Requirement Specifications
Project phase: Business and Data Understanding
You turn the client's wishes into requirements the team can build and test against: functional requirements distinguished from quality requirements such as performance, privacy, security, and maintainability, each expressed in verifiable terms and prioritised explicitly, for example with MoSCoW. You include the requirements specific to a machine learning system — minimum acceptable model performance, the data the system needs, and fallback behaviour when the model is uncertain — traced back to the stakeholders and problem analysis they came from, with what's deliberately out of scope recorded.
Starting Points
Key Points
- Every requirement is written in verifiable terms (you can actually test if it is met).
- There is a clear distinction between functional requirements and quality requirements (performance, security, privacy, etc.).
- Machine Learning specifics are included (e.g., minimum accuracy, data requirements, fallback behavior).
- Every requirement is prioritized using a method like MoSCoW.
- Requirements are traced back to specific stakeholders and the problem analysis.
- The "Out of Scope" section clearly defines the boundaries of the project.
- Each requirement has an assigned owner and a way to measure success.
- The document is a standalone Markdown file, stored in the project's GitLab repository.