Log-Normal Distribution
Cost mean and standard deviation – per GPM analysis (ICE)
Schedule mean and standard deviation – per GPM analysis (ISA)
Cost/Schedule correlation coefficient of 0.8 (based on analysis by Aerospace, Corp.)
Performed at top level for total Core Observatory Satellite
Calculated Joint Distribution and Conditional Probability Curves
P(Cost<x|Schedule<y)
24. GPM Joint Cost Schedule
Distribution
70-80%
Confidence
Band
1.0
0.9-1
0.9 66
0.8 0.8-0.9
0.7 63
0.7-0.8
0.6 60
0.5 0.6-0.7
0.4 57 0.5-0.6
0.3
54 0.4-0.5
0.2
0.1 51 0.3-0.4
0.0 0.2-0.3
66 48
1750
60
1550
550
650
750
850
950
1050
1150
1250
1350
1450
1550
1650
1750
1350
54 0.1-0.2
1150
950
48
750
0-0.1
550
The Point Estimate Cost BY2009$M
A 70% Confidence Solution Schedule Months from PDR
. . . stated that the essence of the new policy is that programs and projects are to be baselined, rebaselined, and
budgeted based on a joint cost and schedule probabilistic analysis; that programs must have a confidence level of 70%
or the level approved by the decision authority, projects must have a confidence level consistent with the program’s
confidence level, and as a minimum, projects are to be funded at a level that is equivalent to a confidence level of 50% or
as approved by the decision authority.
25. GPM Conditional Cost
Distribution
• Two plots of P (Cost < x | Schedule < y) the symbol, | , means “given”
• The blue curve is the original Cost S-curve, i.e., P (Cost < x | Schedule < )
• The pink curve is the modified Cost S-curve, given that we know that the
launch will occur before 01 Oct 13.
P
TY$M
• Cost S-curve becomes steeper with increased certainty of the project’s
duration.
26. GPM Analytic Approach Evaluation
• Strengths
Joint Cost and Schedule Results P(C<c and S<s)
Conditional probability of Cost Given Schedule P(C<c|S<s)
• Weakness
Assumption of Bivariate Log-Normal Model for cost and
schedule variables
Assumption on Cost and Schedule Correlation parameter for
model
Aerospace Study Related Cost Growth and Schedule Growth
Limited Schedule Scope – Did not include the complete program
Included Only Core Observatory
Limited WBS Implementation
Analysis performed at total Satellite Level
28. GPM Simulation Approach
• Analysis focused on the Core Observatory Satellite due
to the lack of available detailed schedules for the Low
Inclination Satellite
• Tools
Cost Model developed in ACEIT
Schedule Model developed in MS Project and Pertmaster
Cost Schedule Interactions Implemented in ACEIT and EXCEL
Risk Analysis Performed in ACEIT
Simulation Draws are Extracted into EXCEL for Analysis and Display
• Method
Simulation of unconstrained Cost and Schedule distributions
Requires Assumption for Correlation between Cost and Schedule
Performed at top level for total Core Observatory Satellite
Calculates Joint Distribution
29. GPM Simulation Results
GPM Core Observatory Total Cost - RY$ vs Launch Date
Costs in RY $M, 5000 Iterations
1400.000
GPM Core Observatory Total Cost - RY$
1200.000
1000.000
JCL=70%
800.000
SRB ICE Conditional
Prob = 69.2%
GPM Project (With
600.000 Reserves)
400.000
26-Jan-13 17-Mar-13 6-May-13 25-Jun-13 14-Aug-13 3-Oct-13 22-Nov-13 11-Jan-14 2-Mar-14 21-Apr-14
Launch Date
30. GPM Simulation Approach
Evaluation
• Strengths
Joint Cost and Schedule Results P(C<c and S<s)
Conditional probability of Cost Given Schedule P(C<c|S<s)
No Assumption Required for form of Joint Distribution
• Weaknesses
Assumption of Cost Growth and Schedule Growth Correlation
Limited Schedule Scope – Did not include the complete program
Included Only Core Observatory
Limited WBS Implementation
Analysis performed at total Satellite Level
32. GPM Resource Loaded Schedule
Approach
• Analysis focused on the Core Observatory Satellite due
to the lack of available detailed schedules for the Low
Inclination Satellite
• Tools
Cost Model developed in ACEIT
Schedule Model developed in MS Project and Pertmaster
Cost Schedule Interactions Implement in Pertmaster
• Method
Estimated Costs/Resources Loaded on Schedule
No attempt was made to segregate fixed and variable costs
Costs are dependent on task duration (i.e. cost increases as schedule
grows)
Focused on Costs-To-Complete
Calculates Joint Distribution
34. GPM Resource Loaded Schedule
Approach Evaluation
• Strengths
Joint Cost and Schedule Results P(C<c and S<s)
No Assumption Required for form of Joint Distribution
Captures Schedule Logic
• Weaknesses
Resource Loading Excludes Cost Estimating Risk and Technical
Risk impact on Costs
Did Not Segregate Fixed and Variable Costs
Costs only scaled by schedule
36. Independent Cost and
Schedule Assessment
• Independent Cost Estimate
Parametric methodology using Price, NICM and SEER-SEM
ICE performed to original schedule capturing risks identified
by the SRB
Adjusted ICE done to capture results of ISA
• Independent Schedule Assessment and Risk
Identification
Available margin was kept in the schedule
Ten risks identified from the Project Risk List
SRB assessed the potential schedule impact due to each
risk
37. Independent Schedule Risk
Assessment Results
• At the 50th % RBSP launch has a potential of slipping 6.7 months
• At the 70th % schedule slip is estimated to be 7.0 months
37
38. RBSP Cost and Schedule Integration
• ISA results applied to specific WBS items
• Determined burn-rates (generally FY10) for each affected WBS
• Time to dollars conversion: ISA Schedule extensions modeled as triangular
distributions with the burn-rate values
• Cost of schedule extension shown at the 70th percentile to be consistent with
cost risk
$25,000
$20,000
PM/SE/MA
ECT
RBSPICE
$15,000 EFW
EMFISIS
Power Distribution
Thermal Control
$10,000 Flight Software
System I&T
Mission Ops Dev
$5,000
$0
2008 2009 2010
40. RBSP Approach Evaluation
• Strengths
Easy to implement in ACEIT
Provided a reasonable result
• Weaknesses
Not a true joint probability distribution
Did not consider time independent costs
41. Conclusions
• Assuming cost and schedule are independent does not
allow for improved estimating confidence.
• Overall cost and schedule risk is reduced by observing
the interaction between cost and schedule.
Cost as a function of Schedule
Conditional Probability of Cost given Schedule
Joint Cost and Schedule Probability
• Correlation between cost and schedule can be modeled
in different ways:
Parametric model
Resource-loaded schedule model
42. Conclusions (Continued)
• Parametric JCL Model
Required Information:
Cost S-curve
Schedule S-curve
Correlation between cost and schedule
Tools
ACEIT
EXCEL
Issues
Does not require Mapping of cost to task durations
Assumption on cost/schedule correlation
Phasing of costs
How much schedule slip is included in parametric data
• Resource-loaded Schedule model
“JCL Experiment” demonstrated feasibility of calculating joint probability
Tools
Pertmaster
Issues
Schedule defined to IOC only
Costs scaled to schedule durations
Excludes Cost Estimating and Technical Risks