Setting contingency at P50 versus P80: what the confidence level actually buys you

Start with what P50 and P80 actually mean

A P50 contingency is the value your simulation output shows as being met or beaten in 50% of simulated outcomes — the median. A P80 is the value met or beaten in 80% of outcomes. Moving from P50 to P80 does not change the underlying risk profile; it changes how much of the upside tail you are choosing to cover.

This distinction matters because P50 and P80 are routinely confused with "average" and "worst case" respectively. Neither is correct. P50 is the median, not the mean, and a right-skewed cost distribution (the normal shape for construction risk) will usually show a mean above the median. P80 is not a worst case — it is simply a more conservative confidence threshold, with real probability remaining above it.

Match the confidence level to what the number is used for

• P50 is appropriate for internal planning and progress tracking, where the organisation can absorb variance across a portfolio of projects and does not need single-project headroom.
• P80 (or higher) is more common for external reporting, funding approval, or any context where being wrong in the expensive direction has consequences beyond the project itself — a regulated utility's price control submission, for example, or a funding ask that cannot easily be revisited.

There is no universally "correct" answer. The choice is a governance decision about risk appetite, not a technical one about which number is more accurate.

Defending the choice in a board paper

1. State the confidence level explicitly, not just the resulting figure — "P80 contingency: £42m" is a defensible sentence; "contingency: £42m" invites the question of what it actually covers.
2. Show the distribution, not just the single figure. A cumulative distribution chart alongside the number lets a board see how much probability sits above and below the chosen threshold.
3. Explain what drives the tail. If a small number of risks account for most of the spread between P50 and P80, name them — a board can engage with "these three risks separate our P50 from our P80" far more usefully than with a bare percentage.
4. Be consistent run to run. Changing the confidence level between reporting periods to manage the headline figure is the fastest way to lose credibility with a board that is tracking contingency drawdown over time.

The number is not static

A contingency set at a given confidence level should be revisited as the register and schedule evolve — new risks, resolved risks and updated estimates all move the underlying distribution. Re-running the simulation at each reporting milestone, rather than only once at baseline, is what keeps the confidence-level statement honest.