94. A supplier reports high overall accuracy for a clinical AI tool. What further questions matter?
HardPatient safety
Technical
Show model answer
I would ask what the tool predicts, how errors are defined and whether evaluation reflects our setting. Overall accuracy can conceal differences between relevant groups, uncommon events or conditions unlike those in its development data.
Qualified clinical, statistical and governance specialists should assess harms, monitoring and accountability. A headline metric is not authorization, and my interview answer cannot personally certify a tool as safe across all patients and clinical uses.
What interviewers listen for
- Define the task and errors
- Examine relevant population performance
- Require qualified governance evaluation