The following diagram represents a process sequence or workflow for conducting effective cost-schedule risk analyses using Oracle’s Primavera Risk Analysis:

Having already discussed the steps necessary to prepare for and perform the schedule risk analysis components of the above in 3.2.1, in the sections that follow, we will be addressing only those components that relate to the preparation of the cost side of the IRA process.

Please ensure that you have read and understood the Schedule Risk Analysis Process before continuing, as the cost process cannot be taken in isolation from the schedule process. The two are inexorably linked.



Having already prepared the project schedule, you’re now ready to review the estimate and risk register in preparation for conducting risk workshops. When conducting an IRA, there are a few things that need to be addressed in terms of estimate preparation:

• First and foremost, it is important to ensure that the schedule and the estimate basis are aligned. The foundation of IRA relies on this condition not being violated. If the estimate and schedule basis are not aligned, when costs are mapped to tasks, the cost ‘burn rate’ per unit time will not be correct, and costs associated with schedule change will be incorrectly calculated. Consider some of the following:

- Have the staffing curves that have been used to generate the estimate been prepared based on the current schedule key milestone dates?

- Do both the estimate and the schedule contain inclement weather allowances? If so, are they both equally treated as above or below the line?

- Do both the estimate and the schedule cover exactly the same scope using the same execution philosophy?

• Next consider whether the level of detail and structure of the estimate is sufficient for the purpose of conducting an IRA. The estimate should be sufficiently detailed so as to be able to accurately map the costs to the relevant portions of the schedule, but not so detailed so as to be overly cumbersome to process within a reasonable timeframe. Ideally, the schedule and the estimate should both use a common Work Breakdown Structure (WBS) so as to ensure alignment.

• Third, establish the breakdown of time independent and time dependent costs. Examples of time dependent costs are labour or hired equipment. Time independent costs are typically things like purchased equipment and materials. However, as discussed in 3.1.6, the commercial conditions in which the project is operating can influence the split of time independent and time dependent costs and must be considered before making these allocations.

• Fourth, ensure that the estimate clearly separates incurred or ‘actual’ costs from remaining costs. In any risk analysis process, it is important to ensure that only remaining costs are ever ranged. Costs that have already been incurred are not subject to uncertainty and must not be ranged during analysis.

• Finally, where possible, existing contingency in the deterministic estimate should be separated out as ‘below the line’ contingency and excluded as an input to the analysis. It is important that all contingency be declared and excluded as an input to any risk analysis. Calculated contingency requirement is an output of the model, not an input to it, and failure to declare contingency will result in overly pessimistic projections.

Risk Register

As described in the SRA Having already assessed those risks having the potential to impact on schedule objectives, the project risk registers should now be re-screened for any risks with the potential to impact on cost. As noted in 1.4 Risk Attributes, it is of paramount importance that the risks are clearly expressed to identify cause, event, & effect.

When considering which costs to include, be mindful of the following two considerations:

• Included cost risk events must not overlap with the cost ranging identified against the estimate (ie. do not ‘double-dip’). Ideally, cost ranges should always omit consideration of things with a probability of existence, as these are better handled as risk events. It is important to be clear before entering a workshop as to where allowances for such uncertainties are to be included; in the ranging or the risk events.

• Ensure that cost impacts of included risk events are not for prolongation costs. Prolongation costs are calculated by the IRA model depending on when and where the risk occurs, and should be omitted as inputs to the model.


For information on available methods of gathering cost uncertainty data, please refer to 1.2.3 Risk Identification.


Use of Hammocks

The first step in mapping costs to a schedule model for IRA is to create cost hammocks spanning sets of continuous related activities against which costs will be allocated. Hammocks should be used instead of mapping costs directly to tasks because of the way that Primavera Risk Analysis software calculates costs during probabilistic calendar downtime (such as when using a probabilistic weather calendar). If costs were to be allocated directly to tasks affected by probabilistic calendar downtime, when downtime occurred, PRA would calculate the time dependent costs for the downtime periods as $0. However, in reality, if for example a construction crew were delayed by inclement weather, they would still need to be paid just for turning up, even though they didn’t actually achieve anything that day.

By creating hammocks that span sets of continuous related activities and not applying any probabilistic downtime to the hammock tasks, when the child tasks are delayed the costs allocated to the hammocks are allowed to correctly increase relative to the amount of downtime incurred.

However, care should be taken in creating cost hammocks in order to ensure that the cost model behaves correctly. Specifically, the following tips should always be followed when creating cost hammocks for an IRA model:

• Hammocks should only span continuous or near contiguous task ranges. Because costs will be spread evenly across the hammock, it doesn’t make sense for the hammock to be distributed across periods in which no activity takes place.

• Hammocks can only ever be as detailed as the level of detail in the estimate. There is no point creating tasks detailed to the N’th degree when the estimate is highly summarized as this will create problems when trying to figure out how to apportion the costs to the hammock tasks.

• Hammocks should ideally only span tasks with a consistent work loading. If there is a distinct change in the work loading levels in one particular area, a separate hammock should be identified for this period with a corresponding change in the unit rate per period cost.

• Hammocks should easily roll up to the required cost reporting structure.

• Unless the analysis aims to produce a probabilistic cash flow as an output, less attention needs to be given to the mapping of time independent costs. As long as the costs roll up to the required reporting structure, the linkages of the hammock to which they’re mapped will not affect their calculated cost.

Use of Budget vs. Remaining Cost Fields

Only uncommitted funds exclusive of contingency should be entered as ‘remaining’ values when entering costs against tasks in PRA. This is to ensure that 1) costs that are already spent are not subject to uncertainty within the model, and 2) contingency is kept as an output of the model, not an input to it, so as to avoid unrealistic degrees of pessimism.

However, it is possible and indeed preferable to still reference the existing levels of project contingency within the model by entering the contingent value within the ‘budget’ field of PRA. In this way, the contingency will not affect calculated remaining costs, but will be reference-able via the reporting histograms as a ‘baseline’ cost such that the probability of achieving this value can be readily seen.

Applying Costs & Cost Uncertainties

Applying costs to the model is a relatively straight-forward, but sometimes time consuming process:

• Having already created the cost hammocks against which costs will be entered, the first job is to ensure that the necessary resources have been allocated to the hammocks as resource assignments against which the costs can be mapped. Most tasks will require at least two resource assignments, a time independent “spread” resource assignment, and a time dependent “normal” resource assignment.

• Next, figure out exactly which estimate line items will be mapped to which cost hammocks. Mappings can be many estimate line items to one cost hammock, but try to avoid mapping one cost line item to multiple hammocks, unless you have the information required to apportion the estimate line item amount between hammocks.

• Once you have figured out all your mappings, you can now sum the budget, remaining, optimistic, most likely, and pessimistic values to be assigned to each task resource assignment.

• Outstanding expenditure excluding contingency can be entered in the ‘remaining’ field, while contingent funds can be entered in the ‘budget’ field.

• For time independent “spread” resources, the optimistic, most likely, and pessimistic costs can be entered directly in the corresponding fields in PRA. However, prior to entering the time dependent “normal” resource distribution values, all three points in the distribution must first be divided by the deterministic remaining duration of the hammock in order to produce a min/likely/max unit rate per time spread.

Distribution Types

Just as with duration uncertainties, resource assignments can also be assigned different distribution types depending on how the input data has been gathered or described. Please refer to 2.2.4 Assigning Schedule Uncertainty Ranges for more information on the use of the available distribution types in Primavera Risk Analysis.


As noted in 2.2.9 Duration Correlation, correlation is an important element of any risk analysis model. It is the means by which we define the degree of relatedness between different uncertainty elements in order to combat the starting assumption of Monte Carlo simulation; that all uncertainties are independent of one another. Correlation is just as important a consideration for cost as it is for schedule in an IRA. It is used to combat the observed effects of the Central Limit Theorem, producing a credible spread on the range of cost outcomes reported from the model.

Correlations can be positive or negative, strong or weak; but without them, the model will produce increasingly narrow or ‘peaky’ results as more and more uncorrelated uncertainties are added.

In IRA, time independent costs should be correlated according to similarities including linked quantity uncertainties & shared raw material costs. For example, if a project had two separate outstanding orders to place for carbon steel piping, the two costs would be correlated because of the shared reliance on the price of steel in determining the cost of the order. Similarly the cost of another order for pipe fittings might be positively correlated with the price of the line pipe orders because of the relationship between the quantity of the line pipe and the quantity of fittings required.

However, time dependent costs in an IRA model have the advantage of already having their correlation in-part defined by the schedule correlation model. If tasks duration uncertainties are correlated, the incurred quantity component of time dependent costs associated with these tasks also become correlated by association. However, it may still be necessary to correlate the rate component of time dependent costs to account for commonality in unit rate per time between resource assignments. This may also be achieved by applying Resource Uncertainty rather than resource assignment uncertainty.

Unfortunately, in the absence of suitable data, correlation is often a matter of judgment. There are few rights or wrongs when it comes to determining what should be correlated (and by how much). This makes applying correlation a fairly subjective process. Thankfully, there are techniques such as the application of Cost Risk Factors (described in the next section) which take away from this subjectivity by effectively eliminating the need to specify correlation between model uncertainties.


Ultimately, most cost uncertainties captured at the task resource assignment level can be attributed to some underlying source or ‘factor’ that may be common across multiple tasks. For example, when discussing Cost Correlation (above), we gave the example of the price of goods being commonly influenced by the underlying price of the raw materials from which they were constructed. This material price is an example of a cost risk factor. Cost risk factors can be applied to tasks in place of traditional cost range uncertainties in order to commonly influence a multitude of elements within the model. Through having a common influence on multiple uncertainties, risk factors also inherently define the degree of correlation between each element, negating the need to make subjective judgments regarding how to apply correlation (provided the use of risk factors is comprehensive and doesn’t leave some elements requiring correlation not covered by a risk factor).

However, the risk factors module that ports natively with Primavera Risk Analysis is unfortunately not without its problems. Specifically, there are two major problems with the native risk factors module in PRA that make it impractical to use for most users:

• First, cost risk factors can only be assigned at the task level, not the resource assignment level. This means that should a task have, for example, a labour resource and a material resource, you are not able to specify that the risk factor should impact on one resource but not the other. This presents a problem for a task containing both supply and install type costs as application of a material price risk factor would also unavoidably affect labour costs, which is incorrect.

RIMPL’s Integrated Risk Factors tool (IRF) works within Primavera Risk Analysis to allow for allocation of risk factors at the resource assignment level as well as introducing risk factor correlation.

• Second, there is a problem with the way cost risk factors in PRA are calculated when combined with cost three point distribution ranges. If no cost distribution is present, the application correctly multiplies the remaining cost of the task by the sampled value of the risk factor. However, if the resource assignments are also given ranges, PRA ignores these, instead using only the deterministic value and the risk factor in determining the calculated cost of the item.

When combined with the fact that the native risk factors module cannot be used in conjunction with the native weather module, this makes the use of this functionality somewhat problematic and unattractive.


Unlike Mapping Schedule Risks (2.2.6), the mapping of cost risks is relatively straight forward by comparison. As defined in Risk Attributes (1.4), all risks have the same common properties of probability and optimistic, most likely, and pessimistic cost impact. For cost risks defined at the project level, if multiple risk task mappings exist, the cost of the risk simply needs be apportioned among the risk tasks such that its overall impact is not over-stated. There is no complex consideration of task criticality or logic, nor series vs. parallel risk task assignments; users must simply ensure that the costs associated with the risk are expressed against the correct reporting roll-up of the cost breakdown structure within the model. However, should a risk have both schedule and cost impact, all the same guidelines outlined in Mapping Schedule Risks (2.2.6) must be followed.


Resource Cost Uncertainty

In addition to the resource assignment uncertainty discussed above, Primavera Risk Analysis is also capable of modelling uncertainty at the resource level. Uncertainty assigned at the resource level affects all resource assignments associated with that resource equally. The cost at the resource assignment level is calculated as the product of the sampled values of both the resource and the resource assignment unit values. A good example of the use of resource cost uncertainty might be where the cost of a recurrently used resource is known +/- 10%. Rather than entering +/- 10% against each of the resource assignments in which the resource is used, the uncertainty can be created at the resource level, which then affects each of the resource assignments equally in turn. This has the secondary benefit of inherently defining the correlation between each of the resource assignment rate uncertainties also.

Escalation Curves

PRA allows for the creation of escalation curves to model predicted changes in the value of resources over time. This feature could be used to model increases in Escalation using periods as small as one day, or as large as the user cares to define within the bounds of the project duration.

However, there are some limitations of the escalation curves functionality in PRA that users should be aware of:

• First, it is important to note that escalation curves are deterministic only, and not subject to uncertainty.

• Second, it should be noted that the effects of escalation curves are expressed only through some reporting features of PRA, but not others. Specifically, escalation will affect the outputs of the Probabilistic Cash Flow, but not the Cost Histogram graphs. This can easily result in confusion if users are not aware of this limitation.

Discounted Cash Flow (DCF)

Discounted Cash Flow is a method of valuing a project incorporating an allowance for the time value of money. Because of inflation and other factors such as opportunity cost, the theory behind DCF is that the present value of money should be discounted the further away from the present one forecasts into the future. For example, in relative terms, a million dollars today might be worth the equivalent of 1.1 million dollars in 3-4 years’ time. As with escalation, DCF in PRA acts only on the Probabilistic Cash Flow outputs of the model, and not on the Cost Histogram outputs.

Net Present Value (NPV), & Internal Rate of Return (IRR)

PRA also comes equipped with two reporting features for evaluating the overall profitability of projects: Net Present Value (NPV) and Internal Rate of Return (IRR). Both NPV & IRR assess the balance between project profit and loss by assessing revenue versus expenditure, including economic inputs such as escalation and DCF. In PRA, one is able to assess the probabilistic range of returns on investment by incorporating uncertainty on capital and operational expenditure, as well as revenue uncertainty.

The following Probabilistic NPV chart shows a bimodal distribution representing a 10% chance of a 20% structural reduction in revenue after 5 years operation of the asset:


Monte Carlo Method simulation is a mathematical technique that uses repeated random sampling within specified distributions to calculate the probability of defined outcomes. The principal of the method is that by simulating a process many times using ranged parameters before doing something in actuality, a mathematically based prediction of how the real process may eventuate can be calculated. The method was invented in the Second World War to simulate nuclear events during the Manhattan Project to develop the atomic bomb and has been adapted to an increasingly widespread range of applications since.

Please refer to Schedule Monte Carlo Simulation (2.2.10) for a detailed explanation of the basics of how Monte Carlo simulation works.

In addition to the processes outlined in the link above, Integrated Cost & Schedule Risk Analysis also requires the distribution sampling of resources & resource assignments. Time dependent costs are sampled not only on their distribution values, but also in response to the duration of the tasks to which they’re assigned, and in accordance with any downtime injected through the calendars assigned to those tasks and the resources themselves.

The net result is a data set derived from many hundreds or thousands of simulations of the project, from which inferences can be made as to the likelihood of project cost and schedule outcomes.


Cost Histograms

In a cost histogram, data from analyses are grouped into collectors called ‘bins’ across the dimension of cost. For example, if you had a minimum cost of $100 and a maximum cost of $200, you might create 5 bins, each with a collection range of $20. The frequency with which the results from the analysis fall into these collectors is represented as height on the vertical axis (shown on the left axis in the example above as ‘Hits’ [iteration results]). By structuring the data in this way, a visual representation of the clustering of results along the measured dimension (costs in this case) is displayed.

In the example above, the cost of the entire project has been used as the bin parameter along the horizontal axis. The dimension is always the same as the metric to be reported.

In simple terms, the above cost histogram shows the results for all iterations (‘Hits’) of a Monte Carlo Cost Risk Analysis for the Cost of the Entire Plan. The results are plotted from the minimum calculated cost of the project, through to the maximum. The height of each bar represents the number of hits that fell within the cost range represented by the width of the bar. The highest bar records that 363 hits occurred between $1,875 million and $1,900 million.

Bin sizes are a flexible variable. It is important that the bins are sized large enough to allow for adequate visual representation of trends, but not so large that they hide important information about the model.

Cumulative Curve Data

The cumulative curve adds up the number of hits in each bar progressively so that it represents the number of iterations up to a particular cost threshold. In effect, an intercept from the curve to the horizontal axis represents the percentage of iterations up to that cost threshold or the probability of the Entire Plan finishing on or under that cost.

Cost Risk Analysis results usually refer to ‘P-values’. These are the percentile values which tell us our confidence in achieving our objectives within given thresholds of project value. For example, a P90 cost value of $200 million indicates 90% confidence that the project will be completed on or within a total value of $200 million.

The minimum and maximum costs are usually of less interest than the P10, P50 and P90 (or P20, Pmean and P80) values. The intercepts used by organisations vary according to their risk policies, “risk tolerance” or “risk appetite”.

For further information please refer to the discussions of Skewness and Kurtosis in the section on Interpretation of Schedule Analysis Results (2.2.11).

Probabilistic Cash Flow

The cash flow shown here compares four cumulative representations of capital expenditure (capex) over the project period:

• The deterministic or Early cash flow (grey curve and monthly bars)

• The P10 (optimistic) probabilistic capex curve (red)

• The Pmean probabilistic capex curve (dotted black)

• The P90 (pessimistic) probabilistic capex curve

Note that the probabilistic curves continue on beyond the end of the deterministic bars and curve.

If the revenue and operational expenditure is added to the model, together with their uncertainty ranges and applicable risk events and the model is extended for the economic life of the asset created by the project, probabilistic profitability for the investment opportunity may be evaluated. The following probabilistic cash flow shows such a model for an FLNG project.

The FLNG project and asset model shown runs for 25 years after startup. It assumes that every 5 years a 6 month shutdown of production occurs for major maintenance. The cash flow assumes a notional balloon payment to the project for the FLNG vessel at the end of the 25 years operation when the FLNG vessel is redeployed elsewhere. The DCF and project finance borrowing rates are independently variable. This functionality involves the use of supplementary software and techniques not available in PRA.

From the probabilistic cash flow data available with the graph can be derived probabilistic NPVs and IRRs to evaluate the balance of probabilities between making a good return, breaking even and an unacceptable loss. Sensitivity analyses can be performed, changing whatever combinations of capex, opex and revenue are desired including assumptions of multiple risk events occurring simultaneously. For example, a structural drop in the selling price of LNG can be applied, from any point of the operational life onwards.


Cost Sensitivity Tornados

Just as sensitivities can be measured for schedule drivers (2.2.12), so can correlation be used to track the change in individual costs versus the change in overall project cost or selected portions of it.

This is a useful indicator of the main drivers of project cost, including in the example shown several risk-task costs (red bars) and a risk treatment added to the model to reduce probabilistic risk. However the cost sensitivities suffer from a couple of significant deficiencies:

• They only report costs at the task level and cannot deal with costs at the resource level

• They are unable to report accurately on driver ranking when correlation has been applied to costs and/or durations. This has already been discussed under Correlation and Causation in the Schedule Risk Analysis section.

Quantitative Exclusion Analysis

The method that is most reliable measure of the drivers of cost in the model is to remove the source of cost uncertainty, whether an individual cost ranging of a resource in a task or a grouping of related cost uncertainties or a whole class of uncertainty (eg, all procurement cost uncertainty or all cost impact risk events), then run a full simulation and report at selected P values the difference in probabilistic cost between the IRA model with and without the excluded source(s) of cost uncertainty. We call this Quantitative Exclusion Analysis (QEA).

The breakdown of differences can be depicted where mutually exclusive classes of uncertainty are being systematically removed and quantified, by pie charts, such as shown below:

Conventional tornado diagrams showing the probabilistic cost for each bar can also be produced, as shown below:

RIMPL’s IntegratedRangeDrivers (IRD) tool manages and automates Quantitative Exclusion Analysis (QEA) by Primavera Risk Analysis and subsequent reporting, to show the ranked effects of the various contributors to schedule and cost outcomes
Such presentations can show the measured value of each contributor or class of uncertainty to probabilistic cost (at the selected P value – usually at a high value such as P80 or P90).

Unlike sensitivities, this information is as reliable as the model itself is in representing the project.

comments powered by Disqus