The question landing in board papers and ministerial briefings this quarter is not whether AI works. It is sharper than that. After eighteen months of program activity — signed vendor contracts, an internal AI strategy, named use cases, a steering committee, and visible executive sponsorship — why has enterprise performance not moved? Boards are now asking, in effect, where the AI investment readiness assessment sits in the record, and who signed it.
In Brief
- Boards are asking why a year of AI investment has not moved enterprise performance, and the answer almost never sits inside the technology stack. The missing input is an AI investment readiness assessment produced before the program was authorised.
- The organisational conditions that determine whether AI investment produces enterprise value are knowable in advance, and were rarely assessed before the investment was authorised.
- Most programs surface the gap between activity and outcomes only after the funding cycle has already committed to the next phase, when upstream correction is no longer available.
- The Australian capability picture is structurally tight: 78% of senior technology leaders name AI as the defining trend, while only 7% say the country has the capability to meet what the trend requires (Tech Council of Australia and Datacom, 2026).
- The question that distinguishes a board with a defensible AI account from one without it is whether the conditions for value were assessed, and named, before the cheque was signed.
The question landing on board agendas this quarter is upstream of the technology
What makes the question uncomfortable is not the technology. The conditions that determine whether AI investment produces enterprise value were knowable before the investment was authorised, and in most organisations they were never assessed. The gap between AI activity and AI outcomes is structural. It surfaces only after the funding cycle has committed to the next phase. By the time the board asks the question, upstream correction is no longer available.
This is the diagnostic that matters now. Not what to do next: what should have been assessed before the program was approved.
The evidence shows the constraint sits outside the technology stack
Gartner’s research published in January 2026 reported that more than half of generative AI projects are abandoned after proof of concept, and that the dominant failure modes are absence of clear business value and inadequate data readiness, rather than technical capability (Chandrasekaran, 2026). The technology is not the constraint. The constraints sit elsewhere in the system: in how the business case was framed, in whether the data the program assumed was actually available at the quality the use case required, and in whether the organisation had a mechanism for converting model output into a decision a human would act on.
McKinsey & Company’s 2026 organisational research describes the operational picture from the other side. Eighty-eight per cent of leaders are deploying AI, but the deployments are stuck in piecemeal use cases, with efficiency improvements at the team level and no movement at the enterprise performance level (McKinsey & Company, 2026). The investment produces activity, the activity produces local wins, and the local wins do not aggregate into anything an executive can present to a board as a return on the program.
The Australian picture sharpens the same diagnosis. Deloitte Australia’s 2026 enterprise AI report found that only 22% of Australian companies have advanced agent governance models in place, meaning the structures that would allow autonomous AI to be deployed at scale, with accountability traceable to a named decision-maker, exist in a small minority of the organisations now investing in agent capability (Deloitte Australia, 2026). The Tech Council of Australia and Datacom 2026 Tech Leaders Survey reports that 78% of senior technology leaders identify AI as the defining trend of the year, while only 7% say Australia has the capability to meet what that trend requires (Tech Council of Australia and Datacom, 2026). Investment is moving ahead of the conditions that would let it land.
A typical AI business case does not assess the organisational conditions for value
The conditions that determine whether an AI investment produces enterprise value are not a mystery. They are also not what most business cases assess. A typical AI business case looks at model accuracy, vendor capability, indicative cost, and a forecast efficiency saving. None of those four predict whether the program will produce enterprise value, because none of them describe the organisation that has to absorb the program’s output.
What predicts enterprise value is a different set of conditions, and an AI investment readiness assessment is the instrument that names them. The data the program depends on has to be governed at the quality the use case actually requires, not the quality the use case assumes. Decision rights for AI-influenced decisions need to be defined before the model is deployed, not after a regulator or auditor has asked who authorised an outcome. The operating model around the AI must have the cycle time and feedback loops to absorb model output and convert it into action at the rate the business case assumed. The leadership team needs the diagnostic literacy to read AI outputs as evidence, rather than as answer. And the program needs a defined enterprise outcome, owned by an accountable executive, that the use cases collectively roll up to, instead of a portfolio of locally justified pilots.
These conditions are knowable. They are also assessable in weeks, by anyone with a structural read of the organisation. The reason they are typically not assessed before the investment is authorised is straightforward: the AI business case has been written by the team that wants the AI program, and that team has no incentive to surface structural conditions which, if absent, would mean the program cannot succeed in its current form.
This is not a failure of capability inside those teams. It is a structural feature of how AI investment gets sponsored. The proposing team is the assessing team. The conditions for success, most of which sit outside that team’s authority, fall outside the scope of the proposal. The board approves it. The conditions remain unassessed. The program runs. Eighteen months later the board asks why performance has not moved, and the structural answer is not available, because nobody upstream of the cheque was asked to produce it.
A board reviewing AI performance is reviewing an upstream assessment that did not happen
If the diagnosis holds, a board reviewing AI investment performance is not looking at a technology question. It is looking at the consequence of an upstream assessment that was either skipped, scoped to the wrong question, or owned by the party with the strongest interest in the answer being yes.
That reframes the conversation. A minister or board chair asking why performance has not moved is asking the right question of the wrong layer of the system. The honest account does not live inside the AI program. It lives in the assessment that was, or was not, produced before the program was authorised. An executive who can produce that account, or who can name what would need to be reassessed before further investment is approved, is in a defensible position. An executive who can only produce program activity reports is not.
There is also a compounding cost. An AI investment that proceeds without the upstream assessment does not stand still. It accumulates further investment cycles on the same structural foundation. Each cycle extends the gap between activity and outcomes, and each one makes the eventual unwinding more expensive. The cost is not the wasted spend on the current cycle. It is the structural debt accruing across cycles, made harder to read by the program’s growing footprint inside the organisation.
And then there is precedent. AI will not be the last general-purpose capability the organisation is asked to absorb. The pattern of approving a strategic investment without assessing the organisational conditions for its success is the pattern that will repeat with the next capability, and the one after that. What gets established now is not an AI program. It is a governance precedent for how the organisation evaluates strategic investment under pressure.
The defensible question to test now
The defensible question for an executive whose board is now asking what the AI investment has delivered is not “what use cases are running?” or “what is the model accuracy?”. Those questions describe activity. The defensible question is upstream of the activity:
Before this investment was authorised, what assessment was produced of the organisational conditions on which its enterprise value depended, and who, with no interest in the answer being yes, signed it?
If the assessment exists and the signature is independent, the executive has a credible account for the board. If the assessment exists but the signature came from the proposing team, the executive knows the conversation that needs to happen with the board now, before the next funding cycle. If no assessment exists at all, the executive knows what to initiate before the next investment decision is taken, and that is the position a board will recognise as leadership.
The executive who can name the conditions that should have been assessed before the cheque was signed is in a different conversation from the one whose only response to the board is what the program is currently doing.
References
- Chandrasekaran, A. (2026, January). Why half of GenAI projects fail: Avoid these 5 common mistakes. Gartner.
- Deloitte Australia. (2026). The state of AI in the enterprise 2026. Deloitte.
- McKinsey & Company. (2026). The state of organizations 2026.
- Tech Council of Australia and Datacom. (2026). Australian tech leaders survey 2026. Tech Council of Australia.
