Purchase orders should be the control point between demand and spend. In practice, requisitions arrive incomplete, catalogues are ignored, and urgent buys skip the process via email to a preferred vendor. By the time a PO is raised, the commitment already exists — procurement becomes a documentation function after the fact.
Cycle time is the other failure. Approvers lack context on category, budget, and vendor history. Requests bounce for missing specs. Buyers re-enter the same data into ERP that already lived in an email thread. Suppliers chase status because nobody can see where the PO sits in the chain.
Without automation, policy is tribal. Thresholds differ by memory. Preferred-vendor rules are optional. Duplicate POs and split orders evade detection until invoice match fails downstream. Finance inherits the mess at payment time.
Ownership is often unclear. Category managers own strategy; buyers own execution; plant ops own urgency; finance owns coding. When nobody owns the intake standard, every team invents a shortcut. Anti-patterns include auto-approving everything under a round number, cloning last year’s PO without checking price drift, and treating “urgent” as a permanent override code.
AI-assisted PO automation improves completeness at intake, checks policy before routing, accelerates approvals with context, and feeds cleaner documents into invoice matching — so procurement control happens before cash leaves, not in the post-mortem.