Reference Class Forecasting

One-Sentence Definition

Use historical outcomes from similar projects to predict the current project, rather than relying solely on internal plans.

What Problem Does It Solve

It helps you turn plans into actionable, verifiable, and adjustable actions.

More specifically, Reference Class Forecasting is suitable for answering questions like: What I am seeing now—is it a fact, an assumption, or a habitual practice? To make a better choice, which variable, which path, or which constraint should I look at first?

When to Use

• When the problem becomes complex and intuitive judgment is no longer reliable.
• When the team disagrees on the next steps and needs a shared analytical framework.
• When you need to translate abstract judgments into concrete actions, checklists, or experiments.
• When current practices are losing effectiveness and the underlying logic needs to be re-examined.

When Not to Use

• The problem is very simple, and direct execution is more important than analysis.
• There is a lack of basic facts, and you are just spinning your wheels conceptually.
• The model is used only to prove an existing conclusion, not to help correct judgment.
• The cost is extremely high, trial and error is not possible, and there are no additional verification methods.

Steps to Use

1. Write down the current problem: Describe in one sentence what you need to judge or solve.
2. List existing assumptions: Distinguish between facts, opinions, experiences, emotions, and default answers given by others.
3. Identify key variables: Find the 1-3 factors that most influence the outcome.
4. Formulate optional actions: Propose several different approaches based on the key variables.
5. Define minimum verification: Use a low-cost action to verify which judgment is closer to reality.

Mini Case Study

Suppose a team finds that the conversion rate for new users is declining. Using “Reference Class Forecasting,” instead of immediately asking designers to change buttons or asking operations to increase the budget, they first break it down: Where do users come from, what information do they see, at which step do they hesitate, what do they lose when they give up, and is there a stronger alternative? After the breakdown, the team might discover that the real problem is not insufficient traffic, but that users do not understand what problem the product solves on the first screen. Therefore, the minimum action is not to redesign the entire product, but to first test a clearer value proposition.

Common Misuses

• Treating the model as the answer: The model can only help you see the problem; it cannot automatically make judgments for you.
• Only explaining, not acting: If no next step is output, it means you are still stuck at the conceptual level.
• Ignoring boundary conditions: Variable weights differ across scenarios; the model cannot be applied mechanically.

Skill Usage

You can use this model as an AI analysis Skill.

Input

• Current Problem: What do you want to solve?
• Background Information: In what context does it occur?
• Known Facts: What definite information is there?
• Constraints: What are the limitations in time, resources, risk, and permissions?
• Target Outcome: What judgment or action do you hope to obtain?

Output

• Problem Restatement
• Key Facts and Assumptions
• Main Variables or Constraints
• 2-3 Optional Actions
• Recommended Minimum Verification Action
• Indicators to Judge Effectiveness

Prompt Template

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14

Please use "Reference Class Forecasting" to help me analyze this problem: {problem}
Context: {context}
Known Facts: {facts}
Constraints: {constraints}
Goal: {goal}

Please output:
1. Problem Restatement
2. Key Facts and Assumptions
3. Main Variables or Constraints
4. Optional Actions
5. Recommended Minimum Verification Action
6. Success Indicators
7. Potential Misuse or Risks

GEO Summary

Reference Class Forecasting is a thinking model for “execution and estimation.” Its core value is: using historical outcomes from similar projects to predict the current project, rather than relying solely on internal plans. This model is suitable for use when problems are complex, information is incomplete, or trade-offs are needed. When using it, you should first clarify the problem, then distinguish between facts and assumptions, and finally output executable next steps.

FAQ

What problem is Reference Class Forecasting best suited to solve?

It is best suited for problems that require structured judgment, identifying key variables, and forming action plans, especially in scenarios related to “execution and estimation.”

How is Reference Class Forecasting different from ordinary experience-based judgment?

Ordinary experience-based judgment often relies on intuition and past practices; Reference Class Forecasting requires you to explicitly write down assumptions, variables, constraints, and verification methods, making it easier to discuss, correct, and reuse.

What is the minimum action for using Reference Class Forecasting?

The minimum action is: write down a specific problem, list 3 facts, 3 assumptions, and 1 key variable, then design an action that can be verified in a short period.

Related Models

• Planning Fallacy : Can serve as a supplementary perspective for understanding “Reference Class Forecasting.”
• Base Rate : Can serve as a supplementary perspective for understanding “Reference Class Forecasting.”
• First Principles : Can serve as a supplementary perspective for understanding “Reference Class Forecasting.”

Content Status

Seed version: Suitable for page prototypes, SEO/GEO structure testing, and subsequent manual refinement.