Outside-view scenario forecast prompt

Written by skills.andrewb-c.com, not by the researchers cited. It puts into prompt form the “outside view” from Kahneman and Lovallo (“Timid Choices and Bold Forecasts”, 1993), reference class forecasting as Flyvbjerg developed it for project planning, and the base-rate-first habit Tetlock and Gardner describe in Superforecasting (2015).

The prompt

You are forecasting: {what will happen / what number}, by {date}.
Context: {facts, data, plan}

Work in this order and label each step.

1. Reference class. Name 2–3 classes of similar past cases (e.g. "Series B SaaS
   companies raising prices", "infrastructure projects over $1B"). For each, give
   the base rate or typical distribution and say where it comes from. If you don't
   know a figure, say so — don't invent one.
2. Pick the anchor. Choose the most comparable class and state its base rate as the
   starting estimate.
3. Adjust. List the specific ways this case differs from the class. For each, give
   direction and rough size of the adjustment, and why. Keep total adjustment modest
   unless the evidence is strong.
4. Scenarios. Give base, upside and downside cases: the key assumptions, the
   resulting value or outcome, and a probability for each (they must sum to 100%).
5. Pre-mortem. Assume the forecast turned out badly wrong. Give the two most likely
   reasons.
6. Final answer. A single probability or a range (10th–90th percentile), plus the
   one indicator that would most change your view, and what it would move you to.

How to use it

• Supply real base rates when you have them. The model’s remembered statistics are the weakest part.
• For numeric what-ifs (revenue, costs, timelines), ask it to show the arithmetic behind each scenario so you can check it or move it into a spreadsheet.
• Pair it with a data connector (warehouse, SEC EDGAR, market data) so the context is current.

Limits

Models make arithmetic and data errors and can make up plausible-sounding base rates. Check every figure that matters. Scenario probabilities are judgments, not measurements. Not investment advice.