PERT (Program Evaluation and Review Technique)

How PERT Works

The Program Evaluation and Review Technique is a statistical estimation method that accounts for uncertainty by using three estimates instead of one. For each task, the estimator provides an optimistic estimate (the best case if everything goes right), a most likely estimate (the duration if normal conditions prevail), and a pessimistic estimate (the worst case if significant problems occur). The PERT formula then calculates a weighted average that gives the most likely estimate four times the weight of the extremes.

PERT was developed by the US Navy in 1958 for the Polaris missile submarine program, which involved over 3,000 contractors and required unprecedented schedule coordination. The technique addressed a fundamental problem: single point estimates hide uncertainty. A task estimated at “10 days” could mean anywhere from 7 to 20 days depending on conditions. PERT makes that uncertainty explicit and computes a statistically grounded expected duration.

The technique assumes a beta distribution for task durations, which is slightly skewed toward the pessimistic end. This reflects the empirical reality that tasks are more likely to take longer than expected than shorter. The weighting (1:4:1 for optimistic, most likely, pessimistic) produces an expected value that accounts for this asymmetry.

The PERT Formula

Expected Duration (tE) = (O + 4M + P) / 6, where O is the optimistic estimate, M is the most likely estimate, and P is the pessimistic estimate. The formula divides by 6 because the weights (1 + 4 + 1) sum to 6.

Standard Deviation (σ) = (P minus O) / 6. This measures the uncertainty of the estimate. A large standard deviation means high uncertainty (wide spread between optimistic and pessimistic). A small standard deviation means the estimate is relatively certain.

Variance = σ squared = ((P minus O) / 6) squared. Variances are additive across sequential tasks, which allows calculation of the total path uncertainty by summing individual task variances along the critical path.

Worked Example

A software migration task has three estimates: optimistic = 5 days, most likely = 8 days, pessimistic = 17 days. Expected duration = (5 + 32 + 17) / 6 = 54 / 6 = 9 days. Standard deviation = (17 minus 5) / 6 = 2 days. The task is expected to take 9 days with a standard deviation of 2 days, meaning there is roughly a 68% probability the task will finish between 7 and 11 days (one standard deviation on either side).

Compare this to a single point estimate of “8 days” (the most likely value). The PERT estimate of 9 days is more conservative because the pessimistic case (17 days) is much further from the most likely (8 days) than the optimistic case (5 days). PERT captures this asymmetry. A single point estimate does not.

Key Characteristics of PERT

PERT produces two outputs per task: an expected duration and a measure of uncertainty (standard deviation). This dual output is its primary advantage over single point estimating. The expected duration feeds the schedule. The standard deviation feeds the risk analysis.

The technique is most valuable when tasks have significant uncertainty: new technology, unfamiliar scope, external dependencies, or limited historical data. For routine tasks with well known durations (running a standard report, deploying a tested build), PERT adds calculation overhead without improving estimate accuracy.

PERT estimates can be aggregated along the critical path to calculate the probability of meeting a target completion date. Sum the expected durations to get the expected path length. Sum the variances to get the path variance. Use the normal distribution to calculate the probability that the actual path duration falls within a target range. This is PERT’s most powerful application: turning a deterministic schedule into a probabilistic one.

When to Use PERT

Use PERT when tasks have significant uncertainty and the schedule needs to account for that uncertainty rather than hiding it behind a single number. Common scenarios include first of a kind projects (no historical data for analogous estimating), tasks with external dependencies (vendor delivery, regulatory approval), complex integration work where problems are likely but unpredictable, and any project where stakeholders need to understand the probability of meeting the target date rather than just seeing a deterministic schedule.

PERT is standard practice on large government and defense programs, construction projects with complex sequencing, and pharmaceutical development where regulatory timelines introduce significant uncertainty.

When Not to Use PERT

Routine tasks with well established durations do not benefit from three point estimation. If the team has done a task 20 times and it always takes 3 to 4 days, a single point estimate of 3.5 days is more efficient than gathering three estimates and running the formula.

PERT assumes task durations follow a beta distribution, which may not hold for tasks with binary outcomes (it either works or it does not) or tasks with multimodal distributions (two distinct scenarios with different durations). For these cases, Monte Carlo simulation provides more accurate modeling.

Teams that lack the discipline to provide honest three point estimates will not benefit from PERT. If the optimistic estimate is always “most likely minus 1” and the pessimistic is always “most likely plus 1,” the technique adds calculation overhead without adding information. The three estimates must be genuinely different scenarios, not arbitrary adjustments around a single guess.