6.1 Schedule Risk Principles, Risk Drivers, and Three-Point Estimates

6.1 Schedule Risk Principles, Risk Drivers, and Three-Point Estimates

A deterministic Critical Path Method (CPM) schedule calculates dates from one duration, one calendar, and one logic model for each activity. It is essential for coordination, but its single finish date is conditional: the date is true only if the modeled assumptions occur. A probabilistic schedule risk analysis repeatedly samples uncertain inputs and shows a range of completion outcomes.

AACE RP 57R-09 addresses integrated cost and schedule risk analysis using risk drivers and Monte Carlo simulation of a CPM model. It is guidance, not a universal standard, and it emphasizes choosing a degree of confidence appropriate to the decision. AACE RP 65R-11 is a different document: it concerns integrated expected-value methods. Do not cite RP 65R-11 as the title for Monte Carlo CPM simulation.

Two different things to model

Treating every risk as independent activity padding usually overstates some paths, understates common-cause effects, and hides the source of contingency. The model should preserve traceability from each material risk to its affected work.

Building three-point estimates

For an uncertain duration, define:

• Optimistic (O): a credible short outcome under favorable but feasible conditions;
• Most likely (M): the modal or most plausible outcome under the stated execution plan; and
• Pessimistic (P): a credible long outcome under unfavorable but feasible conditions.

The labels do not automatically mean P1, P50, or P99. If the organization intends the endpoints to represent particular percentiles, it must say so. The team should document crew size, quantities, production rates, calendars, weather treatment, access assumptions, and whether named risk events are included. Otherwise, two experts can use the same three numbers while describing different uncertainties.

Classic PERT approximation

The traditional PERT approximation is:

$T_e = \frac{O + 4M + P}{6}$

$\sigma \approx \frac{P - O}{6}$

For O = 12, M = 18, and P = 36 working days, the expected duration is 20 days and the approximate standard deviation is 4 days. These are model outputs, not guarantees. The standard-deviation expression is a PERT approximation; it is not the standard deviation of every possible beta distribution sharing the same three inputs.

Triangular distribution

A triangular distribution uses minimum, mode, and maximum directly. Its mean is:

$\mu = \frac{O + M + P}{3}$

It can be useful when a bounded range and a mode are all that can be defended. A beta-PERT form can place more weight near the mode. Neither distribution is automatically “correct” merely because data are scarce or abundant. Historical data may support an empirical distribution, while expert judgment may still support either triangular or beta-PERT after calibration.

Discrete risk drivers

Suppose an environmental permit has a 35% chance of causing delay. A traceable model can represent:

1. a 35% occurrence probability;
2. an impact distribution conditional on occurrence; and
3. the activities and logic interfaces affected.

In iterations where the risk does not occur, its added impact is zero. In iterations where it occurs, the model samples the defined impact. If the event affects design, procurement, and field mobilization together, mapping a common driver preserves that dependence better than applying three unrelated random paddings.

Correlation and model quality

Correlated inputs matter. Weather may affect several civil activities during the same period; a late vendor drawing may affect fabrication and testing together. Ignoring correlation can produce an unrealistically narrow or otherwise distorted output distribution. Conversely, adding arbitrary correlation everywhere can overstate tail risk.

Before trusting the output, verify that the deterministic CPM is usable: logic is complete, calendars are intentional, constraints are justified, resource assumptions are understood, and the status date is correct. Simulation does not repair a defective network; it repeats that defect thousands of times.

Interpreting results

The output usually includes a completion-date distribution, percentile dates, criticality or schedule-sensitivity measures, and risk-driver rankings. These answer different questions. A P80 date is the date met or beaten in about 80% of modeled iterations, given the model assumptions. A high criticality index indicates frequent critical-path participation; high correlation indicates strong association with completion variation. Neither statistic by itself proves causation or prescribes a mitigation.

The practical discipline is to state assumptions, run sensitivity checks, and explain limitations. A risk model supports a decision; it does not convert uncertain inputs into certainty.