"We need a P80 number." If you work on large projects, you have probably heard that in a meeting. But what does it actually mean, and why not P50 or P90?
P50, P80 and P90 are confidence levels: the cost you won't exceed with a 50%, 80% or 90% probability. So a P80 of $143 million means that in 80 out of 100 plausible versions of the project, the final cost ends up at or below $143 million. It doesn't mean "the estimate plus 20%." That's probably the most common misunderstanding about confidence levels.
These numbers come out of a probabilistic estimate, usually a Monte Carlo simulation (the practitioner's guide to quantitative cost risk analysis explains how that works). Here the focus is on what each level means, how to choose between them, and the mistakes that end up costing money.
What each level means
| Level | Meaning | Chance of overrun | Typical use |
|---|---|---|---|
| P50 | Half of the outcomes are lower, half are higher | 1 in 2 | Project budget managed tightly; reserve held above project level |
| P80 | Four out of five outcomes are lower | 1 in 5 | Common choice for a project's total budget including contingency |
| P90 | Nine out of ten outcomes are lower | 1 in 10 | Funding envelope or approval ceiling when a new funding decision would be very costly |
Keep two things in mind. The P50 (the median) is usually lower than the average. Cost outcomes are skewed: a few bad outcomes are really bad, and they pull the average up. And the jump from P80 to P90 is often bigger than you would expect, because the curve flattens out in the tail. Every extra percent of certainty costs more than the one before.
One project, three levels
Let's make it concrete with the fictional Millbrook Connector, a road and bridge project with a point estimate of $120 million. A Monte Carlo analysis gives:
| Level | Total ($M) | Contingency above estimate ($M) |
|---|---|---|
| P50 | 135.8 | 15.8 |
| P80 | 143.4 | 23.4 |
| P90 | 147.6 | 27.6 |
Going from P50 to P80 costs $7.6 million and cuts your chance of overrunning from one in two to one in five. Going from P80 to P90 costs another $4.2 million and halves that chance again. Is that worth it? That isn't a technical question. It depends on what an overrun would cost you: a new approval round, damage to your reputation, or other projects that have to wait.
How to choose the right level
Four questions help:
- What happens if we go over budget? If an overrun means going back to a council, a parliament or an investment committee, a higher level is cheaper in the end.
- Who else draws on the same reserve? In a portfolio, individual project budgets can sit at P50 while a shared reserve covers the whole portfolio. Not every project overruns at the same time, so a pooled reserve needs less money than a separate P80 budget for each project (Kujawski, 2002).
- How mature is the estimate? Early estimates have wider ranges. A P80 at concept stage is a very different number from a P80 at final design.
- What do the rules say? Some funders prescribe a level. The Australian Government asks for both a P50 and a P90 for projects above $25 million that seek Commonwealth funding. NASA budgets at a 70% joint cost and schedule confidence level.
A pattern that works for many owners: the project manager manages to the P50, the sponsor holds the difference up to P80, and anything beyond P80 goes back to the board. It keeps pressure on the project, protects the reserve, and makes escalation automatic instead of political.
Five common mistakes
- Reading P80 as "plus 20%." In the example above, P80 happens to be 20% above the estimate. On a riskier project it could easily be 40%.
- Comparing P-values from different models. A P80 is only as good as the ranges, risks and correlation behind it. Ask what is in the model before you compare numbers.
- Spending to the budget. A P80 budget should be underspent most of the time. If every project uses up its full contingency, the confidence level was never real.
- Splitting contingency across work packages. Money handed out per package tends to get spent per package. Hold it centrally and release it against risks.
- Treating the number as fixed. Run the analysis again at every stage. A falling P80 means risk is being retired. A rising P80 is an early warning.
Frequently asked questions
Is P85 different from P80?
Only in degree. It's the cost with an 85% chance of not being exceeded. Some organizations use P85 for the upper end of a range.
What is a joint confidence level?
The probability of meeting both the cost and the schedule target at the same time (see the glossary). It's always lower than the cost-only or schedule-only confidence level.
Should a contractor and an owner use the same level?
Not necessarily. A contractor prices the risks it carries; the owner budgets for the risks it keeps.
Quantil's RCAT weighs a Monte Carlo P80 against two cross-checks to recommend the reserve, and splits it into project contingency for the project manager and management reserve for the sponsor. See how it works.
References
- Department of Finance, Australian Government. Defining P50 and P80.
- Department of Infrastructure, Australian Government (2023). Guidance Note 3B: Deterministic Contingency, v2.0.
- Kujawski, E. (2002). Why projects often fail even with high cost contingencies. Systems Engineering, 5(2), 151–155.
- NASA (2019). NID 7120.122: JCL Requirements Updates.