Practical Guide to AI-Based Requirements Analysis and Specification: Prompt Strategies and Automation Cases

To leverage AI as a strategic partner rather than a simple generation tool, this guide presents specific prompt strategies and automation cases for each stage of requirements analysis and specification. The core principle is not to let AI directly define business goals or core customers, but for users to set the direction while using AI to enhance their thinking.

Requirements Analysis: Setting Direction and Deriving Insights

In the analysis stage, it is recommended to assign critical roles to AI or have it analyze external data to gain insights.

• Utilizing the Devil's Advocate: Assign the persona of a "PM with excellent critical thinking skills" to the AI to point out hidden assumptions, potential risks, and logical leaps in the proposal. This helps correct biased perspectives and discover edge cases in advance.
• Deriving NFRs Based on Competitor Reviews: Analyze actual user reviews from app markets using AI to quantitatively derive non-functional requirements (NFRs) regarding performance, stability, and usability. For example, verifiable criteria such as "image upload response time within 1 second" can be established.
• VC Interviewer Persona: Instead of having the AI directly propose ideas, guide it to ask sharp questions like a venture investor, helping users clarify the essence of the business and execution strategy themselves.

Specification: Maintaining Consistency and Automation

In the specification stage, productivity is increased by automating repetitive documentation and quality management with AI.

• Generating User Stories and Acceptance Criteria: Use the Gherkin format (Given/When/Then) to generate complete drafts of acceptance criteria including normal scenarios, failure scenarios, and edge cases. When input together with policy documents, it produces specifications ready for immediate development.
• Leveraging Existing Assets: Train the AI on past excellent proposals or technical specifications to receive suggestions for the format or structure of new specifications. At this time, it is crucial to clearly recognize how the organization's key assets are utilized.
• Automating Definition of Ready (DoR): AI reviews whether backlog issues are ready for development (e.g., title clarity, compliance with Gherkin format) and automatically generates polite and specific requests for supplementation to the person in charge for unmet items.

Core Principle: Clear Requests and Review

AI outputs must be treated as "drafts" that require human review. Ambiguous requests yield only ambiguous results, so large questions should be broken down, or meta-prompts (e.g., "Ask questions if information is insufficient") should be used to improve input accuracy. AI should be positioned as an auxiliary tool that helps users focus on valuable judgments and decisions.