Most capital programs report one completion date. It appears in the monthly report, the lender's package and the board paper, and it is almost always the deterministic finish of the current schedule. What it rarely carries is any statement of how likely it is.
That omission matters. A deterministic date assumes every remaining activity finishes on its planned duration and no risk event occurs. On a complex program that is not a forecast. It is the best case, and it is treated as a commitment by everyone downstream.
What P50 and P80 actually mean
A schedule risk analysis applies uncertainty ranges to durations and models discrete risk events, then simulates the program many times. The output is a distribution of possible completion dates. The P50 is the date the program has an even chance of meeting. The P80 is the date it has a four-in-five chance of meeting.
On programs we review, the deterministic date frequently sits well to the left of the P50, sometimes so far left that its own probability is uncomfortably low. That is not a modeling quirk. It is the cumulative effect of merge bias: where several paths converge, the latest one governs, and the more paths there are, the more likely one of them runs late.
If you announce your P50, you have chosen, in advance, to be late half the time.
Which date to commit to
We encourage a simple convention:
- Manage the team to the P50, so that the plan stays ambitious and float is not padded
- Commit externally, to boards, lenders, tenants and the public, at or near the P80
- Hold the gap between them as schedule contingency, owned by the sponsor and released by decision
- Report movement in both dates every period, with the drivers named
- Re-run the analysis whenever the risk register or the logic changes materially
The gap is not waste. It is the honest price of uncertainty, made visible, and it gives the sponsor something tangible to govern.
A risk analysis is only as good as its schedule
Probabilistic forecasting magnifies the quality of the network beneath it. Missing logic, hard constraints and unrealistic durations do not disappear when a simulation is run; they produce a confident-looking distribution that is wrong. A health assessment is the prerequisite, not an optional extra.
Three-point estimates also need discipline. Ranges drawn up in a single workshop tend to cluster around the plan. Better practice is to derive them from historical performance where it exists, to separate uncertainty in durations from discrete risk events, and to model correlation between related activities so the results are not artificially narrow.
Reading the tornado
The most useful output of a risk analysis is often not the dates but the sensitivity ranking: which activities and risks contribute most to the spread. That list should drive the agenda of the next risk meeting. If the top driver is a utility energization date, mitigation effort belongs there, not in shaving days from finishes.
What changes for the board
When boards receive a P50 and a P80, the conversation changes. Questions move from whether the date will hold to what it would take to raise its confidence, and what that is worth. Contingency stops being a hidden buffer and becomes an explicit decision. And when the program does move, the movement is a change in odds that was already on the table, not a surprise.
