A manufacturer wants to "use GenAI to reduce machine downtime". What questions do you ask and what do you propose?

Strong answer

Downtime has two levers: fewer failures and faster repair. Predicting failures is a time-series problem on sensor data; it needs instrumented machines and failure history, and GenAI adds little. Faster repair is a knowledge problem: technicians searching manuals, past work orders and tribal knowledge. That is where retrieval and generation help. Ask: what is mean time to repair today, how much of it is diagnosis versus waiting for parts, what documentation exists and in what languages, and can technicians use a device on the floor (gloves, noise, connectivity). Propose a technician assistant over manuals and historical work orders, answering with citations and refusing when no source covers a safety-critical step. Measure MTTR on a pilot line versus a comparable line. Mention predictive maintenance as a separate, longer track if sensor data exists.

Weak-answer indicators

• Promises GenAI will predict failures
• Ignores floor conditions such as gloves, noise and offline use
• No safety refusal behaviour