Procurement agents the team actually uses.

Spend questions from management waited on a request to the data team, then a few days of pivots before an answer landed. POs walked through the ERP one screen at a time. Order status got chased by email, then re-keyed into the system. MRO requests piled into inboxes, each one requiring a buyer to look up the right SKU, the right supplier, the right history.

The category knowledge that made any of this work — what supplier, what spec, what has been ordered before, who approves at what threshold — lived in the experienced buyers’ heads. Hard to share. Easy to lose when someone moved on.

The team was good at the work. The team was the integration layer between the ERP, the inboxes, the spend data, and the people asking questions.

We built an agentic procurement system around the work the team actually does, not a generic chatbot bolted to the side. Two roles live inside it: an analyst with structured access to ERP and spend data, and a buyer authorised to take actions inside the ERP and on the team’s inboxes.

The analyst answers spend questions in minutes — by category, by supplier, by quarter, by anomaly. Management asks, gets a grounded answer with a reference back to the source data, and makes the decision without queuing the data team.

The buyer creates POs, monitors orders, follows up on exceptions, triages procurement-related email, and handles MRO requests end to end. Category knowledge accumulates inside the system: what has been ordered, what passed quality, which supplier was used for the last version of the part. The agent proposes items for orders; a manager approves; the next request is smarter for it. The buyers stopped spending the day on routine and moved up the stack to contract renewals, sourcing, and supplier relationships.