BA ACADEMY / LESSON
Define an AI product requirement
Specify user value, data boundaries, evaluation and human fallback before building an AI feature.
The idea
An AI feature needs the same problem framing as any product, plus explicit uncertainty management. Define intended use, prohibited use, data and privacy boundaries, quality metrics, evaluation datasets, bias and failure handling, monitoring and human escalation. Avoid promising deterministic answers from a probabilistic model.
Example
For a support assistant, define which policy sources it may use, when it must say it is unsure, what is logged, how users reach a person and how answer quality is measured.
Illustrative BA scenario
Fictional banking support scenario: the team tests common and edge-case questions against approved knowledge, reviews incorrect answers with compliance and launches with escalation rather than unsupervised account advice.
Try it yourself
- Write three non-functional requirements for a customer-facing assistant.
- Define two metrics for answer quality and one for user impact.