A three-point estimate gives an activity three durations instead of one: optimistic (things go well), most likely, and pessimistic (things go badly, short of a discrete risk event). The range shows how uncertain the estimate is. A Monte Carlo simulation samples durations within those bounds, usually with a triangular or PERT (beta) distribution, to produce a range of possible finish dates instead of a single date.
The same idea applies to cost estimates, where the three points are low, most likely and high costs for a line item.
How it works
- Pick the activities. Usually the critical and near-critical work, plus long activities with real uncertainty. Ranging every activity is rarely worth the effort.
- Ask the people who do the work. The superintendent, the trade foreman or the estimator gives the three values and the reasons behind them.
- Record the basis. Note who gave the range and why, for example “two crane wind days per month in that season”.
- Choose a distribution. Triangular treats the range plainly. PERT (beta) concentrates probability around the most likely value. The choice changes the results, so state it.
- Simulate the network. A Monte Carlo run samples each ranged activity, recalculates the critical path for each iteration, and records the finish date. Thousands of iterations give a distribution of outcomes.
- Read the percentiles. The P80 is the date the model finishes on or before in 80% of iterations.
The two common formulas for a single activity:
| Mean | Notes | |
|---|---|---|
| Triangular | (O + M + P) ÷ 3 | Gives the extremes more pull |
| PERT (beta) | (O + 4M + P) ÷ 6 | Standard deviation is often approximated as (P − O) ÷ 6 |
PERT, the Program Evaluation and Review Technique, came out of US Navy program management in the late 1950s. Its weighted mean is still the most widely quoted three-point formula.
Worked example
Illustrative: structural steel erection is planned at 40 working days. The erector gives optimistic 36, most likely 42 and pessimistic 55, because wind days that stop the crane are common in that season.
| Measure | Calculation | Result |
|---|---|---|
| Triangular mean | (36 + 42 + 55) ÷ 3 | 44.3 days |
| PERT mean | (36 + 4 × 42 + 55) ÷ 6 | 43.2 days |
| PERT standard deviation (approx.) | (55 − 36) ÷ 6 | 3.2 days |
| Chance of 40 days or less (triangular) | (40 − 36)² ÷ ((55 − 36) × (42 − 36)) | about 14% |
| P80 for this activity alone (triangular) | 55 − √(0.2 × 19 × 13) | about 48 days |
Three things stand out. The planned 40 days is optimistic against the erector’s own view: under the triangular distribution it has about a one-in-seven chance on this activity alone. The range is skewed to the right, so both means sit later than the most likely value. And an 80% confidence duration for this one activity is around 48 days, eight more than the plan.
Those are single-activity numbers. At project level, the P80 finish comes from simulating the whole network, because steel may or may not be on the longest path in every iteration and other ranged activities interact with it.
Three-point estimates vs risk register events
| Three-point estimate | Risk register event | |
|---|---|---|
| Describes | Ordinary variability in a planned activity | A discrete event that may or may not happen |
| Example | Steel erection 36 / 42 / 55 days | “Utility power energized late”, 30% probability |
| In the model | Every iteration samples a duration | The event occurs only in some iterations |
Keeping them separate prevents double counting. If the pessimistic steel duration already allows for a late steel delivery, the same late delivery shouldn’t also appear as a risk event.
Why it matters
A single-point schedule implicitly claims every activity will take exactly its planned duration. Three-point ranges show how much confidence the plan deserves and where schedule contingency is needed. A risk analysis is only as credible as its ranges. Ranges that come from the people doing the work, recorded with their source, are defensible. Generic default ranges, such as minus 10% and plus 20% on everything, are a sensitivity test, not a forecast.
Common mistakes
- Anchoring on the plan. Setting most likely equal to the scheduled duration by default, even when the crew says otherwise.
- Ranges that are too narrow. People tend to give tight ranges. Asking what the worst realistic case looked like on the last similar job helps.
- Counting risks twice, once in the pessimistic value and again as a register event.
- Adding pessimistic durations along a path and calling it the worst case. That ignores how paths interact and overstates the result.
- Not recording the source, so nobody can defend or update the range later.
- Ignoring correlation. If cold weather slows concrete, it probably slows masonry too. Treating them as independent narrows the result.
How os.construction handles it
The Risk page takes three-point estimates in days or percent, typed, pasted or applied from templates, and records each one’s basis. Every result states whether its estimates were real, mixed, default or fixed. See Risk analysis.
FAQ
What is a three-point estimate?
A three-point estimate describes an activity's duration, or a cost, with three values instead of one: an optimistic value if things go well, a most likely value, and a pessimistic value if things go badly. The spread between them expresses how uncertain the estimate is.
What is the PERT formula for a three-point estimate?
The PERT (beta) mean is (O + 4M + P) ÷ 6, where O is optimistic, M most likely and P pessimistic. A common approximation of the standard deviation is (P − O) ÷ 6. The weighting of four on the most likely value pulls the mean toward it.
What is the difference between triangular and PERT estimates?
Both use the same three points. The triangular distribution's mean is (O + M + P) ÷ 3, which gives the tails more weight. The PERT or beta distribution's mean is (O + 4M + P) ÷ 6, which concentrates probability around the most likely value. For a right-skewed range, the triangular mean is later than the PERT mean.
Should a pessimistic estimate include risk events?
Usually not. The pessimistic value should reflect ordinary bad conditions for that activity, such as slow productivity or weather within normal seasonal ranges. Discrete events that might or might not happen, such as a late permit or a failed inspection, belong in a risk register, so the same delay isn't counted twice.
Can you add up three-point estimates to get the project duration?
Not by simple addition of the pessimistic or optimistic values. Activities on different paths overlap, and the longest path can change from one scenario to the next. A Monte Carlo simulation samples each activity's range, recalculates the network thousands of times, and reports the spread of finish dates, such as the P50 and P80.