I like the idea of predictive analytics, but I do not think it is the first problem. Before an OEE system needs AI, it needs a decent version of reality. That sounds obvious, but this is where a lot of systems get weak.
If I do not know the difference between planned downtime and unplanned downtime, the model is already muddy. If scrap is entered later, or not tied to a product, or not tied to the right time window, then the dashboard may look clean but the operation is still fuzzy. If the only production record lives in SAP, I may know what was reported, but not what happened between start and finish.
The constraint is that operations are messy. People are changing jobs, clearing jams, waiting on material, dealing with maintenance, adjusting equipment, and doing things that do not fit neatly into a dropdown. A system that asks for perfect data will get ignored. A system that asks for useless data will get pencil-whipped. The goal is not maximum data collection. The goal is enough structure to make the operation legible.
What I would want first is a clean event model. Running. Stopped. Planned stop. Unplanned stop. Changeover. No schedule. Starved. Blocked. Faulted. Then counts and scrap tied back to time, product, line, and maybe operator or crew depending on what is appropriate. That is the foundation. AI can sit on top of that later, but it cannot fix a bad definition of downtime.
The surprise is that the boring definitions are usually the strategic work. Everyone wants the dashboard. Fewer people want to argue about what counts as downtime. But that argument is the system. The chart at the end only reflects the definitions underneath it.
Notes for next time: define the event before building the report. Make sure raw material to inventory tracking means something operationally, not just in a database. Do not mistake SAP production records for a full operational model. Let SCADA and MES carry the time-based truth. Then, once the data has shape, predictive analytics might actually have something useful to chew on.