Reference Class Forecasting in Practice

Abstract

Early forecasting errors can follow a project through delivery. Planning fallacy, optimism bias, and habitual estimating shortcuts can conceal uncertainty in a confident-looking budget.

The original OntaBuild research pairs Reference Class Forecasting (RCF) with scope unpacking. Comparable projects provide context; individual scopes are examined as distributions rather than treated as one settled number.

Across eight completed projects, the original paper reports that preconstruction estimates averaged 0.52% below actual direct costs and that one project experienced an overrun. These are findings from a small historical sample, rather than guarantees for future projects.

Key findings

The paper compares those outcomes with cited industry figures: a 73.66% building-project overrun rate, average budget overruns of 27%, and average schedule overruns of 38%. The source populations and definitions should be considered when interpreting that comparison.

Methodology: RCF + unpacking

1. Build the reference classes

   Combine internal project records and independent references into archetype libraries.

2. Unpack the project

   Separate the project into scopes and examine each independently, making differences and assumptions easier to inspect.

3. Investigate scope variation

   Compare estimator and vendor quotes. A wide spread can signal ambiguous drawings, inconsistent interpretations, or missing scope.

4. Apply contingency deliberately

   The original methodology describes a 3% allocation—1% owner, 1% general contractor, and 1% unknowns—tuned by the scope distributions. It is a description of the studied workflow, not a universal contingency prescription.

Make variance visible

Scope distributions, including box-and-whisker views, make disagreement visible. Wider distributions direct attention toward the drawings and specifications that need clarification before award.

Resolving those questions can narrow the range of plausible costs. The objective is a better-grounded estimate, with uncertainty made explicit.

Results and interpretation

The source reports one overrun associated with a landlord-mandated vendor whose pricing was approximately seven times the comparison price. The method had flagged the anomaly.

The paper also describes an observed relationship between high-variance scopes and later change orders, with further study in progress. That relationship is presented as a research direction, rather than an established causal result.

The original statistical discussion reports a proportions comparison at p<0.001 against the 73.66% baseline from a 186-project dataset. Its own caveat is important: eight projects provide limited power and generalizability. Larger samples are needed to strengthen the evidence.

What this means for project teams

• Use disagreement between estimates as a question to investigate.
• Clarify scope before it becomes a contract or field dispute.
• Make the assumptions behind the forecast available for review.
• Compare estimates with actual outcomes to improve the next reference class.

The practical contribution is a review process that helps teams identify uncertainty earlier and use targeted clarification to improve the basis for commitment.

References from the original paper

• Flyvbjerg & Gardner (2023). How Big Things Get Done.
• Flyvbjerg et al. (2025). The Uniqueness of IT Cost Risk.
• McKinsey (2017). Reinventing Construction.
• Rivera et al. (2016). Identifying the Global Performance of the Construction Industry.
• Sassano (2025). The holistic view in forecasting.

Findings and reference descriptions are carried forward from the original OntaBuild whitepaper. The underlying dataset and cited publications have not been independently revalidated as part of this site restyling.