The GPO Is Now the Most Important Role in Your Agentic AI Transformation

“It is not the IT and platform team. Absolutely incorrect thinking.”

That was Uli Erxleben’s reaction, Hypatos' CEO, when a live poll at a recent SSON webinar revealed that most attendees believed IT owned Agentic AI in their organisation. It’s a common answer, and according to Uli and Hans Schut of KPMG Netherlands, it’s one of the central reasons so many Agentic AI projects fail to scale.

Who Should Own Agentic AI?

Ideally, the Global Process Owner should own Agentic AI. Agentic AI doesn’t run on code; it runs on work instructions: natural language descriptions of how your processes should be carried out. These instructions tell the agent how to handle a non-PO invoice, what tax treatment applies to a cross-border goods movement, and when to escalate an exception. That knowledge doesn’t live in an IT system. It lives with your finance and GBS teams.

And critically, those instructions must reflect not how processes are done today, but how they should be done. As Uli put it: “You need to teach this to the AI — not how it’s done today, but how it should be done.” IT can build the platform. Only the business can own the process.

The GPO’s Moment Has Arrived

Owning those work instructions, i.e., writing them, maintaining them, and improving them, is the job of the Global Process Owner. But not the GPO as the role has traditionally been defined. In many organisations, the GPO is a title without teeth: local business units retain the real authority to do things their own way.

Agentic AI makes that model unworkable. An AI agent executes its instructions exactly as written. There is no creative workaround, no informal escalation, no human common sense to paper over a gap. A flaw in a work instruction doesn’t produce a single mistake — it propagates through every transaction the agent touches.

The GPO must therefore have real authority: the mandate to define how processes are standardised across business units, backed by C-level support when local entities push back. As Uli said: “This person needs full backing and empowerment by the group C-level. If every company code develops work instructions independently, you end up with something that is neither scalable nor cost-effective.”

Hans drew on the lessons of RPA: “Time and again, the conversation comes back to the same point — the business team is responsible for the output. The same logic applies here, at much higher stakes.”

Three Things That Make Agentic AI Successful

• Empower the GPO with real authority. Process standardisation is the economic prerequisite for scale. The GPO must be able to make binding decisions — or have clear C-level backing to do so.
• Treat work instructions as a compliance asset. Not a one-time configuration exercise, but a living document that is version-controlled, reviewed regularly, and auditable. Every agent decision should be explainable by reference to the work instruction that drove it.
• Align incentives to the new model. Process owners who have historically been rewarded for managing large teams need to be incentivised instead on automation outcomes. As Uli noted: “Define the vision, make a plan, and put KPIs behind it that go into the bonus of those people.”

Gartner predicts 40% of Agentic AI projects will fail by 2027. The technology is rarely the problem. The operating model is, and at the centre of that operating model is the question of who really owns the process. Get the GPO right, and the rest follows.

To hear more insights on "How To Scale Agentic AI Without Failing," check out the full webinar here.