Skip to content

Outside View / Reference Class Forecasting

Why It Matters

People chronically overweight the details of their specific case while ignoring the broader pattern. "Our startup is different" - but 90% of startups fail regardless of how different they feel. The outside view corrects this by forcing you to start with base rates and then adjust based on what makes your case genuinely unusual.

The Key Move

  1. Identify the reference class - what category of situations does this belong to?
  2. Determine the base rate - what happened in similar cases historically?
  3. Start your estimate at the base rate, not at your intuition.
  4. Adjust only based on factors that are demonstrably different from the reference class and have been shown to affect outcomes. The discipline is resisting the temptation to treat every case as unique when the base rate is highly predictive.

Classic Application

Intelligence estimates of political instability. Analysts naturally focus on country-specific factors (leadership personalities, ethnic tensions, economic conditions). Reference class forecasting starts with: "What percentage of countries with these demographic and economic indicators experienced regime change in the next 5 years?" Then adjusts based on genuine outliers. This massively improves accuracy over pure inside view analysis.

Surprising Application

Major project planning. IT projects routinely overrun timelines and budgets. Rather than estimating based on the specific project, reference class forecasting asks: "What percentage of similar projects (same technology, team size, complexity) finished on time?" The base rate is typically ~30%. Starting there and adjusting beats bottom-up estimation by large margins.

Failure Modes

Choosing a reference class that's too broad (comparing your case to "all businesses" rather than "B2B SaaS companies with similar characteristics"). Adjusting away from base rate without evidence (believing you're special). Ignoring base rates entirely (defeats the purpose). Using base rates when you have specific, causally relevant information that's more diagnostic (sometimes inside view is appropriate). Not updating base rates as new data arrives.

Go Deeper

Kahneman and Tversky, "Intuitive Prediction: Biases and Corrective Procedures" on inside/outside view. Flyvbjerg, "From Nobel Prize to Project Management: Getting Risks Right." Tetlock, "Superforecasting" Chapter 7.