Prompt · Cognitive tools

Reference-Class Forecasting on a Board Estimate

Kahneman’s outside view: locate the decision in its base-rate class before accepting the inside-view projection.

Version 1.0 Tested with Claude, GPT-4-class Category: Cognitive tools

When to use it

Any time the board is asked to approve an estimate — a project cost, a timeline, a recovery value, a contingent-liability exposure, an M&A synergy, a capital-allocation projection. The inside view that management presents is constructed by reasoning about the specific features of this decision. The outside view, first articulated by Kahneman & Tversky and developed into the reference-class-forecasting discipline by Flyvbjerg, asks the different question: what has been the actual outcome distribution of decisions like this one? Where the two views diverge, the outside view is reliably closer to reality. Reference-class forecasting is particularly valuable because it short-circuits over-confidence, optimism bias, and anchoring in a single exercise.

The prompt

You are serving as a senior board advisor conducting a reference-class forecasting exercise on an estimate the board has been asked to approve. The estimate under consideration is: [STATE THE ESTIMATE IN NEUTRAL, REDACTED TERMS. INCLUDE: what is being estimated (cost, timeline, recovery, liability, synergy, growth, etc.); the specific number or range proposed; the time horizon; the implicit confidence level in management’s presentation.] The strategic context is: [DESCRIBE THE DECISION THE ESTIMATE SUPPORTS, AT STRUCTURAL LEVEL. Redact names, specific numbers, jurisdictions where confidentiality requires.] Do the following: 1. Define the reference class. What is the broader class of decisions, projects, or claims of which this one is an instance? Be specific — not “M&A transactions” but “mid-cap cross-border acquisitions in the same sector, with acquirers of comparable size, pursued under time pressure.” State what makes an instance count as in the class. 2. Establish the base rate. Based on publicly available evidence, what is the distribution of actual outcomes for decisions in this reference class? Give the median, the range from the 25th to 75th percentile, and the worst-decile outcome. Cite the source basis for your estimates. 3. Locate the current estimate on the distribution. Is management’s projection at the median of the reference class? The optimistic quartile? Beyond the best decile of actual outcomes? Be specific about where it sits. 4. Identify the features of this specific decision that would justify departing from the base rate, and the features that would argue for regressing toward it. Be concrete — “the company has done this before successfully” is inside-view reasoning; “this class of decision has consistently come in over budget because of regulatory timelines” is outside-view evidence. 5. Produce an outside-view estimate — a range, with explicit uncertainty, anchored on the reference-class distribution and adjusted only for features whose effect is documented. State it alongside the inside-view estimate for comparison. 6. Finally: identify the three questions the board should ask management about the gap between the two estimates, if there is one. Be specific and concrete. Do not hedge with “it depends.” Avoid generic governance language.

What to supply

Redact as in the other prompts. The exercise needs the structure of the estimate — what is being projected, on what horizon, under what method — more than the precise numbers. Names, proprietary data, and specific figures can be abstracted without loss of value. The reference-class forecasting works because the base rate is usually stable across the specific instance, not because of the specific instance.

What to expect

A structured comparison of the management estimate against the actual distribution of outcomes for comparable decisions. The most common output is a gap: management’s estimate sits in the optimistic tail of the reference-class distribution. The useful move is not to reject the estimate out of hand — sometimes the specific decision’s features justify an optimistic placement — but to force the conversation about which specific features justify the departure, and to require the answer to be concrete and testable.

Before using

Read Privilege and confidentiality before using prompts. Do not paste privileged material; redact names and figures.

Related