The Inventory Layer Is the Moat
Models cannot order what they cannot see. Visibility first, then the PO.
Everyone wants to add AI to manufacturing. Almost no one can tell you what to order this week.
This is not a criticism of the teams. It's a structural problem. Inventory was never designed to be queried. It was designed to live in a constructor's notebook and a supplier thread from Tuesday.
What makes inventory hard
Three properties make stock uniquely hostile to decisions:
- Schema chaos: every plant counts differently. Two sites in the same group often cannot agree on what "on hand" means.
- Allocation blindness: a part can be on the shelf and still unavailable because it is promised to another job.
- Rare-parts sparsity: the expensive items are the ones nobody tracks until they are late.
Visibility first
Every team we've worked with that tried to add AI before solving inventory eventually rewrote the AI layer. The teams that made stock visible first shipped faster and kept the gains.
If your model recommends a PO based on a stale count, procurement stops trusting it within weeks. Rebuilding trust takes months. Standing up a live inventory view takes weeks.
What the moat looks like
A manufacturer who can see on-hand vs allocated vs unordered, including long-lead parts, can run every ordering decision on top of it. A manufacturer without that layer has to solve it separately for every PO.
The inventory layer is not a project. It's a capability. And like any capability, the earlier you build it, the more compounding returns you get from it.