byrxj
Working Paper · Operating Models & AI
The Global Process Owner in an Agentic World
As agents absorb the doing, the process owner’s job moves to the three things no agent owns: accountability, ground truth, and judgment.
As agents absorb the doing, the process owner’s job moves to the three things no agent owns: accountability, ground truth, and judgment.
The Global Process Owner has spent two decades standardizing how humans run an end-to-end process and holding the line on cost, cycle time, and service levels. Agents are now taking over the doing. A small human team can already supervise a fleet of specialized agents that runs an entire process.
This does not retire the process owner. It moves the role up. When agents execute, the value migrates to what they cannot own. A named human still has to be accountable for what an agent decides. Someone has to own the machine-readable ground truth the agents run on. And someone has to exercise judgment in the cases agents escalate. The process owner becomes the certifier of agent behavior, not the author of the manual.
The honest version of this story is that the role splits. The part that routes information and enforces process compresses. The part that defines boundaries, owns the ground truth, and makes meaning in the exceptions grows. This paper maps the shift and what the next-generation process owner has to become.
The unit of work is changing under the role. The process owner used to own how people execute. Soon they will own how a fleet of agents executes, and whether it can be trusted.
The signal is loud across the operations world. In SSON’s 2026 industry research, agentic AI is now the top investment priority for Global Business Services, cited by 65% of respondents, well ahead of traditional automation at 40%. Nearly 40% of organizations now run their AI initiatives directly inside the shared-services or GBS unit rather than leaving them to IT. There is a forcing function underneath it: the Hackett Group reports workloads rising faster than budgets, which leaves leaders with a structural shortfall they cannot close by hiring.
McKinsey put the sharpest number on where this goes. A human team of two to five people can already supervise an agent factory of 50 to 100 specialized agents running an end-to-end process such as onboarding a customer or closing the books. Human roles shift toward supervision, exception handling, design, and systemic improvement. Treat the pace as the aggressive end of the range. Treat the direction as settled.
To see what changes, fix what the role has been. The definition is uncontested across the institutions that study it.
SSON defines the Global Process Owner as the person who owns an end-to-end process across functional, geographic, and business-unit boundaries: the glue that keeps a process optimized by cutting through silos. Genpact frames it as concentrating end-to-end accountability in a single person with enough authority to influence resources they do not directly control, charged with driving standardization to deliver lower cost, faster cycle times, and higher satisfaction. Deloitte’s model rests on shared outcome KPIs, a value-stream mandate, and a decision-rights matrix. APQC frames the owner as a single person accountable to senior leadership, empowered to standardize systems, measures, and reporting.
Strip it down and the traditional role is four things: end-to-end accountability, standardization, service-level governance, and decision rights. Hold those four in mind. Each one is about to change shape.
Here is the core claim. Accountability does not transfer to the AI. As agents take the doing, the human who owned the process becomes the human who certifies the agents and answers for them. The role’s value concentrates in three things no agent can own.
Every agent decision still needs a named human owner. The process owner is accountable for the business outcome inside the agent’s authorized scope. The vendor and IT are accountable for the system performing as specified, not for the business decision itself. Accountability shifts from who made the call to who set the boundaries the agent operated within. That is framework accountability, and it lands squarely on the process owner.
Agents do not run on code. They run on work instructions: descriptions of how a process should be carried out, written in language precise enough to execute. Standard procedures are being recompiled into machine-readable agent logic, and the people who must author it are the business process owners, not the platform team. The instruction has to match reality, not the tidy version in the binder. The discipline is to stop documenting what should happen and start capturing what does. Owning that ground truth, and keeping it current, becomes the heart of the job.
Agents handle the volume and escalate the hard cases. The escalation rate is the new edge of the role. What arrives there is contradiction, novelty, and consequence, the work that needs a human who can interpret rather than route. The process owner becomes the named manager every agent needs: the one who sets the guardrails, certifies the agent before it deploys, reviews it on a cadence, and owns the calls the agent cannot make.
A 2026 Berkeley operating-model paper describes autonomous agents that are certified before deployment and reviewed periodically, with guardrail agents blocking high-risk actions in real time. Certification is not a metaphor here. It is becoming a literal step, and someone has to sign it. In a process, that someone is the process owner.
| Dimension | Traditional GPO | Agentic-era GPO |
|---|---|---|
| Owns | The standard operating procedure | The certified, versioned ground truth agents consume |
| Standardizes | How humans execute | How humans and agents execute |
| Measures | Cycle time, cost, SLA | Agent accuracy, autonomous resolution, escalation, drift |
| Role in AI | Gatekeeper of tribal knowledge | Certifier in “AI proposes, human approves” |
| Cadence | Periodic process review | Continuous lifecycle: draft, certified, deprecated |
| Accountable for | The process working | The agents being trustworthy |
This is a bigger job, not a smaller one. The process owner becomes the accountable owner for whether agents in their domain can be trusted, which is a higher-stakes mandate than owning a manual ever was.
Cycle time and SLA measured a human process. They do not capture an autonomous one. The emerging metric stack the process owner now owns:
Context coverage is appearing as an advanced measure but is not yet a standardized KPI. Treat it as a leading-edge instrument, not a settled one.
A balanced view has to hold the disruption case and the growth case at once, because they do not actually contradict. They describe two halves of the same role pulling apart.
The compression is real. Gartner projects that some organizations will use AI to flatten structure and remove a large share of process-routing middle management. Andreessen Horowitz argues the outsourced back-office is being unbundled, because a model can do in-house what a labor-selling vendor used to charge for. The clearest framing comes from independent commentary: the management layer splits, and the people whose job was to route information and enforce process get compressed, while the people who translate, interpret, and absorb contradiction become more important. The job stops being information processing. It becomes meaning-making.
The growth is also real. BCG argues that as agents scale, governance, data integrity, and end-to-end accountability matter more, not less, and that companies which dismantle their process-ownership capability pay for it many times over. Gartner’s own skepticism cuts the other way too: it expects more than 40% of agentic AI projects to be canceled by end of 2027 on cost and weak controls, which says no one is handing whole processes to unsupervised agents soon.
Both camps agree on the split. The transactional, information-routing tier of the role collapses. The governance, accountability, ground-truth, and meaning-making tier grows. The process owner who only routes and enforces is exposed. The one who certifies, curates, and judges becomes more valuable than before.
The research converges tightly here. The next process owner needs domain depth plus AI fluency, the ability to govern an agent fleet, and ownership of a single source of truth. The one skill explicitly devalued is doing the process by hand.
McKinsey frames the new literacy as understanding agent workflows, inputs, and failure modes well enough to evaluate and manage them, not to code them, and says the edge goes to people who combine deep domain expertise with fluency in guiding agentic systems. The Josh Bersin Company puts a number on the People Operations version: HR roles shift from process owners to enablers, with strategic work rising from around 30% to as much as 75%, and recruiters and process managers becoming orchestrators who oversee digital agents. Bersin even sizes the fleet they will manage, up to roughly 130 HR-focused agents across employee, decision, monitoring, action, and rule types. SSON adds that investing in a single source of truth becomes a competitive advantage in its own right.
For People Operations specifically, this is the cleanest career signal in the research. The process owners who run HR are about to become the certifying authority for the company’s people-context, which is the most sensitive and highest-leverage context in the enterprise. That is the move from administrator to owner of the foundation.
The same current is reshaping the teams that author knowledge. Standard-procedure writing and knowledge management are turning into the work of encoding business rules into machine-readable ground truth. Cognizant has committed to deploying a thousand context engineers, on the argument that in this era the lever is context. Anthropic describes the underlying shift as moving from writing individual prompts to curating the information a system runs on. The job is no longer publishing a document a human reads. It is owning a context product an agent consumes, with tests, an owner, and an expiry. The process owner and the knowledge owner are converging on the same work from two directions.
Drawn from 2024 to 2026 sources across institutional bodies, analysts, an academic operating-model paper, a venture firm, an independent skeptic, and vendors. Commercially interested sources are flagged.
Take the GPO-2028 diagnostic: sixteen questions that place you in one of four quadrants and return a quarter's worth of moves.
Take the diagnostic →Email this paper to your inbox, with the illustration and an editable document version attached.