Agentic AI · solution
Procurement agents for manufacturing
A procurement agent decides when to raise a purchase request and at what quantity, using live consumption data from the MES against a stocking policy your supply-chain team owns. The goal is continuity without overstocking, and every request is fully traceable to the demand signal that produced it.
The decision this agent takes
For each managed SKU, the agent watches daily consumption against on-hand inventory and open purchase orders. When projected coverage falls below the policy value, the agent decides: raise a purchase request in the ERP for the economic-order quantity, against the preferred vendor for that SKU, with a delivery window aligned to the production plan.
Data the agent needs
- Live consumption from the MES (BOM explosion + actual production).
- On-hand inventory and open POs from the ERP.
- Vendor lead times and preferred-vendor mapping from the SRM.
- Stocking policy per SKU: safety stock, EOQ, service-level target, owned by the supply-chain planner.
- Historical shortage and expedite events for policy tuning.
What it does unsupervised
- Raise purchase requests in the ERP, subject to a per-request cost cap set by procurement.
- Bundle requests for the same vendor into a single PO where the vendor supports it.
- Notify the buyer when a vendor’s lead-time trend degrades beyond the policy tolerance.
- Reconcile receipts against POs and flag mismatches.
Where a human stays in the loop
- Vendor selection for a new SKU is human-approved, not agent-picked.
- Requests above the per-request cost cap route to procurement approval.
- Any single-source spend decision is human-approved; the agent surfaces alternatives.
- Quarterly stocking-policy review by supply chain.
Where this agent gets it wrong
Honest failure modes and the data-quality prerequisites we insist on before an agent goes live. If your plant does not meet these, the agent runs in advisory-only mode until it does.
- If the BOM in the MES is stale, consumption projections are wrong and the agent will over- or under-request. Data-quality prereq: BOM change-control process with a named owner.
- Vendor master data errors (wrong lead time, wrong MOQ) propagate directly into agent decisions. The agent cannot detect this on its own.
- Bill-of-materials variants and phantom assemblies confuse the projection if not modelled explicitly; the agent falls back to historical consumption which is less accurate.
- The agent optimises for continuity within its SKU list. Cross-SKU trade-offs (e.g. a common material shortage) still require a human planner.
How Ajinkya ships this today
Ajinkya’s existing SAP integration service line already pushes and pulls consumption, PO and inventory data at reference plants. A productionised out-of-the-box procurement agent that closes the loop into PR generation is [NOT YET SHIPPED] and is scoped per client.
FAQ
What ERPs does it work with?
SAP S/4HANA and SAP ECC are the reference stacks. Other major ERPs (Oracle, Dynamics, Odoo) are supported provided they expose the required master data via API or IDoc.
What is the smallest workable pilot?
One product family with 40-80 active SKUs, 3 preferred vendors, and 12+ months of consumption history.
Does the agent negotiate with vendors?
No. Agent scope stops at the PR. Vendor negotiation and rate revisions stay with procurement.
How is dual sourcing handled?
Where a dual-source policy exists, the agent respects the split percentage set by procurement. The agent does not decide sourcing strategy.
Talk to the engineers who ship this
A 30-minute discovery call gets you a scoped answer on whether this agent fits your plant, what data your team would need to prepare, and what a pilot would look like.
Book a 30-min callReviewed by Amey Kadle, Founder, Ajinkya Technologies. Last reviewed: 2026-08-29.