CIO Sponsorship Is Not Executive Ownership
I have yet to come across an organisation genuinely treating AI as operating-model change. But why? Before leaders can decide how AI will alter an operating model, they need a truthful view of the one they already have. Do they genuinely know how the organisation works today?
An operating model is often reduced to boxes on an organisation chart. That is far too narrow. The chart shows who reports to whom. A process map shows how work is supposed to move. Neither necessarily reveals how decisions are really made, where judgement sits, which workarounds keep the process moving or who absorbs the consequences when something changes.
For me, an operating model is a living and breathing network. It includes decision rights, incentives, measures, handoffs, data flows, informal workarounds, tacit judgement and the distribution of power. Change one component and the effects cascade through the rest of the network.
AI exposes this because it rarely stays within the boundary of the tool being introduced. It changes how information moves, where decisions are made, what work people perform, which exceptions matter and how value is created or lost. Yet AI plans can still be built as though the technology can be inserted into an otherwise stable organisation.
This is the underlying problem I see in much of the planning for 2027. AI ambition, technology ownership and expected benefits are spreading through the business, while cost, risk and accountability tend to collect around the CIO. The CIO can make those tensions visible, but visibility is not ownership. An executive team cannot delegate its choices about funding, workforce, risk and operating responsibility simply by placing the CIO at the centre of the plan.
A local decision becomes an enterprise trade-off
Consider a business function adopting an AI assistant to reduce the time spent preparing case responses. The local decision is perfectly rational: the work is repetitive, the tool is available and the potential time saving is credible.
The deployment then needs access to operational data, identity controls, integration, monitoring, support and an approach to handling exceptions. Some of those responsibilities sit outside the function that expects the benefit. The tool reduces drafting time, but the surrounding roles, targets, approval stages and capacity assumptions remain unchanged. Time is saved at task level, although no corresponding budget saving appears. Meanwhile, additional platform, assurance and support costs accumulate elsewhere.
Finance then questions the value. The business function points to faster work. IT points to the growing cost and risk burden. Neither is necessarily wrong; they are measuring different parts of the same system.
What began as a local productivity decision has become an enterprise trade-off, and the CIO is left holding the contradiction.
That cascade is more important than any single technology choice. It shows why apparently separate questions about budgets, ownership, cost and workforce keep converging around the CIO. Ambition and expected benefit can remain distributed, while the technical and risk consequences become concentrated.
Why these tensions keep landing on the CIO
Gartner’s analysis of CIO planning for 2027 gives a useful public expression of this pattern. It reports pressures across budget growth, distributed technology ownership, lifecycle cost and workforce expectations, and places the CIO at the centre as the “reconciling force”.
I agree with the tensions more than I agree with treating their reconciliation as a CIO responsibility. It is understandable that they become most visible from the CIO’s seat: technical dependencies cross functional boundaries, ongoing costs accumulate and risks rarely remain local. However, seeing the whole contradiction does not give the CIO the authority to decide every trade-off within it.
Released capacity is an executive choice
That cascade turns on a question that is often left unanswered: what is the released capacity for?
In the second half of a six-month field experiment reported by the American Economic Association, across 66 firms and 7,137 knowledge workers, the 80% of treated workers who used the generative-AI tool spent two fewer hours each week on email and reduced time working outside regular hours. The researchers did not detect changes in the quantity or composition of their tasks.
The experiment does not tell us what each firm should have done with that time. It exposes the decision that AI cannot make on behalf of its leaders. Should the capacity reduce work outside normal hours, improve quality, absorb more demand, remove cost or change the shape of a role? Each answer has different consequences for targets, staffing, measures and the benefit case.
If nobody chooses, the function can point to time saved while Finance looks for a saving that was never designed to appear. What looks like a disagreement about AI’s value is partly an unresolved decision about the work.
The economics literature offers a related caution. In their account of the Productivity J-Curve, Brynjolfsson, Rock and Syverson argue that general-purpose technologies such as AI require substantial complementary investment, much of it intangible and poorly measured. The implication for an organisation is that a technology business case can understate the work needed around processes, capabilities and organisational design.
That should not be read as reassurance that weak value will inevitably improve with time. It may not. It means the complementary change needs to be chosen, funded and owned rather than treated as something the technology will produce by itself.
The CIO’s role is to make the system legible
The CIO has a central role in resolving this, but central does not mean singular.
The CIO should make the relevant operating model visible enough for the executive team to decide. That means exposing dependencies, technical constraints, lifecycle costs, data conditions, security implications and the consequences of different choices. It means showing where a local decision will create work, cost or risk elsewhere. It may also mean establishing the technical conditions through which decentralised innovation can happen safely.
Decentralisation is not inherently a failure. Business teams are often closer to the work, the customer problem and the operational context. Allowing them to shape and sometimes build AI solutions can improve relevance and speed. The problem appears when authority is decentralised but the consequences are not: one part of the organisation chooses, another pays, and a third remains answerable when something fails.
Nor does executive ownership mean that every decision belongs to a committee. Collective ownership without named owners quickly becomes ambiguity. A CIO may legitimately own an enterprise platform, an information-security outcome or a technical standard. A business leader may own the operating outcome and realised value. A people leader may own workforce redesign. The executive team owns the trade-offs between them and must make sure the named accountabilities form a coherent whole.
Before choosing the change, the executive team also needs a shared view of where the present model is strong, where it is weak, where the AI opportunity sits and how much disruption it is prepared to own. Without that baseline, leaders can agree on the ambition while holding incompatible assumptions about what it will take.
This distinction is reflected in public governance guidance. The NIST AI Risk Management Framework calls for clear roles and lines of communication while stating that executive leadership takes responsibility for decisions about AI development and deployment risks. Gartner has also argued separately for shared CHRO-CIO ownership where work, talent and technology overlap.
Neither source proves that every AI decision belongs to the full executive team. Taken together, I think they support a more defensible principle: accountability for enterprise consequences should not default to the CIO simply because AI has a technical component.
Where the executive team does not share a view of the current operating model, the CIO should not silently reconcile the differences. The job is to surface the competing assumptions, show their consequences and require the people who own the business to choose between them.
Visibility does not require paralysis
There is an obvious risk in this argument. If an organisation must fully understand its operating model before changing it, discovery can become endless and experimentation can stop.
That is not what I am suggesting.
Few organisations could produce a complete and uncontested picture of how they work. The useful question is whether they understand the part of the operating model that a particular AI decision will affect. The depth of discovery should be proportionate to the scale, reversibility and potential impact of the decision.
A bounded experiment can continue while that understanding develops. It should have a named business outcome and owner; limited scope, data, authority and duration; measures that include downstream effort and cost rather than task speed alone; and explicit conditions for stopping, revising or expanding it.
The point of the experiment is not merely to prove that the technology works. It is to expose assumptions about the organisation.
This is also why learning and scaling need different approval standards. A limited trial involving a small group and reversible workflow change should not face the same gate as an enterprise deployment that changes roles, reallocates budgets or places important decisions behind an automated system. The UK Government AI Playbook calls for a named senior responsible owner for a specific AI project and clear responsibilities across the AI lifecycle. Its separate Digital, Data and Technology Playbook says testing should be proportionate to the size, complexity and uncertainty involved, with measurable objectives, defined scope, clear timescales and time to consider results before scaling.
Organisations should keep learning. What they should pause are irreversible or high-impact commitments whose operating consequences have not been understood or owned.
The five executive decisions hidden inside the AI plan
Once this is understood as an operating-model problem rather than a collection of technology problems, five executive decisions emerge.
What part of today’s operating model will this alter? Make the relevant current state visible by mapping how the affected work actually happens, including decision rights, handoffs, information flows, exceptions, measures and informal judgement. Do not assume the documented process is the real process.
Where do we disagree? Ask each executive to state what they believe will change, where value will appear, which costs will increase, what risk is acceptable and what will happen to any capacity released. Differences should become explicit before they become delivery problems.
Who owns each outcome, and which trade-offs require a collective choice? Name the accountable owner for each specific outcome, then record the decisions that must be made together. Shared ownership should connect accountabilities, not dilute them.
What can we learn reversibly? Use bounded experiments to test both the technology and the operating assumptions around it. Measure local performance alongside downstream workload, lifecycle cost, quality, risk and adoption.
What must be understood and accepted before scale? The answer does not need to remove uncertainty. It does need to identify which roles, budgets, controls, measures and behaviours will change, and which named leaders accept responsibility for those consequences.
The 2027 planning question is therefore not simply whether the CIO can reconcile AI ambition with IT reality. It is whether the executive team has mistaken CIO sponsorship for executive ownership.
Before approving the AI plan, do we genuinely share a view of how the organisation works today, and which trade-offs are we collectively prepared to own?
Comments