The 74% Problem: Why AI Investment Failures Are Commitment Failures, Not Technology Failures

In October 2024, Boston Consulting Group published a survey of 1,000 senior executives across 59 countries and 20 sectors. The finding that made headlines: 74% of companies had yet to show tangible value from their use of AI. This, despite the fact that AI investment was accelerating, adoption was spreading, and almost every major organisation had at least one AI initiative in flight.

The industry's response to this finding has been, broadly, to diagnose the wrong problem. Data quality is insufficient. Change management was neglected. The technology is more complex than anticipated. These answers have one thing in common: they locate the failure in the execution phase. They treat the commitment as sound and the delivery as flawed.

In my experience, that is the wrong frame. The failure is not in the execution. It is in the decision to commit.

The Gap That Shouldn't Exist

McKinsey's 2025 State of AI report found that 88% of enterprises had deployed AI in at least one business function. That number is striking. Almost universal adoption. Yet the same research found that only 6% qualify as high performers — defined as organisations where AI contributes more than 5% of EBIT. Ninety-four percent of enterprises with active AI deployments are not seeing significant value.

The RAND Corporation studied this failure pattern directly. Their 2024 research — based on interviews with 65 data scientists and engineers with five or more years of experience — found that AI projects fail at more than twice the rate of conventional IT projects. That doubling is not explained by technical complexity alone. Conventional IT projects carry technical complexity too. Something specific to the AI commitment decision is driving the divergence.

"The mandate failed. Not because the technology didn't work. Because the organisation was never actually committed to what success required."

Why The Industry Has The Wrong Diagnosis

RAND's research identified the top cause of AI project failure as "misunderstanding or miscommunicating the problem the AI needs to solve." Read that carefully. The leading failure cause is not a data problem, a security problem, or a change management problem. It is a problem of not understanding — or not being honest about — what the organisation is actually trying to do.

That is a commitment architecture failure. It means the decision to invest was made without sufficient clarity about the outcome being committed to. The pilot was approved, the budget was allocated, the vendor was selected — and the organisation was not aligned on what success looked like, what it required, or whether it was actually prepared to deliver it.

Data quality, change management, and technical complexity are real challenges. But they are predictable challenges. They are visible in advance. Organisations that commit to AI investments without stress-testing their readiness for these challenges are not the victims of unexpected complexity. They are the product of a commitment decision that was never properly validated.

The Commitment Failure Pattern

In 25 years operating in technology and AI environments across 50+ countries, I have watched this pattern repeat. It follows a consistent sequence:

• The investment thesis is compelling and the pressure to act is real
• The question of whether the technology can work gets answered — through pilots, vendor assessments, technical reviews
• The question of whether the organisation is committed to what success requires does not get asked
• Capital is deployed
• Execution reveals the gaps that validation would have found
• The failure gets attributed to execution when the cause was the commitment decision

The distinction matters because the remedies are different. If the failure is in execution, the answer is better project management, more experienced teams, stronger governance. If the failure is in the commitment decision, the answer is a different kind of evaluation before the commitment is made.

What Changes When You Ask The Right Question First

BCG's research identified 26% of organisations as having developed the capabilities needed to generate tangible value from AI. The common characteristic of that cohort is not superior technology, better data, or more AI experience. It is decision architecture: these organisations evaluate the commitment to the outcome before the investment in the technology.

The question is not "can we deploy this AI capability?" That question is almost always yes. The question is "are we committed to what success with this capability actually requires?" That question is far harder, and most organisations avoid it because the honest answer creates friction with the investment thesis.

Commitment to an AI outcome means committing to the workflow redesign that makes it valuable. Committing to the data infrastructure that makes it reliable. Committing to the governance architecture that makes it defensible. Committing to the leadership alignment that holds it through execution. These are not technical commitments. They are organisational commitments. And they can be assessed — rigorously, before capital is deployed.

That is what decision intelligence is. Not better technology evaluation. A fundamentally different question: not whether the investment can work, but whether the organisation is committed to what working actually requires. The 74% are not failing because AI is hard. They are failing because that question was never seriously asked.