Every AI initiative that went to the investment committee came with a governance mechanism: a sponsor, a business case, and a set of metrics to show progress. Three years of AI investment later, 37% of enterprises globally report unchanged EBIT despite scaling AI adoption by six percentage points over the same period (Tinkoff et al., 2026). The individual projects were on track; the portfolio was not.
In Brief
- AI portfolios were built with an approval mechanism; the stopping mechanism was never included.
- 37% of enterprises report unchanged EBIT despite scaling AI investment; the governance model can approve projects but has no mechanism to stop them.
- Capital approval, AI strategy, and project execution sit in different accountabilities; no one owns the stopping decision.
- Gartner projects more than 40% of AI projects will be cancelled by 2027 due to cost, unclear value and risk gaps.
- The CFO who can name which AI projects should stop changes the board meeting from a project defence to a reallocation discussion.
No one owned the stopping decision
Most AI portfolios were designed to approve investment, not to reallocate it. The governance model, with investment committee approvals, project-by-project business cases and Chief Digital Officer (CDO)-led strategy oversight, is built for the question ‘should we start this?’ It has no mechanism for the question ‘should we stop this, or redirect the capital to something with greater portfolio impact?’ Those are structurally different governance acts, and most AI portfolio governance models contain only the first one.
Most AI portfolios were designed to approve investment, not to reallocate it.
The accountability structure does not include a single, accountable owner of the portfolio reallocation decision, and that is the structural condition that keeps the gap invisible until the board raises it. Capital approval sits with the CFO, AI strategy with the CDO, and project execution with the business unit sponsors. None of these accountabilities covers the question of which projects to defund when the collective outcome is not materialising. That accountability was never assigned because the governance model assumed continuation rather than reallocation. The result is a portfolio that can measure project-level progress in precise detail and cannot produce a defensible answer to the question ‘what should stop?’ Capital infrastructure programmes have faced the same design gap: individual project sponsors account for their own milestones while no portfolio authority holds accountability for the overall programme’s investment return.
Project approval metrics in most portfolios measure each initiative’s forward progress, and have no mechanism to identify when any initiative has passed its point of marginal return. A predictive maintenance initiative that reduced unplanned downtime in year one is producing value by its own metric. Six similar initiatives running in parallel, each meeting its own progress measures, do not automatically produce an aggregate EBIT result. The governance model processes each case individually; it has no mechanism to compare them against each other at the portfolio level, or to ask whether the aggregate investment thesis, that AI at scale would produce EBIT impact, is still holding. Gartner projects that more than 40% of AI projects will be cancelled by 2027, due to cost, unclear value, and risk management gaps (Gartner, 2026). That projection describes what happens when external pressure forces stopping decisions that portfolio governance was never designed to make.
Only 6% of organisations qualify as AI high performers: those generating measurable EBIT returns from their investment (Tinkoff et al., 2026). The structural distinction that separates them is that they built, alongside the approval mechanism, a governance model that could identify and defund under-performing projects and redirect the capital.
The portfolio runs without a stop
The board’s question, why has three years of AI investment not moved the EBIT numbers, requires a governance answer the portfolio was never designed to produce. Deciding which projects should stop is a reallocation act. It requires a single, accountable owner of the stopping decision, portfolio-level metrics that aggregate project outcomes against a shared EBIT target, and the authority to defund projects that are meeting their own milestones while contributing nothing to the collective result. Most AI portfolio governance models contain project approval, project monitoring and CDO oversight; none includes a mechanism for the stopping decision.
The longer that governance gap remains open, additional capital continues to be committed to continuation. Each quarterly reporting cycle, projects meeting their own metrics receive continued investment, and the board’s account of portfolio progress diverges further from the EBIT result. 84% of Australian CIOs report they are not strongly confident AI spending has delivered measurable value (Logicalis, 2026). That is the natural consequence of a governance model built entirely for approval: projects report progress accurately while the portfolio produces no aggregate account of EBIT impact. The board’s request for a portfolio review before the next capital cycle is the stopping mechanism arriving from outside the governance structure, because the governance structure never built one inside it.
The governance question before the review
The diagnostic question to put to the portfolio governance structure before the board review begins is a governance design question: does the governance model currently contain a single, accountable owner of the reallocation decision, someone whose accountability covers what stops as well as what was approved? In most AI portfolios, that role does not exist. The people who built the governance model built it for the right early-stage purpose: an approval-oriented investment process is a rational response to the early years of AI investment, when the priority was building capability and exploring use cases. The stopping mechanism becomes necessary when the portfolio moves from exploration to accountability: when the board is no longer asking ‘what are we trying?’ and starts asking ‘what is this producing?’
What this means for senior leaders: A CFO arriving at the board with a reallocation argument needs a different kind of analysis. Most individual initiatives will show progress against their own metrics, and that CFO needs to answer one structural question first: does any role in the governance model hold the authority to stop a project and redirect its capital. An independent diagnostic perspective on the portfolio governance question, applied before the board review is framed, changes what the review can produce.
The CFO who arrives at the board with a portfolio stopping mechanism is answering the governance question the investment cycle was never designed to ask.
References
- Gartner. (2026). Hype Cycle for Artificial Intelligence, 2026. Gartner Inc.
- Logicalis. (2026). Global CIO Report 2026: Australia findings. Logicalis. https://itbrief.com.au/story/australian-cios-warn-ai-ambition-outpaces-governance
- Tinkoff, K., Van der Veken, A., Chui, M., & Balakrishnan, S. (2026, August). State of AI in 2026: On the Road to ROI. McKinsey & Company.
