What a P80 date really promises

Why one finish date is not enough

A critical path schedule gives one finish date, calculated from one duration per activity. Those durations are estimates, and most of them are optimistic. When two or more paths meet at a milestone, the later one wins every time, so the chance that the milestone is late is higher than the chance that either path is late. This is called merge bias, and it is why the single calculated date is usually the best case, not the likely one.

How the simulation works

1. Give each uncertain activity three durations: optimistic, most likely and pessimistic.
2. Choose a distribution. Triangular is simple; PERT puts more weight on the most likely value.
3. Add risk events: things that may or may not happen, each with a probability and a delay.
4. Run the schedule a thousand times or more. Each run draws a duration for every activity and recalculates the critical path.
5. Sort the finish dates. The date that half the runs meet is P50; the date that 80% meet is P80.

Two details decide whether a result can be trusted. The sampling method should cover the whole range evenly; Latin Hypercube does that with fewer runs. And the run should be repeatable: a fixed seed gives the same answer for the same inputs, so a reviewer can check it.

What P80 means, and what it does not

P80 means: given the ranges and risks we entered, 80% of the simulated outcomes finished on or before this date.

It does not mean that there is an 80% chance in the real world. The model only knows the uncertainty you gave it. Risks nobody listed, scope changes and a late decision by the client are not in it.

A fair way to say it to a sponsor: "The schedule says 3 May. With the risks we know about, we have an even chance of 22 June and four chances in five of 17 July. I recommend we commit to July and manage to May." That sentence gives a target, a commitment and the size of the gap between them.

Read the drivers, not only the date

The most useful output is the list of what moves the date.

• Criticality index. The share of runs in which an activity was on the critical path. An activity with float in the plan and 40% criticality deserves a second look.
• Sensitivity. How strongly the finish date follows each activity's duration. The top five are where mitigation pays.
• Risk events. Remove one risk and run again. The change in P80 is what that risk costs in days.

Four ways the result misleads

1. Narrow ranges. If every estimate is plus or minus 10%, the result looks certain. Ask the people doing the work what the bad case really was last time.
2. A weak schedule underneath. Open ends and hard constraints stop delay from flowing through. Run the DCMA checks first.
3. Independent durations. If one cause, such as weather or a slow approver, hits many activities, treating them as independent makes the range too narrow. Model the cause as a risk event.
4. Too much precision. Report a month or a week, not a day. The inputs are not that good.

Critova runs the simulation on your schedule with triangular or PERT durations, risk events, Latin Hypercube sampling and a fixed seed, and reports P50, P80 and P90 with drivers and criticality. See risk and Monte Carlo.