INTEGRATED COST / SCHEDULE RISK ANALYSIS

David T. Hulett, Ph.D.

Hulett & Associates, LLC
info@projectrisk.com
Bill Campbell III, CFA
Northrop Grumman Corporation
Bill.Campbell@NorthropGrumman.com

Introduction
Background
Project cost and schedule estimates often seem to be disconnected. When the optimistic estimate of
schedule is retained, in the face of facts to the contrary, while producing an estimate of cost, cost is
underestimated. Further, when the risk of schedule is disregarded in estimating cost risk, that cost
risk is underestimated. The current paper stems from cost and schedule risk consulting assignments
on the US Army Chemical Demilitarization project and was refined in analysis of a payload project for
the US Air Force.

Cost and Schedule are Related

The estimate at completion (EAC) for a project makes many assumptions. Key among those are the
resources that will be working on the project activities, their rate of compensation and their durations
for the activities that react to duration. Each of these assumptions contains potential risk, where risk
is defined as an uncertain event or condition that, if it occurs has a positive or negative impact on the
cost objective. (PMBOK Guide, PMI 2000) Of course, even for those activities where cost does not
respond to duration, there is uncertainty.

Integrating Risk in Cost and Schedule

Cost estimates often become disconnected from the facts of the project that are sometimes different
from the assumptions. The risk that project cost will exceed the estimate is assessed by taking into
account several types of risk, most notably the risk that project activities durations may differ from
those originally assumed (schedule risk). Other cost risks include the resources to be applied per unit
time (burn rate) and compensation per hour for those resources, and uncertainty in the cost of
activities whose costs do not depend on elapsed time.

Schedule Risk Analysis

Schedule Risk
The strategy for integration of schedule and cost risk begins with an analysis of the risk of the
schedule. The dates computed in Figure 1 assume all activity durations are known with certainty. The
scheduling software indicates a finish date for system delivery of January 21. Unfortunately we
cannot estimate durations with accuracy. Often the uncertainty in the durations is represented by a 3point estimate.

Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 1

Name
Start

Duratio
0d

Start
1/2

Finish 4th Qua 1st Qua 2nd Qua3rd Qua 4th Qua 1st Qu
1/2
1/2

Software Subsystem

265 d

1/2

9/23

Design Software

65 d

1/2

3/7

Code Software

150 d

3/8

8/4

Test Software

50 d

8/5

9/23

Hardware Platform

260 d

1/2

9/18

Design Hardware

60 d

1/2

3/2

Build Hardware

165 d

3/3

8/14

QC Hardware

35 d

8/15

9/18

120 d

9/24

1/21

Figure 1. The schedule

before risk analysis

Integration

1/2

80 d

9/24

40 d

12/13
1/21

8/14
8/15

9/18

9/24

1/21

0d

9/23

3/2
3/3

12/12

Test Integrated Syste

8/4
8/5

1/2

Integrate Hardware /

Collecting Risk Data

3/7
3/8

12/12

1/21

Deliver System

12/13

1/2

The collection of data about project risk is usually an involved process. Most organizations do not
have databases that include information suitable for quantitative risk analysis, so interview techniques
are used, and the data are usually estimated judgmentally. The data to be collected consist of 3-point
estimates of duration (optimistic, most likely and pessimistic scenarios), the probability of failure for
activities that may fail, and correlations among uncertain estimates. These scenarios are typically
discovered by in-depth interviews of those who understand the project and its risks. Gathering this
information is difficult, particularly when the interviewees (1) want to bias the results, and (2) are new
to the concepts of risk analysis data and the process of its collection.
Figure 2 shows a possible set of 3-point estimates that are assumed to represent triangular
distributions.

Figure 2. Add risk to each

activity

Duration
0d

ID
1

Name
Start

Software Subsystem

265 d

Finish=RiskOUTPUT()

Design Software

65 d

Duration=RiskTRIANG(50,65,80)

Code Software

150 d

Duration=RiskTRIANG(140,150,165)

Test Software

50 d

Duration=RiskTRIANG(45,55,70)

Hardware Platform

260 d

@RISK: Functions

Finish=RiskOUTPUT()

Design Hardware

60 d

Duration=RiskTRIANG(50,60,80)

Build Hardware

165 d

Duration=RiskTRIANG(135,165,190)

QC Hardware

35 d

Duration=RiskTRIANG(30,40,60)

120 d

Finish=RiskOUTPUT()

10

Integration

11

Integrate Hardware / Softw

80 d

Duration=RiskTRIANG(65,80,100)

12

Test Integrated System

40 d

Duration=RiskTRIANG(30,40,65)

13

Deliver System

1/2

0d

Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 2

Schedule Risk Analysis Results

The risk results provide possible dates and their probabilities for system delivery. This distribution of
completion dates is found by simulating the schedule many times. In each iteration, duration is picked
at random from the distributions shown in Figure 2 for each activity. Because we do not know which
combination of durations will actually occur, we tell the computer to conduct, say, 3,000 iterations,
each with durations selected at random according to the probability distributions for each tasks
duration. A probability distribution of possible delivery dates results and is shown in Figure 3.

Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 3

3/29/03

3/18/03

3/7/03

2/24/03

2/13/03

2/2/03

1/22/03

1/11/03

12/20/02

Delivery Dates

Baseline
Date is
very
Unlikely

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

12/31/02

Results, Distribution of

Value

Figure 3, Schedule Risk

Prob of Value <= X-axis

D is trib u tio n fo r F in is h o f In te g ra tio n

D a te

These results indicate that the date of 2/21 is very unlikely. There are other results from the schedule
risk analysis. The software shows which activities and paths are most likely to delay the project to
assist risk response planning. Figure 4 shows that Integration is always on the critical path and that
software is the next most likely to delay the project. This information is quite helpful to the project
manager.

Collecting Risk Data

The collection of data about project risk is usually an involved process. Most organizations do not
have databases that include information suitable for quantitative risk analysis, so interview techniques
are used, and the data are usually estimated judgmentally. The data to be collected consist of 3-point
estimates of duration (optimistic, most likely and pessimistic scenarios), the probability of failure for
activities that may fail, and correlations among uncertain estimates. These scenarios are typically
These results indicate that the date of 2/21 is very unlikely. There are other results from the schedule
risk analysis. The software shows which activities and paths are most likely to delay the project to
assist risk response planning. Figure 4 shows that Integration is always on the critical path and that
software is the next most likely to delay the project. This information is quite helpful to the project
manager.

Figure 4. Risk Criticality of

Activities and Paths

Activity
Design Software
Code Software
Test Software
Design Hardware
Build Hardware
QC Hardware
Integrate Hardware / Software
Test Integrated System
Deliver System

Risk Critical
Index
58%
58%
58%
42%
42%
42%
100%
100%
100%

Integration of Schedule Risk into the Cost Estimate

Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 4

Matching Cost and Schedule Concepts

The cost estimates for each activity should be driven by a realistic estimate of the duration of the
tasks with which they are connected. Similarly, the risk in the cost estimate is driven by the risk in the
estimate of duration for the same activities. To drive the cost risk for each activity by the correct
schedule risk, we need to match the cost estimates to their corresponding schedule items.
Finding schedule activities that are the same as the entries in the cost estimate is not as easy as it
should be. Often the cost estimate is related to the work breakdown structure (WBS) but the schedule
is not. We found that it was easier to take the dollars in the cost estimate and apportion them into the
schedule summary tasks than vice versa. The duration uncertainty in each summary activity can then
drive the appropriate number of dollars. Mapping was conducted based on interviews of the team
leaders who examined the cost estimates and put rounded percentages of the cost into specific
schedule activities. The results of this mapping are shown in Figure 5.
Map to Schedule
Summary Activity:

Figure 5. Map Costs to

Schedule Activities

Cost Element

A
S/W
B
H/W
C
Integ.
A/B/C
Mat.
A/B/C
P.M & Support
Total Cost per EAC

($ 000)
4,594
3,804
2,077
1,976
320
12,770

Schedule Element
A
B
C

S/W
H/W
Integ.

($ 000)
5,359
4,569
2,842

Total Cost per Sch.

12,770

Cost Risk Analysis Taking Schedule Risk Into Account

Schedule Element Input to the Cost Model
Total cost is the sum of variable and independent costs. Each of these has uncertainty. Variable
costs are the product of three items and each is uncertain.
Input
Item

Uncertain

Source of Data

Duration

Yes

Burn
Rate

Yes

Schedule Risk
Anaylsis Histogram
Interviews

Yes

Interviews

Compens
ation rate

The first problem is to find a common denominator for these three items. Hours can be used, but
integration with the schedule is easier if they are translated to days.
The second problem is to determine the uncertain duration of the schedule summary elements. Each
of these inputs is the probability distribution of the appropriate summary activity, such as those listed
in Figure 5, above calculated from the simulation of the schedule. This distribution or histogram
shows the different duration that the summary activity could take and their relative likelihood.
Calculation of the number of days the Software Subsystem and Hardware Platform take can be found
directly from the simulation of the schedule because these two summary activities start on a fixed day,
the first day of the project. (We have used 7-day weeks in the schedule because the days will be
translated to Excel that does not distinguish weekends from weekdays in computing the date
mathematics. Thus the schedule date will be wrong, unless it is a plant turnaround with 7-day
schedules but the duration in days should be accurate.)

Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 5

The histogram from the schedule simulation for the Integration summary task will not represent
faithfully the number of days of integration work because its starting date is not fixed but depends on
the completion of the longer of its two uncertain predecessor paths, software or hardware. We
simulated the Integration tasks separately, starting from an artificial fixed date, to develop the
histogram of uncertain integration days that abstracts from the uncertain start date for those tasks.

Integration in the Original Schedule

Figure 7. Simulate a
Downstream Path

Integration Summary
Integrate H/W & S/W
Test System
Deliver

120 d
80 d
40 d
0d

9/24
9/24
12/13
1/21

1/21
12/12
1/21
1/21

Finish=RiskOUTPUT()
Duration=RiskTRIANG(65,80,100)
Duration=RiskTRIANG(30,40,65)

Integration Simulated Separately to Derive a Histogram in Elapsed Days

Integration Summary
Integrate H/W & S/W
Test System
Deliver

120 d
80 d
40 d
0d

1/2
1/2
3/23
5/1

5/1
3/22
5/1
5/1

Finish=RiskOUTPUT()
Duration=RiskTRIANG(65,80,110)
Duration=RiskTRIANG(30,40,65)

The histogram is transferred to the cost model from the schedule simulation easily because we used
the same makers software for each of the simulations. Being specific, the schedule was simulated
using @RISK for (Microsoft) Project that produced a histogram as an output. We used a 30-point
distribution in this analysis for accuracy. An example of that histogram might be the following:
Duration=RiskHISTOGRM(242,383,0.001,0.004,0.009,0.017,0.0255,0.032,0.053,0.0715,0.0655,0.089
5,0.088,0.09,0.0955,0.0885,0.056,0.0545,0.0485,0.032,0.027,0.019,0.013,0.0115,0.0015,0.0045,0.00
15,0,0.001,0,0,0}).
The cost risk model was programmed in Microsoft (MS) Excel. The simulation program for the cost
risk model was @RISK for Excel, which accepts the histograms from its cousin in Project. If the
software that simluates the Excel model does not accept the outputs of the schedule risk model, for
instance if Risk+ is used to simulate MS Project and Crystal Ball simulates the Excel model, an
approximation of the schedule risk outputs must be made and input into the cost risk model,
introducing some error that may be large or small. A larger error might occur if there are probabilistic
branches in the schedule and the histogram from the schedule risk analysis has two maxima. It would
be difficult to estimate a single distribution or to make a custom distribution in Excel for such a shape.

Collecting Risk Data on Time-Dependent Costs

The data collection process starts by defining the components of cost that are needed to simulate the
cost risk analysis. These items are all uncertain unless they are in the past. In fact, if the analysis is
done in the middle of the project, there should be a history of actuals that would be certain and
added onto any uncertain estimate of costs yet to go.
The first element to be collected from the project team leaders is the burn rate or number of
resources (hours, heads) applied to the task per unit time (day). The burn rate (e.g., heads per day)
will be multiplied by the duration (e.g., days) to compute the resources (e.g., total hours) for the
schedule summary task.
The burn rate is a partly-new concept for the teams because they tend to think in time-phased terms.
We could not incorporate any time phasing of staffing in the simulation model, so the concept had to
be the equivalent based on an assumption of level staffing. It turned out that this concept was easy
enough to estimate once it was explained and the team was given encouragement.
The team leader was expected to meet with the team on all of these data collection tasks and develop
3-point estimates that could be turned into probability distributions for simulation. For burn rate we
provided data on the estimate of burn rate computed from the cost estimate. This is found by
computing the total hours for the tasks in the base estimate and dividing by the number of days

Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 6

assumed in that estimate. The result is either in hours or headcount. The team fills out the table with
their first estimate of optimistic (low), most likely and pessimistic (high) number of hours or heads per
day.
Note that the number of resources that emerges from this exercise need not be the number in the
base estimate. Sometimes the base estimate is out of date. Sometimes it was squeezed to fit some
target handed down by management or the customer and was never a true estimate of cost. Given
these factors, it is not surprising that the most likely estimate of resources per day is often not that
implied by the original estimate. Infrequently, but sometimes, the optimistic (low) estimate of
resources is above that implied by the base estimates.
The team is then interviewed on their first guess at the three-point estimates. The most common
problem with these estimates is that they are too narrow. It is well-established that people who have
no or little experience in providing data on risk will initially underestimate the true uncertainty in the
numbers. For this reason, an in-depth interview is needed to check both their understanding of the
concepts and also their representation of the optimistic, pessimistic and most likely estimate. The
following table shows the result of one of those interviews. The hours and equivalent headcount are
both provided and teams may differ in the way they want to provide the data. The translation can be
made in the simulation.
Uncertainty in Variable Hours
Nominal
Low

Figure 8. Interview
Data on
Burn Rate

Duration Dependent Labor

EAC Burn Rate (hours/day)
(Equivalent headcount)

118.5
14.8

115.0
14.4

Most Likely

High

120.0
15.0

165.0
20.6

To review, the total variable cost is found by multiplying (1) the number of days duration by (2) the
hours per day and then by (3) the compensation per hour. Each input comes from a specific source
as shown in Figure 6, above.
The compensation per hour really represents the skill mix of the people presumed to be on the job. If
skill levels 2, 3 and 4 are used in the original estimate, an average compensation can be computed
from the weighted average of those to be assigned in the baseline plan. During the teams
deliberations they often noticed that there were a predominance of levels 4 and 5 on the job. This
finding, or projection, implied that the rate per hour in the estimate was not accurate. In several
instances, the skill mix assumed in the initial baseline estimate was quite different from that elicited
from the interviews. The table below is a typical entry for the uncertain compensation rate for variable
costs.

Figure 9. Interview
Data on
Labor Rate

Uncertainty in Labor Rate

Labor Rate
Dollars/Hour (Using current data)

$122.12

$110.00

$130.00

$160.00

Collecting Risk Data on Time-Independent Costs

The term independent cost refers to the cost of the project elements that do not rely on the duration
of the activity. Cost elements in this category might include software releases that are provided by a
vendor, a component or piece of equipment provided by a subcontractor, or a piece of purchased
equipment. The basic characteristic of these could be that their cost, while uncertain, does not
depend on the date of arrival. Thus, while we might not know how much a piece of equipment may
cost at an early point in the project before we get a firm quotation, its cost will not necessarily be
higher if it arrives later rather than earlier in the project.
There were really two types of independent costs elements in the analysis. One, the simpler, would
be represented by a piece of equipment or purchased item that had no time dimension at all. The 3-

Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 7

point estimate for this element of cost has the same concept as is usually found in direct estimation of
cost risk. The interviewee prepares three scenarios representing the low, most likely and high
estimates of cost. These scenarios are discussed and sometimes challenged during the risk interview
and three estimates of cost are developed that correspond to the scenarios. These three estimates
are used to develop a probability distribution, often a triangular distribution, of the risk that the element
will cost more or less than estimated. The cost model simulation uses samples directly from this
distribution.
Three-point estimates for other so-called independent costs were really derived from assumptions
about how long activities would take. Some of these independent cost items were actually estimated
with assumptions of hours, but not necessarily calendar duration. This category included items that
did not have any clear linkage to an available schedule concept. (Risk analysis is often the art of the
possible, and sometimes even the best concepts cannot be implemented.)

Correlation of Uncertain Cost Concepts

Correlation often occurs in projects when the elements costs might be expected to move together in
real projects. Elements costs could move together if a common outside influence is driving them.
One example of correlation between elements costs can be expected when a particular task or group
of tasks is more difficult than expected in the original estimate. For instance, suppose softwarecoding activities are taking more than 265 days in the current schedule. This might occur if the
software turns out to be more difficult than originally expected, or if the estimates of software coding
durations were unwarrantedly short as in a success-oriented schedule. Perhaps the original
assumptions, both for cost and schedule, included reuse of existing software, but that turns out not to
be feasible.
If software (or hardware or integration) tasks are taking longer several things might happen: (1) the
schedule is in the high end of its estimated distribution, (2) project management puts more people on
the activity increasing the per/time unit burn rate, and (3) the people added are more skilled than the
average assumed in the estimate, which drives up the compensation or labor rate.
In this scenario, it would be possible and even expected that the three items in question would all be
above their level in the original cost estimate. That is, the time would be longer, the hours per day
more and the labor rate higher, and these items would all be expected together in the same project.
The effect could also be expected to operate in reverse. That is, if the software (hardware,
integration) tasks turn out to be easier than originally estimated, then duration would be shorter, the
head count lower and the skill level (hence the labor rate) lower together.
It would therefore be said that these three items are positively correlated. That is, if many projects
were done and some had harder software (hardware, integration) tasks than originally estimated, they
would find that all three types of cost elements were moving against them. These simultaneous
adverse events would reinforce each other when multiplied for variable costs to produce a sort of
magnified high cost of the project. The same effect would work in reverse to produce significantly low
cost results. This is the working of correlation.
In this demonstration cost risk model we installed correlations between the three pairs of elements as
follows.
Sample Correlation
Correlation Pairs

Correlation

Duration: Burn Rate

0.5

Duration: Labor Rate

0.5

Burn Rate: Labor Rate

0.5

Figure 10. Example

Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 8

Correlations

In the real project the independent costs could be correlated with these as well. Also, in the real
project the teams were queried about the correlations. This concept was unfamiliar to them and had
to be explained at some length. Some interviewees did not expect to see any correlation and others
did.

Results
Main Findings
The cost estimation model was simulated using uncertainty in the schedule duration in days
represented by the histogram from the schedule risk analysis, variable cost uncertainties represented
by burn rate and labor rate uncertainties, and independent cost uncertainties. Correlations were
inserted. The results show that it is very important to represent uncertain durations along with the
traditional cost risk analysis variables.
The model was set up so that different combinations of assumption uncertainties could be turned on
and off. This allowed us to identify sensitivities and compare the impact of cost risk variables against
the schedule-related uncertainty. For instance, in the first sensitivity, one of the scenarios, the
Schedule Risk Only scenario, looked like this:

Risk-Type Variables

Turned On

Schedule Variation

Yes

Independent Hour Variation

No

Burn Rate Variation

No

Labor Rate Variation

No

Material Variation

No

The first question was whether the traditional cost risk items such as independent cost, burn rate,
labor rate or material variations were more important in determining the overall cost risk than schedule
risk (histograms from the schedule risk analysis) alone. For this sensitivity, we specified the scenario
in Figure 11 above, than reversed the items for the cost risk analysis alone. These results are shown
in Figure 12.
Distribution of Total Program Cost vs. EAC
Schedule Risk Only, Cost Risk Only

Figure 12. Scenarios of

100.0%

80.0%

Schedule Risk Separately

60.0%

Percentile

Cost Risk and

Schedule Risk Only

Cost Risk Only
EAC

40.0%

20.0%

0.0%
11.0

12.0

13.0

14.0

15.0

16.0

17.0

18.0

19.0

$Millions

Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 9

The curves show that the cost risk scenario has the greater impact on the uncertainty in project costs,
with some iterations being somewhat below and others being quite a bit higher than the EAC, which is
shown as a triangle. This shows that even if only cost risk items were considered and schedule
assumptions were held fixed at their baseline levels, costs could be two million higher at least 5% of
the time. The curve for the schedule risk (histograms) only is much steeper than that for cost risk
elements only.
The next question is whether the interaction between cost and schedule risk adds much to the results
when using cost risk. The answer to this question will demonstrate whether the integration of cost and
schedule risk elements into one model and simulation has been worth the efforts described above.
This issue is addressed in Figure 13 below.
Distribution of Total Program Cost vs. EAC
Cost Risk Only, Schedule Risk Only, Cost & Schedule Risk

Figure 13. Simulating

100.0%

80.0%

Elements Together

60.0%

Percentile

Schedule and Cost Risk

Schedule Risk Only

Cost Risk Only
Cost & Schedule Risk

40.0%

EAC

20.0%

0.0%
11.0

13.0

15.0

17.0

19.0

$Millions

The results in Figure 13 indicate that the total cost risk of this project would be incompletely specified
if only cost risk elements or only schedule risk elements were used to represent project risk. The
cumulative distribution of project cost uncertainty is further to the right than either cost or schedule
elements alone would indicate. Of course it is very unlikely that a cost risk analyst would incorporate
only the uncertain schedule data in a cost risk, but it has been known to happen. Usually the cost risk
is represented by the uncertain cost risk elements alone. If schedule risk is not taken into account the
cost risk will be underestimated, in some cases by a substantial margin.
Figure 14 below indicates the degree to which the integrating schedule risk makes a difference.

Figure 14. Components

Of Project Cost Risk

Cost &
Cost Risk Schedule
Schedule
Only
Risk Only
Risk
Minimum
11.7
11.5
11.3
Maximum
16.2
15.0
18.3
Mean
13.6
13.1
14.1
Standard
0.7
0.5
1.0
Deviation
10%
12.7
12.5
12.9
20%
13.0
12.7
13.3
30%
13.2
12.9
13.5
40%
13.4
13.0
13.8
50%
13.5
13.1
14.1
60%
13.7
13.2
14.3
70%
13.9
13.4
14.6
80%
14.2
13.5
14.9
90%
14.6
13.8
15.4

One measure is the likelihood of overrunning the EAC, which was $12.8 million in this example. The
table shows that this value could possibly be achieved with either cost risk elements or schedule risk
elements alone, but it is not very likely with both cost and schedule risk elements taken into account.
Another measure of the contribution of integrating cost and schedule risk elements into the cost risk
estimate is the mean of the distributions. The EAC of $12.8 million increases to $13.6 million when all
cost risk elements are considered, but it increases to $14.3 million when schedule risk elements are
Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 10

added. Thus, the overrun at the mean is only $0.6 million with cost risk elements alone, but it is $1.5
million, more than twice that, when schedule risk elements are added.
The same story is told at a greater level of safety, taking the 80% level. Here cost risk elements alone
have an estimate of $14.2 million but a complete analysis with schedule risk elements shows $15.0
million.
Clearly the inclusion of schedule risk in the cost risk analysis is meaningful and necessary. Without
the schedule uncertainty, the cost risk analysis would underestimate the risk of meeting cost
objectives and not provide the project manager complete information.

Using the Results to Assist Risk Response Planning

The cost risk model was configured to allow each type of risk to be simulated separately or to be
subtracted separately from the total cost risk. The first example of this use of the model is to see
which type of cost risk variable provides the most risk. There are two ways to look at risk in this
context: (1) the greatest increase in cost risk from the EAC representing overruns, and (2) the
greatest spread of risk from lowest to highest representing uncertainty or imprecision in any particular
estimate of cost.
Distribution of Cost Risk Elements and Total Cost Risk vs. EAC
100%
90%
80%
70%
Percentile

Figure 15. Burn Rate

Contributes the Most Cost Risk
Difference from the EAC

EAC

60%

Fixed Hour
Burn Rate

50%

Labor Rate

40%

Material

30%

All Cost Risk

20%
10%
0%
$11.0

$12.0

$13.0

$14.0

$15.0

$16.0

$17.0

$18.0

$19.0

$Millions

Figure 15 above shows that the burn rate of all cost risk elements contributes the most to increases of
the risk over the EAC. Control of the rate of resources per unit time may help to manage the risk of
cost overruns the most. Of course, controlling this factor may be difficult and the ratio benefit / cost of
burn rate control might be very low.
The cost risk element that contributes the greatest total uncertainty from positive (under runs) to
negative (over runs) is the labor rate in compensation per hour. Uncertainty about the skill mix of
people on the job can cause the cost to under run or over run the estimate as shown in Figure 16.

Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 11

Distribution of Cost Risk Elements and Total Cost Risk vs. EAC
100%

Figure 16. Labor Rate

Contributes the Most
Total Uncertainty
in the Cost Risk

90%
80%

Percentile

70%

EAC

60%

Fixed Hour
Burn Rate

50%

Labor Rate

40%

Material

30%

All Cost Risk

20%
10%
0%
$11.0

$12.0

$13.0

$14.0

$15.0

$16.0

$17.0

$18.0

$19.0

$Millions

Another way to examine the benefit of risk mitigation is to subtract each cost risk type from the total
risk representing all cost risk types and total schedule risk. In this way the project manager may see
what risk responses might reduce total project cost risk and prioritize risk mitigation actions. The
information in Figure 17 shows that the greatest improvement in total cost risk might come from
improving the burn rate. It shows what would happen to total cost risk if the project manager might
hold the burn rate to baseline levels. Whether this is possible or not is another story. Still, controlling
the number of resources applied per unit time would seem to offer significant cost risk reduction.
Distribution of Total Program Cost vs. EAC
Risk Mitigation -- Control the Burn Rate Uncertainty to Baseline

Figure 17. Controlling the

Burn Rate Could Contribute
the Most to
Reducing Cost Risk

100%
90%
80%

Percentile

70%
60%

EAC

50%

Cost & Schedule

40%

No Burn Rate

30%
20%
10%
0%
$11.0

$12.0

$13.0

$14.0

$15.0

$16.0

$17.0

$18.0

$19.0

$Millions

Observations
Methodology
There are several issues with the methodology of integrating cost and schedule risk analysis.
Cost and schedule structures can become disjoint, making linkage of the two difficult. Serious
effort was required to link the structures so the costs could be allocated to schedule concepts. The
recommendation is to structure both at a common starting point, probably the WBS.
Modelling issues were serious as well. Decomposing the baseline estimate into its components
and discovering what must have been the baseline assumptions was tricky. Keeping the estimates
clean was not possible, and many approximations were needed. Making the model work when many

Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 12

different cost and schedule elements are present, well beyond the simple example described here,
implies a very complicated spreadsheet model. Including the facility to simulate one or another type
of variable alone was complex modelling as well.

Data Collection Issues

Collecting high-quality data that are as accurate as possible is the most important part of the analysis.
There are new concepts for the interviewees, such as the burn rate, average compensation and
correlation between elements. We simplified the burn rate to a level-loading concept, which further
complicated it for the interviewees. Still, the teams were able to understand and respond.
The explicit use of burn rate and labor rate highlights the basic forces of the cost risk elements of
variable costs. It was found that the participants could look at both elements separately, which we
believe improved the representation of variable cost risk.
Data collection is most difficult when the participants have never been part of a risk analysis, as is
often the case. Expert facilitation is often necessary. The good news is that this is not brain surgery,
and participants can learn while doing.
Individual and organizational maturity comes from treating data collection seriously. It includes
preparing the teams and giving them a sense of importance for the effort. It includes providing
enough time for the exercise and making it part of the regular work, not an extra. It also includes
recognizing that the team leaders and members may not provide perfect data the first time they are
interviewed. They may answer one way the first time and a different way in a later interview. Usually
the second time more risk will be reported. The mature organization will understand that people
become more comfortable discussing risk as they learn how to do it.

Risk Analysis Tools

Risk analysis tools used here are @RISK for Microsoft Project and @RISK for Excel. These work
well with scheduling and spreadsheet packages that are the most commonly used today, but they
require cumbersome translation of results (histograms) from schedule to cost. A more integrated
approach is to load all resources and their costs into a scheduling package that will compute the cost
risk analysis in the same simulation as the schedule risk. The main tool for this analysis is Monte
Carlo from Primavera that takes the input data from Primaveras P3 scheduling program. The results
can be shown as a scatter of result pairs such as those below in Figure 18.

Figure 18. Cost and Schedule

Risk in One Simulation
with Monte Carlo (Primavera)

PERTMASTER can also compute cost risks from schedule uncertainties in a single simulation. One
approach that is not recommended is to try to represent the schedule in a spreadsheet to get the
schedule and cost results simulated in a spreadsheet. Scheduling packages are much preferred to
spreadsheets in representing project schedules.

Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 13

CONCLUSIONS
This paper has demonstrated that a good cost risk analysis needs to incorporate schedule risk as well
as the traditional cost risk elements. Integrating the results from the schedule risk into the cost risk
model provides a more complete picture of cost risk than if it were excluded. The picture of cost risk
that emerges will depend significantly on the degree of schedule risk that is found in the project.
Often cost risk analysis looks at individual cost elements and tries to figure out what drives their
uncertainty. Sometimes participants in a traditional cost risk analysis reveal uncertainty in activity
durations, but those concerns are usually ad hoc and incomplete. The disciplined approach of this
paper makes the consideration of schedule risk explicit and unavoidable.
Often a troubled project is having as much difficulty keeping on schedule as it is on budget. These
two problems reinforce each other, magnifying the problems. This magnification is represented in the
present model by two things: (1) multiplication of time, and therefore time uncertainty by burn rate and
therefore burn rate uncertainty, and (2) correlation between burn rate and duration.
The cost estimates often lose track of schedule realities, causing the estimates of cost to be
unrealistically low in many cases. Estimators feel that they did not have a full kit of information when
the facts of the schedule are made clear to them, making them look bad at estimating and causing
them to lose credibility with the project executives. Insisting on a close communication between the
scheduling and the cost estimation functions will help to reduce this problem. Insisting on looking at
the schedule risk when evaluating the cost risk will improve the accuracy of the estimates of project
cost and of cost risk.

Presented at the Fifth European Project Management Conference, PMI Europe 2002, Cannes France, 19-20 June 2002
Organised by the PMI France Sud
Page 14