Almost every large production manager asks this question every week.
The answer is complex. A bill of materials has five levels, with components shared across products. One second-level component appears in three finished products. Supplier A runs five or six days late for one material category; Supplier B is punctual but has minimum quantities mismatched to consumption. Others are involved too.
Add some materials in stock and others to order, and the challenge grows.
MRP calculates what to order. But from which supplier when several provide the material? Which is cheaper or faster? Is speed or cost more important? It cannot predict unreliable suppliers or seasonal anomalies.
We struggled with this for years, achieving only partial success. Purchasing teams deserve credit for managing additional records and mental calculations. A bad day can delay a response; people notice the 1% failure, not the 99% that went smoothly.
Everyone talks about AI agents, but AI-powered ERP requires strong foundations; otherwise it is empty hype.
An AI agent grounded in transaction data and intelligently integrated with ERP can support people. Human involvement remains essential, but the work becomes easier.
A layer above solid MRP learns supplier behaviour, seasonal consumption and internal reporting delays, adjusting proposals before they become purchase orders.
Less downtime, fewer urgent purchases, less stock sitting for 90 days and fewer of those 1% incidents.
We are developing this with our academic partner FERIT.