Eighteen months in, the board has asked why the AI investment has not moved the numbers. That is a fair question, but it is almost always the second one. The first is whether the AI investment approval that authorised the program was itself examined before the money was released. In most organisations, it was not, and that omission is where the diagnostic case usually begins. It is not where the executive is currently being asked to end it.
In Brief
- When an AI program underperforms, the reflex diagnosis is implementation failure — but the load-bearing question sits one layer up, at the approval decision.
- Most approval papers specify vendor, cost, and timeline but leave the benefit case general and untestable.
- A second investment cycle approved on the same terms produces the same result — and quietly erodes the executive committee's own diagnostic authority.
- An independent review of the approval decision, initiated by the executive before the next paper comes to committee, is the treatment the program should have had at the start.
Blaming implementation stops one layer short
The reflex diagnosis when an AI program has underperformed is that implementation was the problem. The pilots were too narrow. The vendor moved too slowly. The workforce did not adopt fast enough, or the data was not ready. Any of these can be true, and most of them are, at least partly, always true. The trouble is that they only explain the last mile of a decision chain that began much further upstream, with a set of approval choices that specified what the investment was expected to prove, who would decide it had, and against what benchmark. If those choices were thin at approval, no amount of implementation improvement will now produce the answer the board wants.
According to PwC (2026), 56% of CEOs surveyed for the 29th Annual Global CEO Survey report that they realised neither revenue nor cost benefits from AI in the past twelve months. More than half the CEOs surveyed. The scale of the miss is what makes the reflex diagnosis so attractive, because an implementation story is a controllable story. The vendor can be replaced. The pilot can be re-scoped. The training program can be expanded. What tends to sit unexamined is the choice that authorised the investment at the start: the paper the executive committee saw, the assumptions inside it, the benefit case it argued, and whatever mechanism was going to let someone later say whether that case had held.
The Australian version of the reflex diagnosis lands even more precisely. According to PwC Australia (2026), 45% of Australian CEOs identify internal skills and capability as the single largest barrier to AI adoption. The skills-gap framing sits downstream of the real question. The upstream question it obscures is whether the original AI investment approval defined which skills the investment would build, which it would displace, and how the organisation would know either had happened. Where that specification was thin at approval, the skills conversation two years later has no anchor against which to test itself.
The approval process specifies the wrong thing
According to McKinsey & Company (2026), nearly 40% of senior leaders identify redefining process flows as the single most valuable change available to them over the next one to two years, ahead of restructuring, headcount changes, or the AI implementation itself. That finding — drawn from the State of Organizations 2026 research covering more than 10,000 senior executives across fifteen countries and sixteen industries — points to a structural condition beneath the implementation story. Redefining a process flow is a decision, not a delivery activity. It requires the executive committee to have first specified what the AI investment was going to change about how work moves through the organisation, and to have approved the investment on that basis. Where that specification was missing at approval, the workflow redesign that would have made the investment pay off is not something implementation can now go back and add.
The reason this displacement is nearly inevitable has little to do with executives being careless with capital. The approval process around large technology investments has tended to test the wrong thing. The paper that comes to the committee usually specifies vendor selection, cost, and delivery timeline. What it rarely specifies is the mechanism by which the organisation will later know whether the investment answered the question that motivated it, because in most cases the question was never made precise enough to test. The investment gets approved against a general expectation of value. The review, when it comes, gets conducted against a specific record of delivery. Those are not the same object, and the space between them widens quietly for eighteen months before the board notices.
Each cycle erodes the executive's authority
McKinsey & Company (2026) also reports, in the same State of Organizations 2026 research, that two-thirds of leaders acknowledge their organisations are overly complex and inefficient, and that the traditional remedies of structural redesign, cost cutting, and flatter hierarchies are achieving diminishing returns. Those remedies yield less because each one treats the visible symptom as the object of intervention. Meanwhile the decision-making layer that produced the symptom keeps operating in the background, approving the next investment against the same thin specification.
If the reflex diagnosis is accepted, the immediate consequence is a second investment paper that specifies more implementation rigour and is approved against the same specification that produced the first result. The delivery this time will be tighter. The pilots will be smaller. The workforce readiness will be more carefully staged. But the paper still does not specify what would count as evidence that the approval layer had learned anything from the first cycle. Two cycles later, the board has spent twice the money and is asking the same question, and the executive has less confidence in the answer than they did the first time.
The liability that builds through those cycles is not primarily financial, although it is that as well. What erodes is the executive’s own diagnostic authority. Each cycle in which the diagnosis stops at implementation removes one further layer of the committee’s ability to see how it is actually deciding. By the third cycle, the committee’s own decision record is the least reliable evidence available to it about what the organisation has been trying to do.
Executives are now the capability gap
According to LHH (2026), digital and emerging technologies rose seven places in the 2026 View from the C-Suite to become the number one perceived development gap for executives globally. That finding is the most useful diagnostic prompt in this whole set. The gap is not sitting below the executive layer, waiting to be closed by workforce training. It is sitting inside the executive layer itself: in the committee that approves the investment, in the paper that specifies what it is expected to change, and in the review mechanism that will later be asked to say whether it did. Examining how the approval decision is structured means examining the executive layer’s own capacity to specify what a technology investment is being asked to prove, before the investment is approved.
The executive who examines that structure rather than the implementation record is not admitting a mistake. They are refusing to accept a diagnosis that stops one layer too shallow. That refusal is a governance act, the same act by which any large capital investment gets its integrity checked at approval. Bringing the AI investment back to approval-level scrutiny is a treatment the program should have had at the start, not a demotion of the program now.
Independent review targets the approval decision
The board’s next question, if the implementation diagnosis is accepted, will be what is being done to fix the implementation. That question is answerable, and it is not the question the executive should let the board settle on. The question that keeps the executive’s diagnostic authority intact is what an independent review of the AI investment approval would actually find, and whether the executive would still initiate it if they knew. If the honest answer is that such a review would surface an approval process the committee could not defend, then the review is exactly what the executive should initiate, so that the next investment paper is better than the last one.
That review does not sit inside the delivery partner, and it does not sit inside the assurance program that governs the current implementation. It sits with a firm that has no stake in the next investment cycle and no interest in defending the shape of the current one. That is where Zen Ex Machina works: at the approval layer, before the next paper comes to committee, so that the specification the investment is approved against is one the committee can later test. The organisation eighteen months from now looks different, because the decision that authorised the investment was made against a specification the committee can now defend.
What this means for senior leaders
- When an AI program underperforms, implementation is the second question. The first is whether the investment approval that authorised it was ever examined — and in most organisations it was not.
- Approval papers typically specify vendor, cost, and timeline, but leave the benefit case general enough that no later review can test whether it held.
- A second cycle approved on the same specification produces the same result at twice the cost, and each cycle removes another layer of the committee’s ability to see how it is actually deciding.
- The development gap is inside the executive layer — the committee’s capacity to specify what an investment must prove — not below it in the workforce, waiting to be closed by training.
- The review that protects your diagnostic authority sits outside the delivery partner and the current assurance program, and is initiated before the next paper reaches committee, so the specification the investment is approved against is one the committee can later defend.
References
- LHH. (2026, March 26). LHH 2026 View from the C-Suite [Press release]. LHH.
- McKinsey & Company. (2026). The state of organizations 2026: Three tectonic forces that are reshaping organizations. McKinsey & Company.
- PwC. (2026). PwC 2026 Global CEO Survey: Leading through uncertainty in the age of AI [Press release]. PricewaterhouseCoopers.
- PwC Australia. (2026). Bridging the AI skills gap: Why your workforce capability strategy needs attention. PricewaterhouseCoopers Australia.
