Hi {{first_name}} ,
Have you seen this pattern? An AI pilot impresses in a demo, spreads informally, and a few weeks later a supplier assessment or an externally visible message exists that nobody formally authorised. This week's CPO Path is about the leadership move that prevents that moment: governance by decision rights.
That being written - REMINDER:
Don’t miss out your registration to our tomorrow’s meeting - The Procurement AI Agent Prototype in action (start 18:00 CEST / 9:00 am PT, 90 minutes max.)
Registration: Zoom registration.
Executive premise
Most procurement functions are past the question of whether AI will touch their work. In Deloitte's Global CPO survey research, 92% of surveyed CPOs said they plan to invest in generative AI capabilities, while only 37% were piloting or deploying them in procurement. ([Deloitte], 2024) The gap between intention and impact is rarely a model problem. It is usually a governance problem: nobody has decided, in writing, what the system may calculate, what it may propose, and what remains a human decision.
Why this matters
AI adoption is one of the few current topics where a procurement leader is watched simultaneously by the CFO, the CIO, Legal and often the board. How the function introduces these tools is read as a signal of judgment, not of technical taste.
Two findings frame the leadership stakes. Across organisations, McKinsey's State of AI research finds that redesigning workflows has the largest effect on whether generative AI produces bottom-line impact — yet only around a fifth of organisations report having fundamentally redesigned any workflow. ([McKinsey & Company], 2025) The same research finds that senior-executive oversight of AI governance is among the elements most correlated with reported bottom-line impact. ([McKinsey & Company], 2025) Read together: the value does not come from the tool. It comes from a redesigned, governed way of deciding.
The regulatory direction points the same way. Under the EU AI Act, transparency obligations begin to apply from 2 August 2026, while the omnibus agreement of May 2026 deferred the high-risk obligations for stand-alone systems to December 2027. ([European Commission], 2026; [Gibson Dunn], 2026) The dates moved; the expectation of documented human oversight did not.
The real dynamic
The pattern that stalls strong operators is recognisable. A pilot impresses in a demo. It spreads informally. Then a pricing figure, a supplier assessment or an externally visible message is produced by a system nobody formally authorised — and the function spends its credibility explaining, after the fact, who approved what.
The market data suggests this is not a marginal risk. Gartner predicts that over 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. ([Gartner], 2025) Inadequate risk controls is the category a procurement leader can influence directly.
A more useful frame than "How much can we automate?" is a decision-rights question with three layers:
First, what the system calculates. Numbers, flags and KPIs can be produced deterministically and verified. Calculation can be delegated with confidence when it is checkable.
Second, what the system proposes. Recommendations — a supplier status change, a task, a draft message — can be prepared by software, provided the evidence behind the proposal is visible.
Third, what only a human approves. Anything that changes a supplier relationship, commits money, or leaves the company carries a named human owner. Not as a workaround, but as the designed operating model.
In many organisations, the difference between an AI initiative that builds trust and one that quietly erodes it is whether this three-layer split exists on paper before the first tool goes live — and whether every approval is logged well enough to survive an audit question.
What strong leaders do differently
They write the decision-rights map first. One page per workflow: calculated by the system, proposed by the system, approved by a named role, logged where. This document does more for executive trust than any tool demonstration.
They redesign one workflow end-to-end rather than sprinkling AI across many. ([McKinsey & Company], 2025) Spend analysis, inbox triage or supplier lifecycle reviews are natural candidates in procurement: high frequency, measurable, and controllable through approval gates.
They report AI in enterprise language: decisions supported, approval coverage, time from signal to decision, incidents, and auditability — the same categories a CFO applies to any control system.
They treat vendor claims with sourcing discipline. Gartner describes widespread "agent washing" — existing products rebranded as agentic. ([Gartner], 2025) A supplier-qualification mindset applies: evidence over demo.
A practical note: on Wednesday evening (start: 18:00 CEST / 9:00 am PT) I will be demonstrating a working procurement AI setup live, including the approval gates and decision-rights logic described here, applied to spend analysis and supplier lifecycle decisions.
Details and registration: Zoom registration.
this is for our community only as discussed and shaped by your survey participation during the last weeks - so don’t miss out, see you in the meeting :-)
P.S.: The recording will be shared exclusively with registered participants.
CPO Path Diagnostic
Question | Fully true | Partly true | Not true |
|---|---|---|---|
For each AI-supported workflow, we can show in writing what is calculated, what is proposed, and what a human approves. | ☐ | ☐ | ☐ |
No AI-generated output reaches a supplier, customer or ERP system without a named human approval. | ☐ | ☐ | ☐ |
We have redesigned at least one procurement workflow end-to-end around these decision rights. | ☐ | ☐ | ☐ |
Every consequential approval is logged and could be reconstructed in an audit. | ☐ | ☐ | ☐ |
We can brief the CFO or board on our AI governance in their language, in one page. | ☐ | ☐ | ☐ |
One-line verdict
AI earns a durable place in procurement when every consequential action still has a named human owner — by design, not by exception.
Hey {{first_name}}, how is your path going?
I read every single message, and past posts already shaped this space. Reach out to me and tell me what you’re up to and the path you’ve chosen and please share your thoughts about our journey on you preferred platform. CPO Path is free, help me keep it that way.
Talk soon,
Pascal
Did this resonate with you?
Quick wins to implement today (digital products I use and offer to you):
SOURCES
Deloitte. (2024). CPOs Steering GenAI in Procurement Through Uncharted Waters — 2024 Global CPO GenAI Survey. Retrieved from https://www.deloitte.com/us/en/services/consulting/blogs/business-operations-room/generative-ai-in-procurement-cpo-survey.html
European Commission, AI Act Service Desk. (2026). Timeline for the Implementation of the EU AI Act. Retrieved from https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act
Gartner. (2025, June 25). Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027. Retrieved from https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027
Gibson Dunn. (2026). EU AI Act Omnibus Agreement — Postponed High-Risk Deadlines and Other Key Changes. Retrieved from https://www.gibsondunn.com/eu-ai-act-omnibus-agreement-postponed-high-risk-deadlines-and-other-key-changes/
McKinsey & Company. (2025, March). The State of AI: How Organizations Are Rewiring to Capture Value. Retrieved from https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-how-organizations-are-rewiring-to-capture-value
Thank you for reading,
Pascal Hecker | Editor-In-Chief, CPO Path.