Cost Risk Impacts of Schedule Interdependencies Peter Frederic June 2004 SCEA Conference

1
Cost Risk Impacts of Schedule InterdependenciesP
eter FredericJune 2004SCEA Conference
2
Outline

• Introduction
• Caveats
• Implementation Algorithm
• Real-world Example
• Conclusions

3
Introduction

• Correlations between cost element errors cause
  uncertainty in overall estimate to increase
• In development programs, schedule
  interdependencies are a significant cause of
  correlation
• Some advocate correlation factors capture this
  effect
• No WBS item can be completely de-staffed until
  integrated testing completed
• Schedule slip/cost overrun in any one WBS item
  will cause overrun in all WBS items
• Unfortunately, no degree of correlation can
  capture the full effect

4
Point Estimate with Schedule Durations and End
Date Distributions

• Sum of most likely estimates for each WBS item
• No specific statistical significance

5
Monte Carlo Iteration, No Correlation

• Monte Carlo used to establish statistical
  significance
• Single iteration shown
• Software very late, sensor slightly late,
  processor early
• No attempt to model interdependencies

6
Monte Carlo Iteration, 100 Correlation

• Same as previous page, except 100 Correlation
• All item results pick from the same place in
  their individual distributions
• Cost significantly higher, but
• Still no reflection of the delay before IAT
  starts

7
Monte Carlo Iteration, Schedule Links Enforced

• Same as previous page, except schedule links
  enforced directly instead of through correlation
• Sensor and Processor stretched until Software
  finish (IAT start)
• IAT stretched from left to reflect idle staff
  and equipment
• Total cost 34 higher than 100 correlation

8
Caveats

• WBS item cost growth-to-schedule growth
  relationship not always 1-to-1
• Some parts of cost growth not linked to schedule
  materials, purchased parts, or FP subcontracts
• If 50 of cost is non-labor, then 50 cost growth
  100 schedule growth
• Staffing may be increased to prevent schedule
  slip (though often too late)
• Schedule growth-to-cost growth not always 1-to-1
• If 50 of cost is non-labor, then 100 schedule
  growth 50 cost growth
• Staffing may be reduced to same money in case of
  external schedule delay
• Schedule and cost estimate not always in concert
• Methodology described here depends on schedule
  and cost estimate being related through a
  realistic staffing plan

9
Implementation Algorithm

• Using ACE RIK, within each Monte Carlo
  iteration
• Infer schedule slips based on cost growth for
  each WBS item
• Recalculate schedule milestones based on slipped
  durations and logical precedence relationships
• Recalculate costs based on slipped schedule
  milestones

Recalculate schedule milestones
Calculate individual duration growth factor
Adjust individual element based on sched slips
10
Inferring Schedule Slips

• Translate cost growths for individual cost
  elements into equivalent schedule duration growth
• Duration growth ? effort growth ? cost growth

Duration Growth Factor 1.0 (Cost With
Risk/Cost Without Risk - 1.0) Schedule-Related
Cost Factor where Cost With Risk the cost
selected from the risk distribution for the cost
element in a particular Monte Carlo
iteration Cost Without Risk the point estimate
for the cost element Schedule-Related Cost
Factor a factor indicating the portion of the
cost element that is actually related to duration

• Assume cost growth for an individual cost element
  includes CER Risk, input variable risk, and
  technical risk, but not system-level schedule
  risk impacts (systems with known schedule
  anomalies eliminated from CER databases)

11
Recalculating Schedule Milestones

• Determine impact to overall program schedule
• Requires dynamic model of program schedule built
  into cost model
• Not excruciatingly detailed -- capture the slips
  or (advances) in key program-level milestones
• Typical development program
• Preliminary Design
• System PDR
• Detailed Design
• System CDR
• Fabrication, Assembly, and Unit Test
• System Integration, Assembly, and Test (IAT)
• Cost impact of slips apparent at System PDR,
  System CDR, and System IAT

12
Recalculating Schedule Milestones (Cont.)

• Simple finish-to-start
• Milestone end dates calculated using existing MAX
  function in ACEIT

Activity Finish Date MAX(P1 Finish, P2 Finish,
P3 Finish, Pn Finish) Duration Duration
Growth Factor where Px Finish the finish date
of each predecessor activity Duration the
duration of the schedule activity whose finish
date we are calculating Duration Growth Factor
the duration growth factor for this activity as
calculated in the previous step (Inferring
Schedule Slips)
13
Recalculating Costs Based on Schedule Slips

• Calculate cost impacts of major milestone slips
  on individual cost elements
• Each element must have time-phased expenditure
  profile based point estimate initial planned
  schedule
• As major milestones slip, expenditure profile
  stretches and cost increases
• Stretch User Defined Function (UDF) developed
  in ACEIT to model expenditure stretching effect

14
Stretch Function
15
Real-world Example

• Applied schedule risk methodology to an actual
  cost estimate in progress at Tecolote
• Test case estimate is for development, including
  first flight, of an advanced space vehicle
• All cost numbers and technical parameters have
  been adjusted to protect identity of system
• Key WBS items and schedule activities have been
  renamed

16
Example Estimating WBS
17
Schedule for Example Program
18
Risk Distributions for Example
19
Monte Carlo Results for Different Methods of
Modeling Schedule Interactions
20
Conclusions

• Modeling schedule interactions directly more
  completely captures cost risk impacts of these
  interactions than statistical correlation
• Correlation still a necessary capability in risk
  assessment tools e.g. hardware commonality and
  engineering relationships between technical
  variables
• Direct approach to modeling schedule interactions
  can be implemented without adding an unreasonable
  amount of complication to an ACEIT cost model

21
Back-up
22
Historical Cost Growth Data

• Intended to analyze historical cost growth data
  to uncover statistical evidence to support the
  relationship between schedule growth and cost
  growth
• However, Schedule and Cost Growth by Coleman,
  Summerville, and Dameron 1 found no statistic
  evidence of any relationship between cost growth
  and schedule growth!
• Why?
• Much of schedule growth in database may have been
  caused by budget constraints
• E.g. Programs schedule based on spending 1M in
  one year, but only 500K available in that year
  and 500K in the next year Program will take two
  years to complete. However, overall cost will
  not necessarily grow
• baseline schedule and baseline cost estimate for
  a program may not be strongly tied
• Schedules and cost estimates developed by
  separate teams
• Initial schedules determined by working backwards
  from perceived need dates
• Initial cost estimates developed using parametric
  CERs that are not often sensitive to schedule
  milestones