People don't buy products, they hire them for a job — find out what the job is.
Use when Use it when demographic segmentation has stopped generating insight, when you cannot explain why customers churn or convert, when entering an adjacent market, or when a roadmap has become a queue of the loudest requests with no organising logic.
Avoid when Avoid it for incremental optimisation of a well-understood product — you know the job, and reopening it is procrastination. It is weak for infrastructure and compliance work with no discretionary hiring decision. Done badly it collapses into vague poetry about customer aspirations, and 'the job' becomes whatever the loudest person already wanted, now with better rhetoric.
Understand the person, frame the real problem, build something rough, and learn from their reaction.
Use when Use it on ill-defined problems where the need is genuinely unclear, where the people affected are not you, and where the cost of building the wrong thing is high. It is strongest early, when the framing is still open and cheap to change.
Avoid when Avoid it when the problem is well-specified and the answer is known — it is expensive ceremony for work that needs execution. It struggles with problems whose constraints are technical or regulatory rather than human. And it is the most ritualised framework in common use: workshops, sticky notes and a five-stage poster routinely produce the appearance of the method with none of the substance, because nobody left the building to observe anyone.