A prompt engineer deployed a voice AI for a healthcare clinic's appointment system, feeding it 15 years of medical documents. The result was disastrous: the AI gave patients lengthy medical lectures instead of booking slots. Callers hung up, causing a 40% drop in traffic within a week. The core issue was context confusion—the AI couldn't distinguish between helpful knowledge and essential scheduling information.

The initial fix—simply restricting medical explanations—failed. The AI became awkward, announcing it wouldn't discuss topics while doing so. The breakthrough came from redefining the AI's role, not just its knowledge. It was programmed as "Sarah," a receptionist whose sole job was scheduling. Medical knowledge was only used for logistics, like matching specialists or assessing urgency, never for diagnosis or explanation.

The transformation was immediate. Calls shortened to under two minutes, abandonment rates fell, and patient feedback improved. The key lesson: role definition matters more than intelligence. For any AI with a vast knowledge base, success depends on ruthless focus. The goal isn't to showcase knowledge, but to be effortlessly helpful—knowing when to talk and when to just get the job done.