Daniel Barnes · Updated 11 October 2026
AI in procurement uses software models to interpret information, draft work and, when authorised, act within a buying process. The useful starting point is a specific task, the evidence it needs and the decisions that remain with people.
Define the task, test exceptions and record critical control failures before expanding an agent’s authority.
A fictional £10,000 approval threshold needs tests below, at and above the boundary. If an agent bypasses Finance once, record a critical failure even when its other tests pass. The linked template shows how to retain the evidence.
This is editorial guidance, not a claim of independently measured performance. Worked examples are labelled. Vendor announcements and survey findings remain attributed to their sources.
What a 55% rubric score can and cannot tell a procurement buyer.
Check the buying route, scope and availability of the solution extension.
A named pilot and the evidence buyers should ask to see.
Separate survey findings from the controls your own organisation needs.
WOP combines Daniel Barnes’ practitioner experience, sourced reporting and original working templates. Commercial relationships are disclosed on relevant articles. Publication and update dates describe editorial work, not product availability.