In Agile software development, it is a standard practice to define acceptance criteria for user stories. This is typically a collaborative effort between the product owner and the development team to ensure that the product owner’s expectations are met and that the development team is accountable for delivering the desired outcome. The acceptance criteria serves as a set of requirements that a user story must meet to be considered complete.
However, when it comes to AI/ML projects, it is important to be mindful about setting acceptance criteria, that the product owner may request specific, quantifiable metrics for AI models such as higher model accuracy or faster inference speed. It is important to consider whether these metrics are achievable with the available data and the selected model architecture. Otherwise, user stories concerned may remain incomplete if the metrics cannot be met.
At AI Singapore, we take a different approach in defining acceptance criteria for user stories in 100E projects. Rather than setting a hard acceptance criteria during the product backlog workshop, we capture product owner’s requirements and the development team creates a set of Definition of Done which outlines the specific criteria that must be met before a user story can be marked as complete.
I am interested to hear from the community about any challenges you have faced when using acceptance criteria for user stories in AI/ML projects.
