A P80 contingency is the reserve that brings the budget up to the amount your project stays under in 80% of simulated outcomes. Setting one takes seven steps. Most of the work is in the first three.
1. Turn every cost item into a range
For each item in the cost breakdown, record three values: a realistic minimum, the most likely value and a realistic maximum. These three points become a Beta-PERT distribution, a curve that allows the upper tail to be longer than the lower one. That fits construction costs, which overrun more easily than they underrun.
Ask the right question. "How much margin is in it?" invites a reflex answer. "What would have to happen to end up at the maximum, and is that credible?" gets a real one. A range nobody can back up with a story isn't a range.
2. Add the risk events, without counting twice
Discrete risks come on top as a separate layer: events that either happen or do not, each with a probability and an impact range. A stalled permit procedure or a supplier going under belongs here.
The key rule: only include risks that aren't already inside the range of an item. If slow progress on the bored tunnel is already in that item's maximum, a separate "slow progress" risk counts it twice. A double count inflates the reserve and is just as hard to defend as one that's too low.
3. Record which items move together
When the steel price rises, every steel-heavy item gets more expensive at the same time. A model that treats items as independent lets good and bad luck cancel out, and produces a distribution that's too narrow. That shows up exactly where it hurts, in the P80 and P90. You don't need precise figures: the difference between zero and a substantial correlation is what matters. Our guide on correlation shows how large the effect is.
4. Simulate, and fix the seed
Run the project at least 10,000 times. Each run draws a value for every item and decides for every risk whether it occurs. Two settings matter for anything you'll report:
- A fixed seed, so the same input always gives exactly the same figure on any machine.
- Enough iterations. A simulated P80 is itself an estimate with a small sampling margin. For a working session 10,000 runs is fine. For a figure in a funding or procurement document, 25,000 to 50,000 runs keeps two runs from drifting apart.
5. Read the S-curve
The S-curve shows, for every amount, the chance of staying below it. Pick a confidence level on the vertical axis, go across to the curve and down to the amount. The difference between that amount and the base estimate is your contingency. In the tunnel case:
| Level | Budget | Contingency above base | Typical use |
|---|---|---|---|
| P50 | $476.1M | $60.1M (14.5%) | Median. A 50% chance of overrun, too tight as a budget |
| P80 | $507.7M | $91.7M (22.0%) | Common control level; anchor under shared risk |
| P85 | $515.3M | $99.3M (23.9%) | Upper end of the range for some public owners |
| P90 | $524.6M | $108.6M (26.1%) | Toward fixed-price risk |
Note how the last few percent of confidence get expensive. That trade-off is the decision the board has to make.
6. Choose the level, and say why
No standard picks the number for you. AACE International provides the structure for contingency, but leaves the confidence level to the organization's risk appetite and the decision context. Frameworks differ. Some owners use P85 as the upper end; the forecast for the Dutch national flood defense program, for example, was reported as a range from P15 to P85 (Horvat & Partners, 2024). Australian Commonwealth practice ties P50 to the first decision point and P80 to the second. P50 and P80 are common elsewhere.
The contract form matters too. Under a fixed-price contract, a contractor carrying the cost risk will price toward the upper tail. Under an alliance, P80 is often the anchor for the shared budget, with agreed sharing above and below it. Whatever you choose, write down which framework you follow. In the tunnel case P50 and P85 are $39M apart.
7. Structure it, then repeat it at every gate
A percentile isn't yet a budget an auditor accepts. Split the reserve into parts with their own origin and owner: an allowance tied to the maturity of the estimate, the probabilistic contingency from the simulation, and a management reserve held above the project. Record the assumptions behind each range, so the reserve can be explained in six months' time.
Then rerun the analysis at every gate. As the design matures, the ranges narrow and the contingency drops. That isn't because the project became safer, but because you know more.
Do it in Excel
QRCIM follows these seven steps in one workbook: ranges per item, risk events, correlation, a seeded simulation, the S-curve and an AACE-style budget build-up, with quality checks before anything is reported. If you start from a risk register rather than a full estimate, the Cost & Schedule Risk Analyzer gives you the reserve in money and weeks at P50, P80 and P90. To see the principle first, run the simulator on our homepage. Prefer to build the simulation yourself? How to calculate contingency with Monte Carlo gives the Excel formulas.