Project Confidence Simulator

What the simulator answers

A plan gives you one date and one budget. It does not tell you how likely either is, because the single-point estimate behind it has already thrown that information away. The simulator puts the uncertainty back: you give a range for each remaining phase instead of a single number, and it works out what range of outcomes those ranges actually produce.

• How likely is my current committed completion date?
• What dates correspond to P50, P80 and P90 confidence?
• How much contingency separates my commitment from a higher-confidence outcome?
• Which parts of the project are driving the uncertainty?
• If cost is enabled: how likely is my current budget, and what budget would be more defensible?

Three-point estimates

Each workstream is described with three durations rather than one. Optimistic is a genuinely good run where the things that usually go wrong do not. Most likely is what you would put in the plan. Pessimistic is a bad but credible run, not a disaster scenario. The gap between optimistic and pessimistic is the honest statement of how much you do not yet know, and it is the single input that most affects the answer.

Beta-PERT

Those three numbers are turned into a probability distribution using Beta-PERT, the standard distribution for project estimation. It concentrates most of the probability around your most likely figure while still allowing the extremes, which matches how project phases behave: they usually land near the plan, occasionally run long, and rarely finish dramatically early. A triangular distribution, the other common choice, puts a sharp corner on the most likely value and gives your bounds more weight than they deserve.

• The weighting on the most likely estimate is 4, which is the conventional PERT weighting and the reason the classical PERT mean is (optimistic + 4 x most likely + pessimistic) divided by 6.
• Where all three values are the same, the phase is treated as a fixed duration and contributes no uncertainty.

How dependencies are evaluated

Workstreams are connected finish to start: a phase begins when the last of its predecessors finishes. On every single simulation the whole network is resolved again, so two phases running in parallel cost what the longer one costs rather than what both cost added together. This is why the simulator will often return a shorter project than a spreadsheet that totals every phase, and why a long phase off the critical path can turn out not to matter.

How known risks are incorporated

A known risk is an event that may or may not happen. On each simulation it is tested against its probability, and where it fires, a delay is drawn from the range you gave and applied either to the whole project or to one workstream, whichever you chose. Across ten thousand simulations a 30% risk fires in roughly three thousand of them, which is what lets its effect appear in the spread of outcomes rather than as a single padded number.

• A risk with a financial impact uses the SAME trigger for both consequences, so it never delays the project in a simulation where it cost nothing.
• A risk assigned to a workstream lengthens that workstream, so whether it delays the project depends on whether that workstream was on the critical path in that simulation.

How the simulation works

Each simulation draws one duration for every workstream, tests every risk, resolves the dependency network and records when the project finished. Doing that 10,000 times produces a distribution of completion dates rather than a single answer, and the percentiles are read straight off it. This is Monte Carlo simulation: no formula predicts the outcome, the outcome is measured from many simulated runs.

Why 10,000 simulations

Enough that the percentile dates stop moving between runs at the resolution this product reports them, which is whole working days, and few enough that the whole analysis finishes while you are still looking at the screen. More simulations would cost time without changing any answer you would act on differently.

What modelled confidence means

Confidence in your target date is the share of simulated outcomes that finished on or before it. If four thousand three hundred of ten thousand simulations finished by 30 June, the target has approximately 43% modelled confidence. It is a statement about this model under the assumptions you entered, not a probability about the world and not a guarantee. Change the ranges and the number changes, which is exactly as it should be.

P50, P80 and P90

A percentile is the point by which that share of simulated outcomes had finished. P50 is the balanced planning level: half the simulations finished by then, so it is as likely to be early as late. P80 is the level most delivery organisations commit externally, because one run in five going past it is a risk many sponsors accept. P90 is a conservative level for commitments that are expensive to miss. None of them is universally correct, and the simulator deliberately does not tell you which to pick.

Why a most-likely plan is usually less likely than it looks

Teams are often surprised that a plan built from most likely durations does not come out at around 50% confidence. It usually comes out far lower, and the reason is that simulated outcomes are not spread evenly around the plan. Three effects push the distribution later, and they compound across a project.

• Each phase's own average sits later than its most likely value whenever there is more room above that value than below it, which is how almost everyone estimates: a phase can overrun badly but can only finish so early.
• Risk events only ever add time. There is no equivalent event that gives time back.
• Where branches converge, the project waits for the latest of them. A branch running late is inherited by the whole project, while a branch running early is simply not used.

Schedule contingency and the target gap

The target gap is the difference in working days between your current commitment and the P80 outcome. Where P80 is later, it is the contingency you would need to add to reach that confidence level. Where your target is already later than P80, the gap is reported the other way round, as contingency you already hold, rather than as a negative number that reads like a problem.

Cost confidence

Cost confidence is optional and adds sections rather than changing the schedule answer. It samples a Beta-PERT distribution over your optimistic, most likely and pessimistic forecast cost on every simulation, adds the financial impact of any risk that fired in that same simulation, and reports the probability of staying within your budget along with P50, P80 and P90 cost outcomes. The P80 figure is described as a higher-confidence planning level, never as the correct budget.

Uncertainty drivers

The simulator ranks what is actually driving the spread of outcomes, which is not the same as what is longest. It measures the rank correlation between each workstream's sampled duration and the total project duration across every simulation, so a phase whose variation consistently moves the project ranks highly and a long phase with a tight range, or one off the critical path, does not.

• Scores are relative, with the strongest driver at the top. They are correlations and deliberately not presented as percentage shares of delay, because they do not add up to the overrun and labelling them that way would be precise-looking and false.
• Risk events are ranked the same way, by how strongly their firing moves the project.

Scenario re-simulation

Change an assumption and simulate again, and the previous result is kept beside the new one so you can see what moved. This is how the simulator is meant to be used: not to produce one number, but to test whether a specific intervention, tightening an estimate, removing a dependency, mitigating a risk, would actually buy back the confidence you need.

Reproducibility and the simulation seed

Random does not mean unrepeatable. Every analysis carries a seed, and the same inputs with the same seed and the same model version reproduce the same result exactly, on any machine, at any time. That is what makes it safe to put a simulated date in a steering pack: the figure can be reproduced later by anyone who has the analysis.

The working calendar

All durations are working days, and dates are calculated Monday to Friday. Public holidays are not applied, so a project running through a holiday period will finish later than the dates shown. This is stated rather than defaulted to one country's calendar, because inventing somebody else's public holidays would be inventing a fact about their project.

How to read the result responsibly

Treat the confidence figure as a measurement of your own assumptions, not as a forecast handed down from outside. If the number is uncomfortably low, the useful responses are to improve the estimate ranges, to attack the top uncertainty drivers, or to move the commitment, in that order. The least useful response is to re-enter narrower ranges until the number looks better, which changes the output without changing the project.