Duration Estimation, PERT and Schedule Uncertainty

Introduction

A network schedule is only as credible as its duration estimates. In project environments, activity durations are rarely known with certainty: requirements may change, productivity may vary, approvals may be delayed, and technical problems may arise. Consequently, the project manager must distinguish between a deterministic schedule—one based on single-point durations—and a probabilistic schedule that explicitly represents uncertainty.

This chapter develops four related capabilities: selecting appropriate estimation methods, applying three-point and PERT estimation, calculating completion probabilities, and interpreting simulation-based measures such as criticality indices and confidence levels. These techniques do not eliminate uncertainty. Rather, they make assumptions visible and support better decisions concerning contingency, commitments, risk responses and stakeholder communication.

Learning Objectives

By the end of this chapter, the learner should be able to:

1. Select analogous, parametric, bottom-up and three-point estimation methods.
2. Explain the assumptions underlying PERT expected duration and variance.
3. Calculate project duration and completion probabilities for suitable networks.
4. Interpret Monte Carlo simulation, criticality indices and confidence levels.
5. Evaluate estimation limitations and design an appropriate schedule-risk response.

1. Duration Estimation in Network Scheduling

An activity duration is the time required to complete an activity under stated conditions. It should be expressed with its unit—hours, days or weeks—and with assumptions concerning resources, calendars, working time, productivity and scope. Duration is not the same as effort: effort measures labour or resource time, whereas duration measures elapsed calendar time.

For example, 40 labour-hours may represent five working days for one person or two and a half days for two people. The latter conclusion is valid only if the work is divisible, resources are available, coordination costs are negligible and additional personnel do not reduce productivity.

A professional estimate normally records:

• the activity scope and deliverable;
• the estimating method and source data;
• resource assumptions and calendar;
• uncertainty range and confidence level;
• exclusions, dependencies and constraints; and
• the estimator’s rationale and date of preparation.

1.1 Analogous Estimation

Analogous estimation uses the actual duration of a comparable previous activity or project, adjusted for differences in size, complexity, technology, location, team capability or constraints. It is rapid and useful during initiation when detailed information is limited.

Its main strength is managerial efficiency. Its main weakness is dependence on genuine comparability. A previous software module may not be a valid analogue for a safety-critical module even if both contain a similar number of functions. Analogous estimates should therefore identify the adjustment logic rather than simply copy historical duration.

1.2 Parametric Estimation

Parametric estimation relates duration to measurable quantities using a statistical or productivity model. A simple form is:

where D is estimated duration, q is the quantity of work, and r is duration per unit. Units must be consistent. If installation productivity is 0.25 hours per metre and 800 metres must be installed, the estimated effort is 200 hours before calendar and resource adjustments.

A more general regression form is:

where a is fixed setup time and b is incremental duration per unit. Parametric models require relevant, sufficiently stable historical data. Their apparent precision can be misleading when the model omits learning, congestion, rework or scale effects.

1.3 Bottom-Up Estimation

Bottom-up estimation decomposes a deliverable into work packages and activities, estimates each component, and aggregates them. If components are sequential, durations are added. If components can proceed in parallel, the network logic—not simple addition—determines the elapsed duration.

Bottom-up estimation is usually the most defensible approach when scope is sufficiently defined. It is also resource-intensive and vulnerable to decomposition bias: omitted integration, testing, procurement, management and rework activities can create systematic underestimation.

1.4 Three-Point Estimation

Three-point estimation represents uncertainty through three values:

• optimistic duration, a: a favourable but credible outcome;
• most likely duration, m: the modal or best professional assessment; and
• pessimistic duration, b: an adverse but credible outcome.

These values should be based on explicit assumptions. “Optimistic” does not mean physically impossible, and “pessimistic” should not be an unlimited catastrophe scenario unless the model is specifically designed to include such risks.

2. PERT Expected Duration and Variance

The Program Evaluation and Review Technique (PERT) uses a beta-shaped approximation to represent activity-duration uncertainty. The traditional PERT expected duration is:

The corresponding variance is:

and the standard deviation is:

Here, T is activity duration; a, m and b are optimistic, most-likely and pessimistic durations in the same unit. Variance is measured in squared time units, whereas standard deviation is measured in time units.

The weighting of m reflects the traditional PERT assumption that the most-likely estimate is more informative than either extreme. The formulas are approximations, not empirical laws. They are most useful when the three estimates are credible, the distribution is reasonably unimodal, and activity relationships are represented correctly.

Worked Example 1: Activity PERT Estimate

Suppose an activity has a = 4 days, m = 7 days and b = 16 days.

The expected duration is 8 days, with a standard deviation of 2 days. The expected value is not the same as the most-likely value: the pessimistic tail pulls the weighted mean above 7 days.

Alternative Three-Point Conventions

Some organizations use a triangular mean:

This is not the traditional PERT estimate. The project team must document which convention is being used, because the two methods produce different means. For the example above, the triangular mean is 9 days, whereas the PERT mean is 8 days. Neither should be presented as universally correct without stating the underlying distributional assumption.

3. From Activity Estimates to Project Duration

For a purely serial critical path, expected durations and variances are commonly aggregated as follows:

where μ_P is expected project duration and σ²_P is project-duration variance for activities on the assumed critical path. The second relationship requires, in its simplest form, independent activity durations. If durations are correlated, covariance terms are required:

A positive covariance—such as several activities being delayed by the same supplier—raises total variance. Ignoring correlation can understate schedule risk.

For networks with parallel paths, the project completion time is the maximum completion time among paths, not the sum of all network activities. Therefore, simply adding expected durations across one initially identified critical path is an approximation. As uncertainty changes, a different path may become critical.

4. Completion Probabilities

When a project-duration distribution is approximated by a normal distribution, the probability of completing by target date S can be estimated using:

and

where μ_P is mean project duration, σ_P is standard deviation, and Φ is the cumulative standard normal distribution function. The normal approximation is most defensible when the project contains many approximately independent contributing activities and no severe skewness dominates the result.

Worked Example 2: Probability of Meeting a Date

Assume a serial critical path has expected duration 40 working days and standard deviation 5 working days. The sponsor requests completion within 47 working days.

The standard normal cumulative probability at Z = 1.4 is approximately 0.9192. Thus, under the stated assumptions:

This result does not mean that the project has a guaranteed 91.9% chance. It means that the probability is approximately 91.9% under the chosen estimates, independence assumptions and normal approximation.

A target date can also be derived from a desired confidence level c:

where z_c is the standard normal quantile associated with confidence level c. For a one-sided 90% confidence level, z_c is approximately 1.282; for 95%, it is approximately 1.645. These values should not be confused with a project-management “contingency reserve” unless the reserve has been explicitly calculated and governed.

5. Monte Carlo Simulation for Schedule Risk

Monte Carlo simulation repeatedly samples activity durations from specified probability distributions, applies network logic, and records project completion times. A generic procedure is:

1. Define the network, calendars and precedence relationships.
2. Assign a distribution or three-point model to each uncertain activity.
3. Sample one duration for every activity in an iteration.
4. Perform the schedule calculation to obtain project completion time and activity dates.
5. Repeat for many iterations.
6. Summarize the resulting completion-time distribution and activity behaviour.

Typical outputs include:

• mean and median completion date;
• percentiles, such as P50, P80 and P90;
• probability of meeting a fixed target date;
• finish-date histograms or cumulative curves;
• probability that an activity lies on the critical path; and
• sensitivity measures linking uncertain activities to total duration.

A P80 date means that approximately 80% of simulated outcomes finish on or before that date, assuming the model is valid. It does not mean that the remaining 20% are impossible, nor does it automatically specify management reserve.

Criticality Index

For activity i, the criticality index is commonly defined as:

where n_i is the number of simulation iterations in which activity i is critical, and N is the total number of iterations. If an activity is critical in 3,200 of 10,000 iterations, its criticality index is 0.32, or 32%.

Criticality index is not the same as duration sensitivity, cost impact or risk probability. An activity can have a high criticality index but little effect on the completion date if its duration variation is small. Conversely, an activity with a moderate criticality index may have substantial impact when it becomes critical.

Simulation Quality and Convergence

More iterations do not correct a poor model. Model validity depends on realistic distributions, correct logic, credible correlations, appropriate calendars and avoidance of double-counting risk. Simulation results should be checked for stability: increasing the number of iterations should not materially change key percentiles. Random seeds and model versions should be recorded to support reproducibility.

6. Confidence, Contingency and Decision Use

Confidence levels must be communicated precisely. “The schedule is 90% confident” is incomplete unless it states the event, model and date—for example, “The P80 completion date is 30 September, conditional on the approved scope and model assumptions.”

Three concepts should be distinguished:

• Estimate uncertainty: natural variation in how long planned work may take.
• Known-unknown risk: identified events whose probability and impact can be modelled.
• Unknown-unknown risk: events not represented in the model and therefore not captured by its percentile.

Contingency reserve is intended for in-scope uncertainty and identified risks, whereas management reserve is generally controlled for unforeseen work within the project’s governance rules. Terminology differs among organizations, so the project management plan should define the policy.

7. Illustrative Case: Digital Claims Transformation

Background

An insurance company is implementing a digital claims platform. The principal network is:

• A: configure workflow, then B: integrate payment service, then D: system test;
• A also precedes C: migrate claims data, and C also precedes D;
• D precedes E: user acceptance and deployment.

Thus, the two principal paths are A–B–D–E and A–C–D–E.

Initial Estimates

Using traditional PERT:

The expected duration of path A–B–D–E is approximately 29.33 days. The expected duration of A–C–D–E is approximately 32.00 days, so the data-migration path is initially longer. Its approximate path variance is 1 + 4 + 1.78 + 1 = 7.78 days², giving a standard deviation of approximately 2.79 days.

Analysis and Approach

The team should not conclude that the project duration is exactly 32 days. The two paths share A, D and E, and the path that controls completion may change across outcomes. A Monte Carlo model is therefore constructed with the precedence relationships, activity distributions and a positive correlation between C and D because both depend on the same test environment and data quality.

The simulation produces a P50 completion of 32 days, a P80 completion of 36 days and a P90 completion of 38 days. Activity C has a criticality index of 61%, D 74%, E 58%, A 49% and B 27%. These figures indicate that D is a particularly important schedule-control point, while B is less frequently critical than C. However, the team also examines sensitivity: C’s large duration range makes it a major source of completion-date variation.

Decision and Outcome

The sponsor’s public launch commitment is set at the P80 date rather than the mean. The team then reduces uncertainty by funding an early data-quality assessment, securing a test environment, and splitting the migration rehearsal from the production migration. A subsequent simulation shows the P80 date falling by two days and the P90 date by three days. The project has not merely added arbitrary buffer; it has reduced the variance-producing mechanisms.

Learning Points

1. The longest expected path is a starting hypothesis, not a permanent fact.
2. Shared activities and correlations matter.
3. A P80 commitment is a governance choice supported by analysis, not a guarantee.
4. Risk response can improve the distribution, not only move the selected date.
5. Criticality index should be interpreted alongside duration sensitivity and business impact.

8. Limitations and Professional Judgement

PERT has several limitations. The beta approximation may not represent a real activity distribution, especially when risk events create a long right tail. Three-point estimates can be affected by optimism bias, anchoring and strategic pressure. Experts may also provide correlated estimates without recognizing the correlation.

Network logic is equally important. Missing approval gates, procurement lead times, rework loops or resource constraints can invalidate a sophisticated simulation. Standard CPM calculations often assume unlimited resource availability; if a resource can perform only one activity at a time, resource-constrained scheduling may produce a longer and different critical sequence.

Schedule risk analysis should therefore be integrated with risk registers, assumptions logs, change control, resource planning and earned-value or milestone reporting. Results should be reviewed at major scope, design, procurement and testing gates.

9. Technology, AI and Governance

Modern scheduling tools can automate distribution fitting, simulation, sensitivity analysis and visualization. AI systems may assist by extracting historical durations, identifying similar projects, detecting anomalous estimates or suggesting missing dependencies. Their output remains advisory. Historical data may encode obsolete processes, selective reporting or organizational bias, and an AI-generated estimate may appear precise without being valid.

Good governance requires traceable data sources, version control, explainable assumptions, human approval, access controls and protection of commercially sensitive information. When AI uses employee or supplier performance data, privacy, fairness and contractual obligations must be considered. A model should never be used to conceal uncertainty from sponsors or to manufacture an apparently objective commitment.

Best-Practice Checklist

• Define duration units, calendars and resource assumptions.
• Use the estimation method appropriate to scope maturity and data quality.
• Record optimistic, most-likely and pessimistic assumptions explicitly.
• Distinguish effort from elapsed duration.
• Use PERT formulas only with clearly stated assumptions.
• Include covariance or correlated sampling where common causes exist.
• Model parallel paths, constraints, lags and rework realistically.
• Report percentiles and target-date probabilities, not only a single finish date.
• Interpret criticality index with sensitivity and impact measures.
• Validate simulations through scenario reviews and convergence checks.
• Re-estimate when scope, design, resources or external conditions change.
• Communicate confidence as conditional model information, not certainty.

Chapter Summary

Duration estimation progresses from rapid analogous estimates to data-based parametric estimates, detailed bottom-up estimates and uncertainty-aware three-point estimates. Traditional PERT calculates an activity’s weighted expected duration and approximates its variance, but its results depend on credible estimates and distributional assumptions. For a serial path, means and variances can be aggregated under independence; for realistic networks, parallel paths, shared activities and correlations require greater care.

Completion probabilities can be approximated with a normal model when its assumptions are defensible. Monte Carlo simulation is more flexible because it repeatedly applies uncertain durations to the actual network and produces percentiles, target-date probabilities and criticality indices. These outputs support decisions about commitments and contingency, but they do not replace professional judgement. The quality of schedule-risk analysis depends primarily on scope definition, network logic, data quality, correlation treatment, governance and honest communication of assumptions.

Key Terms

Duration: Elapsed time required to complete an activity.

Effort: Quantity of labour or resource time applied to work.

Analogous estimation: Estimation based on an adjusted comparison with a similar past activity or project.

Parametric estimation: Estimation using a quantitative relationship between work quantity and duration.

Bottom-up estimation: Estimation made by aggregating detailed component estimates.

Three-point estimation: Representation of duration using optimistic, most-likely and pessimistic values.

PERT: A probabilistic estimating technique using weighted three-point estimates.

Optimistic estimate: A favourable but credible activity duration.

Most-likely estimate: The duration judged most probable under stated conditions.

Pessimistic estimate: An adverse but credible activity duration.

Variance: A measure of dispersion expressed in squared time units.

Standard deviation: The square root of variance, expressed in time units.

Critical path: A precedence path controlling the project completion date under specified conditions.

Monte Carlo simulation: Repeated random sampling and schedule calculation to estimate outcome distributions.

Percentile: A value below which a specified percentage of simulated outcomes fall.

Confidence level: A stated probability associated with achieving a specified outcome under a model.

Criticality index: The proportion of simulations in which an activity is critical.

Correlation: Statistical association between uncertain durations.

Contingency reserve: Time or budget held for identified in-scope uncertainty or risk.

Management reserve: Governed reserve for unforeseen work, according to organizational policy.

Review Questions

Short Answer Questions

1. Distinguish duration from effort and explain why the distinction matters.
2. When is analogous estimation appropriate, and what is its principal weakness?
3. State the traditional PERT expected-duration and variance formulas.
4. What does a P80 completion date mean?
5. Define criticality index and explain what it does not measure.

Long Answer Questions

1. Compare analogous, parametric, bottom-up and three-point estimation in terms of data requirements, strengths and limitations.
2. Explain the assumptions behind aggregating activity variances on a critical path.
3. Describe a complete Monte Carlo schedule-risk analysis from model construction to reporting.
4. Critically evaluate the use of normal approximation for project completion probabilities.
5. Discuss how governance and AI-assisted estimation can improve or damage schedule credibility.

Analytical / Application Questions

1. An activity has optimistic, most-likely and pessimistic durations of 5, 8 and 17 days. Calculate its PERT expected duration, variance and standard deviation. Interpret the result.
2. A serial path has mean duration fifty days and standard deviation 6 days. Estimate the probability of finishing within 59 days using a normal approximation, and state the assumptions required.
3. Design a schedule-risk response for a project in which data migration has a high criticality index, strong positive correlation with testing, and a wide pessimistic estimate. Explain how your response would affect the simulation model and management decision.

Multiple Choice Questions with Answers

Question 1

An activity has optimistic, most-likely and pessimistic durations of 4, 7 and 16 days. What is its traditional PERT expected duration?

A. 7 days

B. 8 days

C. 9 days

D. 10 days

Correct Answer: B. 8 days

Question 2

For the activity with a = 4 and b = 16 days, what is the PERT variance?

A. 2 days²

B. 3 days²

C. 4 days²

D. 12 days²

Correct Answer: C. 4 days²

Question 3

Which estimation method uses a measurable relationship such as duration per metre installed?

A. Analogous estimation

B. Parametric estimation

C. Bottom-up estimation

D. Expert voting

Correct Answer: B. Parametric estimation

Question 4

A project has mean duration 40 days and standard deviation 5 days. What is the Z-value for a 47-day target?

A. 0.7

B. 1.0

C. 1.4

D. 2.4

Correct Answer: C. 1.4

Question 5

A criticality index of 0.32 means that an activity is critical in:

A. 32% of simulation iterations

B. 32% of the project duration

C. 32% of the activity’s work

D. 32 consecutive days

Correct Answer: A. 32% of simulation iterations

Question 6

Which condition most directly invalidates simple addition of path variances?

A. All durations are measured in days

B. Activities are positively correlated

C. The network contains written assumptions

D. The project has a baseline schedule

Correct Answer: B. Activities are positively correlated

Question 7

In a network with parallel paths, project completion time is generally determined by:

A. The shortest path

B. The average path duration

C. The maximum path completion time

D. The sum of every network activity

Correct Answer: C. The maximum path completion time

Question 8

What does a P80 completion date indicate?

A. The project must finish exactly on that date

B. Approximately 80% of modelled outcomes finish by that date

C. The estimate has an 80-day contingency

D. Every activity has 80% certainty

Correct Answer: B. Approximately 80% of modelled outcomes finish by that date

Question 9

If an activity is critical in 3,200 of 10,000 iterations, its criticality index is:

A. 0.032

B. 0.32

C. 3.2

D. 32.0

Correct Answer: B. 0.32

Question 10

Which is a principal limitation of Monte Carlo simulation?

A. It cannot represent uncertainty

B. It always assumes one critical path

C. It cannot use precedence relationships

D. It cannot correct unrealistic inputs or network logic

Correct Answer: D. It cannot correct unrealistic inputs or network logic