A schedule gives you one finish date. A quantitative schedule risk analysis, or QSRA, tells you how likely that date is, which later date you can commit to with confidence, and which activities and risks decide the difference. This guide explains what you need, the steps, how to read the results, and when Excel is enough.
A QSRA runs your schedule many times with uncertain durations and risk events, and reads the finish date from every run. The result is a range of finish dates with a probability for each one, not a single date. It's decision support: it shows which dates are realistic given what the team knows today. It doesn't predict the future, and it doesn't make a date certain.
What a QSRA is, and why it matters
A deterministic schedule assumes every activity takes its most likely duration and no risk occurs. That future almost never happens. Durations run long more often than short, and risks that hit the critical path push the end date out. Where parallel paths meet, the delays pile up (see merge bias).
A QSRA makes this visible with a Monte Carlo simulation of the schedule network. The US Government Accountability Office lists it as one of its ten scheduling best practices in the Schedule Assessment Guide. GAO's point is simple: without a schedule risk analysis, you can't say how likely the planned completion date is. It recommends running one before the baseline is set and again as the schedule is updated.
What you need
| Input | What it means | Why it matters |
|---|---|---|
| A CPM schedule with logic | Activities linked by predecessors and successors, with lags and calendars | The simulation recalculates the network every run. Without logic, a delay can't travel to the finish. |
| Three-point durations | A minimum, most likely and maximum duration per activity | This is the ordinary estimate uncertainty: how long the work could take |
| Risk events | Discrete risks with a probability and an impact range, linked to the activities they hit | Things that may or may not happen, such as a permit delay or a late delivery |
| Correlation | Which durations tend to run long together | Activities with a shared cause (same contractor, same weather) move together. Ignoring that makes the range too narrow. |
AACE International's Recommended Practice 57R-09 builds the analysis on the same foundation: a CPM schedule with logic tight enough that the simulation finds the right dates and critical paths by itself, with risks used as drivers of the activity durations.
The steps
- Check the schedule. Every activity needs a predecessor and a successor, except the start and finish. Remove hard date constraints that aren't real, and fix negative float and missing links.
- Collect three-point durations. Ask the people who do the work for a realistic minimum and maximum, not a few percent either side. Focus on the long and critical activities first.
- Map the risks. Take the risks from the register that affect time, give each a probability and a three-point impact, and link them to the activities they delay.
- Set correlation. Group activities that share a cause and give them a correlation. Keep it simple: a few groups beat a large matrix nobody can explain.
- Run the simulation. Thousands of iterations, each one sampling the durations and risks and recalculating the network. Use a fixed random seed so the result can be reproduced.
- Read the outputs and act. Pick a commitment level, find the drivers, test mitigations, and repeat at every phase gate.
Worked example: Hollin Sands Wind Phase 2
Illustrative case. The project and figures are fictional, taken from our SRA Pro sample report.
An offshore wind project has 45 activities, 57 logic relations and 25 risks, simulated 10,000 times. The target finish is 30 Jun 2028. The deterministic schedule, with every activity at its most likely duration and no risks, finishes on 26 Oct 2028. The simulation says:
| Level | Finish date | vs deterministic |
|---|---|---|
| Deterministic | 26 Oct 2028 | - |
| P50 | 10 May 2029 | +196 days |
| P80 | 21 Jul 2029 | +268 days |
| P90 | 29 Aug 2029 | +307 days |
The target lies 386 days before the P80 date. Holding it as an external commitment means accepting a much higher than 20% chance of overrun. The report puts three options on the table: move the commitment to P80, buy part of the gap back with mitigation, or change scope or logic.
What the outputs mean
P50, P80 and P90 dates. The P80 date is the date the project finishes on or before in 8 out of 10 simulated runs. P50 is a coin toss. P90 is for hard deadlines with penalties. In the example, going from P50 to P80 costs 72 days, and P80 to P90 another 39. The 111 days between P50 and P90 is the practical width of the risk. P50, P80 and P90 explained goes deeper.
Criticality. The criticality index is the share of runs in which an activity is on the critical path. In the example, 20 of the 45 activities are critical in more than 80% of runs, and 13 are never critical. The ones in between deserve attention: they only become critical when things go wrong.
Sensitivity (tornado). A tornado chart ranks activities by how strongly their duration moves the finish date. Here the top driver is jacket fabrication, with a correlation of 0.27. A better estimate for that one activity, such as a firm supplier quote, narrows the whole distribution.
Risk contribution and what-if. The risks can be ranked the same way. The offshore weather window risk accounts for 21.1 days of the 198-day mean delay. Removing it entirely would bring the P80 forward by 53 days. That's the number to weigh against the cost of the mitigation.
Pitfalls that make a QSRA wrong
- Open ends and missing logic. An activity without a successor can slip as far as it likes without moving the finish. The simulation then looks calmer than the project is.
- Hard constraints. A "finish no later than" date pins an activity in place, so the delay simply disappears. Replace constraints with logic, or with a milestone you track separately.
- Ignoring merge bias. Adding up durations along the critical path, or using PERT on one path, misses the parallel paths that overtake it. Only simulating the full network catches it.
- Ranges that are too narrow. Plus or minus 10% on everything is a sign the ranges weren't really discussed. Narrow inputs give a steep S-curve and false comfort.
- No correlation. When durations are treated as independent, the long ones and the short ones cancel out. The result is a range that's narrower than reality, especially in the P80 to P90 tail. See correlation in risk analysis.
- Counting a risk twice. If a risk is already in the maximum duration, don't add it again as a risk event.
When Excel is enough, and when it isn't
You can build a QSRA in a plain Excel workbook for a handful of activities in series: sample each duration, add them up, and read the percentiles. Once there are parallel paths, lags, different link types and calendars, you need a real CPM calculation in every iteration. Writing and testing that in formulas is slow and error-prone.
A purpose-built Excel tool fits when you want to stay in Excel, the schedule has tens to a few hundred activities, and you need a board-ready result. A dedicated schedule risk product fits better when the schedule has thousands of activities, several schedulers work on it at once, or you need integrated cost and schedule risk across a program. Excel tools, add-ins or dedicated risk software? compares the options fairly.
Do it in Excel
Schedule Risk Analyzer Pro runs a QSRA in Excel on your real CPM network. It gives P50, P80 and P90 finish dates, criticality, a tornado and a what-if table for mitigations. It imports Primavera P6 XER files and produces a board pack. To understand why parallel paths make you late, read merge bias: why parallel paths make you late.
References
- US Government Accountability Office (2015). GAO-16-89G, Schedule Assessment Guide: Best Practices for Project Schedules, Best Practice 8, "Conducting a Schedule Risk Analysis".
- AACE International. Recommended Practice 57R-09: Integrated Cost and Schedule Risk Analysis Using Risk Drivers and Monte Carlo Simulation of a CPM Model.