EXPOSED
The bulk of the modal purchasing agent's day — issuing POs, comparing quotes, chasing delivery dates, maintaining vendor records, running spend and price-variance reports — is structured screen work that procurement software plus AI already handles, and e-procurement platforms have been eating this tier for two decades before LLMs arrived. What persists is supplier negotiation with real leverage, qualifying new sources under quality and geopolitical risk, resolving supply disruptions, and owning the make-or-buy and total-cost call. There is no license or signature requirement here, so nothing protects the transactional tier; the occupation consolidates into fewer, more strategic category managers.
Mixed — a routine tier and a judgment tier. Requisition-to-PO conversion, three-bid comparisons, expediting calls, and reorder-point maintenance are already executed by Coupa/SAP Ariba workflows and RPA, which pulls the score down to 8; what holds it above the automatable band is the non-scriptable work — sitting across from a sole-source supplier during an allocation shortage, auditing a new contract manufacturer's tooling capability, or deciding whether to qualify a second source in a different country.
Some physical or field component. A 5 reflects the occasional plant walk, trade show, supplier site audit, and receiving-dock inspection of a rejected lot — real but episodic, since the requisition queue, ERP, and the negotiation call all happen at a desk.
No licence, no signature requirement. No state licenses buyers; CPSM and CPSD are resume items, not practice requirements, and the purchase order binds the employer under agency law with the company — not the agent's personal signature — answering for a bad contract, which leaves the score at 2 rather than 0 only because government buyers work under FAR/state procurement codes with audit and debarment consequences.
Some relationship component. An 8 recognizes that a category manager's decade of relationships is what gets your order pulled forward during an allocation or gets a price held through a raw-material spike — but the buyer is interchangeable to the requisitioner internally, and most transactional spend is placed against a catalog or contract where the counterparty cares about the terms, not who typed them.
Meaningful discretion. Make-or-buy calls, dual-sourcing decisions, and accepting a deviation on incoming material carry real dollars and schedule risk, which puts this at 11; it stops short of the top band because approval thresholds, engineering-specified specs, and management sign-off on capital and long-term agreements mean the buyer recommends and executes far more often than they unilaterally own the ambiguous call.
Has AI actually changed your work?