A Number That Should Sting
A survey of 700 senior business leaders by the technology and management consultancy Emergn, published on July 2, 2026, puts a hard figure on a soft problem: United States organizations waste an average of 2.4 percent of annual revenue on AI initiatives that fail to deliver expected returns. For a company with a billion dollars in revenue, that is 24 million dollars a year evaporating on projects that do not work, and the survey suggests it is happening not through bad luck but through bad discipline.
The finding lands at an awkward moment. Enterprises have spent two years being told that AI is existential, that the risk is moving too slowly rather than too fast. That message drove a surge of initiatives, and it also created a reluctance to admit when any given one has failed. The Emergn data quantifies the cost of that reluctance. The waste is not the price of experimentation, it is the price of experiments nobody is willing to end.
The Real Failure Is the Refusal to Stop
The most damning statistics in the survey are about stopping, not starting. Only 30 percent of organizations consider shutting down an underperforming AI initiative to be normal practice. Nearly half stop projects only after significant time and money are already sunk. The average organization is running more than six transformation and AI initiatives at once, and one in ten operate with no formal oversight or governance structure at all. The picture is of portfolios that expand but never contract.
Emergn's chief executive, Alex Adamopoulos, put the diagnosis sharply, saying the problem is not that companies are taking risks on AI, it is that they are funding activity and calling it progress. That distinction is the heart of the matter. Activity is easy to generate and easy to celebrate: a pilot launched, a proof of concept demoed, a vendor engaged. Progress is harder, because it requires killing the things that are not working to free resources for the things that are, and killing projects is politically painful.
Boards Are Flying Blind
The governance gap the survey exposes is startling. Just 27 percent of United States leaders say they can give their board real time visibility into every transformation and AI program. More than 20 percent admit that status reports paint an overly optimistic picture compared with reality, and nearly a quarter of senior leaders concede they are reluctant to admit when an AI project has failed. Put together, these numbers describe organizations where the people funding AI cannot see clearly what they are funding.
This is a controls problem dressed as a technology problem. When status reports are optimistic by default, when failure is culturally unspeakable, and when boards lack real time visibility, capital keeps flowing to dead initiatives because no honest signal ever reaches the people who could redirect it. The absence of that signal is not a minor inefficiency, it is the mechanism by which 2.4 percent of revenue disappears. The fix is not better models, it is better instrumentation of the truth.
Accountability Beats Enthusiasm
The survey's constructive thread is about accountability. Christopher Panneck of KPMG, quoted in the research, argued that clear accountability matters, and that when business owners are tied to outcomes, decisions become more pragmatic. That is the crux. AI initiatives owned by no one in particular, championed by enthusiasm rather than a named executive answerable for results, are the ones that drift. Tie a project to an owner whose credibility depends on its outcome, and the incentive to keep a failing effort alive evaporates.
We would go further. The discipline that distinguishes organizations getting value from AI is not superior technical talent, it is the willingness to establish clear metrics, defined decision points, and honest reporting, and then to act on what those things reveal. That is unglamorous governance work, the sort that does not make a keynote. But it is precisely what separates a portfolio that compounds value from one that compounds waste, and the survey suggests most companies have not yet done it.
Why Sunk Cost Thinking Runs So Deep
The behavior the survey documents, funding failing projects rather than stopping them, is not stupidity, it is human nature operating inside organizational incentives. The sunk cost fallacy is one of the most robust findings in behavioral science: people and institutions escalate commitment to failing endeavors precisely because they have already invested in them. Layer on the reputational stakes of a public AI initiative, championed by a named executive, and the incentive to keep a dying project on life support becomes overwhelming.
Fixing it therefore requires structural countermeasures, not exhortations to be more rational. Predefined kill criteria set before a project starts, stage gates that force an explicit continue or stop decision, and a culture that treats a well reasoned shutdown as a success rather than an admission of failure all work against the natural pull of sunk cost thinking. The organizations that manage AI portfolios well are the ones that have engineered these countermeasures into their process, so that stopping does not depend on any individual finding the courage to say the project failed.
The Uncomfortable Mandate for CIOs
For the CIO, this research is both cover and challenge. It is cover because it demonstrates that AI waste is systemic, not a personal failing, and that the fix is a discipline the whole enterprise lacks. It is a challenge because governance of the AI portfolio is landing squarely on the technology leader's desk, and doing it well means becoming the person who says no, who insists on kill criteria, and who makes failure discussable rather than shameful.
The organizations that will pull ahead are the ones that treat AI portfolio management like any other capital allocation: with stage gates, honest metrics, and the routine discontinuation of what does not work. That sounds obvious, and the survey shows it is rare. The 2.4 percent of revenue at stake is not a technology cost, it is the price of governance that has not caught up to ambition. Closing that gap is the least exciting and most valuable AI project most companies could run this year.
