Why it matters: a buyer needs an adviser whose method can produce a no-build result. Ask this before discussing a favourite technique.
A strong answer should cover: baseline optimisation already tried; the remaining failure class; whether it is behavioural, informational or operational; cost of the smallest useful experiment; and a stop rule. Request an anonymised counterexample where simpler work was preferable, with permission to describe it.
Follow-up: “Which observation would reverse your current recommendation?” Bring three sanitised errors and the metric dictionary, not a vague request to make the model better. Error-analysis practice is the grounding reference.
Interview status: Not yet interviewed.