Uncertainty Analysis

Definition

Uncertainty Analysis quantifies the range of plausible outcomes for cost and schedule estimates by modelling both inherent variability (aleatory uncertainty) and knowledge gaps (epistemic uncertainty), together with the impact of identified risk events. The output is a probability distribution — typically expressed as P10, P50, P80, and P90 confidence levels — not a single deterministic number. Definitions of these terms are maintained in the PMMilestone PM Glossary.

History

Formal uncertainty analysis entered project controls through the U.S. Department of Energy's cost-estimating practice in the 1990s and was codified in AACE International Recommended Practice 40R-08 (Contingency Estimating — General Principles) and 57R-09 (Integrated Cost and Schedule Risk Analysis). The U.K. HM Treasury Green Book and the Australian Department of Infrastructure's P50/P90 guidelines extended the discipline into public-sector project assurance. The dedicated risk-analysis modules in the Project Controls Academy walk through both recommended practices.

Principles

• Separate base estimate uncertainty from discrete risk events — never double-count.
• Use three-point estimates (optimistic, most likely, pessimistic) anchored in reference-class data, not gut feel.
• Correlate variables that move together (e.g. steel price and steel-erection productivity) — ignoring correlation under-states the tail.
• Report the full S-curve, not just the P50 — owners need to see the P80 for contingency sizing.
• Refresh at every stage gate; uncertainty narrows as engineering matures.

Real-World Construction Example

An offshore wind developer ran an integrated cost-and-schedule uncertainty analysis at FID. The base estimate stood at EUR 2.1 billion with a 36-month schedule. The Monte Carlo simulation, drawing on a reference-class dataset of 24 prior offshore wind projects, produced a P50 of EUR 2.28 billion and a P80 of EUR 2.47 billion. The board sanctioned at P80 with EUR 370 million contingency. Two years later the project landed at EUR 2.41 billion — inside the contingency envelope. The discipline of analysing rather than negotiating contingency paid off twice: at sanction, and at outturn.

IT / Agile Example

A national health-records migration program treated its 36-month timeline as a point estimate at first. After the first integration slipped six weeks, the PMO ran an uncertainty analysis using sprint-velocity distributions from analogous historical programs. The P50 dropped from 36 to 41 months and P80 to 47 months. The board added a six-month buffer and a contingent funding tranche of GBP 28 million. When two further integration surprises emerged later, the program absorbed them without breaching its envelope.

Project Controls Perspective

Uncertainty analysis is where the controls function earns its credibility with the board. It demands disciplined use of reference-class forecasting, transparent assumptions, and integration of cost and schedule. The controls team owns the model, the data, and the documentation; the project director owns the decisions taken on the back of it. Cross-validate with the EVM Calculator and the Schedule Health Checker.

Practical Lessons Learned

• Reference-class data beats expert judgement every time. Without it, ranges drift to the optimistic.
• Correlation is the silent killer of bad analyses — assume zero correlation and you understate the tail by 20–40 percent.
• The output of the model is a conversation, not an answer. The board needs to understand why the P80 is what it is.
• Static analyses age within months. Build the model to refresh in hours, not weeks.

Common Mistakes

• Treating uncertainty analysis as a one-off exercise at sanction instead of a living model.
• Using symmetric ranges when project data clearly shows a long right tail.
• Double-counting risks already in the base estimate by adding them again in the risk register.
• Mistaking precision for accuracy — a four-decimal-place P80 from a thinly-sourced model is still a guess.

Expert Tips

• Capture identified events on the Risk Register Template and link each one to a cost and schedule impact node.
• Tornado charts beat S-curves for executive audiences. They show which assumptions matter, not just the spread.
• Have the cost lead and the schedule lead build the model together. Cost-only or schedule-only analyses are systematically optimistic.

Key Takeaways

• Uncertainty analysis converts a single number into a range with confidence levels.
• P80 is the convention for sanctioned capital projects; P50 is for internal stretch.
• Aleatory and epistemic uncertainty are different — model them differently.
• The discipline is only credible if it is refreshed at every gate and reconciled to outturn.

Further Reading

Detailed treatment is given in the risk and contingency titles listed in PMMilestone Books & Publications.