Prompt Alchemy: Building a Decision Stress-Test Flow with AI via "Multi-Dimensional Role Simulation"

When facing major decisions (such as career transitions, product pricing, or strategic adjustments), the most common trap we fall into is "Confirmation Bias"—subconsciously seeking only evidence that supports our existing views while ignoring potential risks.

Even when using AI, if you simply ask, "Is this plan feasible?", AI often provides a polite but useless affirmative answer due to its built-in sycophancy.

To truly leverage AI for improving decision quality, you need to upgrade it from an "assistant" to a "virtual board of directors." This article shares a workflow for achieving decision stress testing by constructing multi-dimensional role simulations.

Core Logic: From "Single-Point Consultation" to "Multi-Dimensional Adversarial Debate"

The core of this workflow lies in using prompts to force AI to switch between different cognitive dimensions within the same conversation, generating insights through conflict between roles.

Phase 1: Constructing the Virtual Board of Directors

Do not let AI answer from a single identity. Instead, define a set of roles with conflicting interests and cognitive perspectives.

Operation Instruction:

"I will provide you with a [specific decision plan]. Please simultaneously扮演 (play) the following three roles and provide evaluations from each perspective:
1. The Aggressive Growth Hacker: Focuses solely on scalability, speed, and potential explosive growth, with extremely high risk tolerance.
2. The Conservative Risk Officer: Extremely risk-averse, focusing on compliance, stability, and contingency plans for worst-case scenarios.
3. The Objective UX Advocate: Stands entirely from the end-user's perspective, focusing on friction points, emotional value, and actual usability.
Please present their intense debate regarding this plan in a dialogue format."

Purpose: By forcing perspective switching, you drag the dimensions subconsciously ignored into the spotlight.

Phase 2: Guiding Deep-Level Gaming

After the initial role debate, do not rush to conclusions. Instead, guide the roles into a second round of gaming focused on the points of contention.

Operation Instruction:

"Now, please have the [Risk Officer] list three specific failure scenarios in response to the most optimistic predictions made by the [Growth Hacker]; then, have the [UX Advocate] evaluate which of these two perspectives aligns more closely with real user psychology."

Execution Actions:
- Observe which role's arguments make you feel the most uncomfortable—this usually indicates the weakest part of your decision.
- Input your genuine concerns to the AI and let the three roles collaboratively explore solutions.

Purpose: To transform vague intuitions into concrete risk points and mitigation strategies.

Phase 3: Synthesizing the Final Decision Matrix

Finally, require the AI to convert the results of this debate into a quantifiable decision matrix.

Operation Instruction:

"Based on the multi-dimensional discussion above, please generate a decision matrix table for me:
- Rows: Different options/dimensions of the plan.
- Columns: Growth Potential, Risk Level, User Satisfaction, Implementation Cost.
- Each cell should include a score (1-10) and a brief rationale.
Finally, provide comprehensive recommendations and the mandatory 'prerequisites' that must be met."

---

Practical Checklist

• [ ] Reject Single Identity: Avoid broad instructions like "Please analyze this plan."
• [ ] Define Conflicting Personas: Ensure there are natural conflicts of interest between the defined roles (e.g., Growth vs. Risk Control).
• [ ] Capture Discomfort: Pay close attention to arguments that challenge you or make you feel uneasy.
• [ ] Quantify Output: The final result must include a comparison matrix rather than just a summary paragraph.

Gotchas & Precautions

1. Preventing Role Collapse: In long conversations, AI may gradually forget the distinctions between roles and drift toward mediocre, neutral commentary. The solution is to periodically send: "Please remember your current identity is [Role Name], and re-examine the above content using that role's tone."
2. Beware of Pseudo-Logic: AI sometimes fabricates reasons that seem logical but lack factual support to maintain its persona. For key data points, you must require AI to note: "This is a simulated deduction, not factual data."
3. Scope of Application: This workflow is suitable for strategic planning, complex product definition, and high-stakes personal decisions; it is not suitable for simple technical questions or factual queries.

Conclusion

The highest value of AI is not making decisions for you, but helping you walk through all possible paths to failure in advance by simulating different personalities and perspectives. Experiencing a fierce argument in the digital world is far cheaper than enduring a painful failure in the real world.