Using user roles as a test dimension
AI systems are often tested from a single perspective: the developer’s or tester’s own. Rarely does anyone put themselves in the shoes of the actual user with their specific needs. What gets missed: an IT-shy nurse under time pressure has to manage the same interface as the experienced tester with deep tech know-how working under lab conditions. The defects that show up in daily use only surface in operations.
Persona-based testing brings user roles into test design
The approach adds a dimension to classic test strategies: test cases are also derived from real user roles. A persona describes goals, prior knowledge and typical behaviour of a user group — and produces scenarios a purely functional approach does not generate.
For AI applications this weighs especially heavy. A language model that performs cleanly in the test lab can show completely different weaknesses in the hands of a different user group. When someone with little technical background phrases an input differently from expected and the model returns unusable answers, the system does not report an error. The user still has not reached their goal — and that does not show up in a classic test run, because no one tested from that perspective.
From UX artefact to an active test dimension
In many organisations personas for the applications already exist — as the output of UX workshops, filed and well hidden in Confluence or Miro. Testing rarely sees them. The actual value sits exactly there: personas can steer which failure paths get tested. Where does an occasional user drop off, which permission gaps open up via guest access? Those questions almost write themselves once you think about test scenarios from the role.
For most apps a few personas are enough, derived from analytics and support data to get started. The number matters less than the question of whether they actually feed into sprint planning and test automation as a real target group. If you are currently looking for a way to extend your test strategy with this dimension, that is a typical entry point for structured quality assurance.
Functional correctness is a necessary foundation, but says little about whether an AI system actually works in the daily lives of its users. That is exactly the gap persona-based testing closes.
