The RAND Corporation's 2024 research into AI project failure found something that standard technology due diligence cannot explain. AI projects fail at more than twice the rate of conventional IT projects — not because AI technology is twice as technically complex, but because the commitment decisions that approve AI investments are held to the same standard as lower-stakes IT decisions.
Those two things are not the same standard. They look identical from the outside. Both involve technical review, vendor assessment, cost-benefit modelling, and board sign-off. But the standard technology due diligence framework was built to evaluate whether the technology can work. It was not built to evaluate whether the organisation is committed to what working requires. For a conventional IT deployment, that gap is manageable. For an AI investment, it is where the failure lives.
AI projects fail at twice the rate of conventional IT projects
Despite adequate technical capability. The leading failure cause: "misunderstanding or miscommunicating the problem the AI needs to solve" — a commitment architecture failure, not a technical one.
A Good Technology, A Bad Commitment
The failure pattern I have observed across 25 years of technology and investment environments is consistent: the technology assessment passes. The vendor is credible. The use case is sound. The business case is compelling. The commitment is approved.
Then execution begins, and the gaps emerge. The data the AI system needs does not exist in the form the model requires. The workflow change the deployment demands is larger than the organisation anticipated. The governance architecture that was assumed to be in place was never actually built. The leadership accountable for the outcome changes eighteen months in. The investment fails — and the failure gets attributed to execution.
None of these are surprises. Every one of them is visible in advance with the right analytical framework. The reason they are not found is that the due diligence process does not ask for them. It asks whether the technology can work. That question is answered. The questions that determine whether the investment will work are not on the checklist.
"The due diligence approved the technology. The commitment was never validated. These are not the same thing, and the difference is where most AI investment losses originate."
The Five Questions Standard Due Diligence Doesn't Ask
RAND's top failure cause is "misunderstanding the problem the AI needs to solve." This is not a technical misunderstanding. It is a commitment misalignment. The organisation committed to the investment without agreeing on what success looked like or what delivering it would require. Every subsequent decision — about data, talent, workflow, governance — flows from this foundational clarity. If it does not exist at the point of commitment, execution will expose it at the worst possible time.
Deployment plans describe how the technology will be implemented under favourable conditions. Execution reality describes what the organisation will actually encounter: the data gaps, the integration complexity, the organisational resistance, the timeline compression. BCG's research found that organisations generating value from AI share one characteristic — they have honest clarity about the execution requirements before committing. Standard DD evaluates the plan. Decision intelligence evaluates the reality.
Gartner identified "inadequate risk controls" as a primary cause of GenAI project abandonment. The word "inadequate" is important. Risk controls were present. They were inadequate — which means the risk itself was underestimated or mischaracterised. In AI investments, embedded risk typically lives in three places: data integrity (the training and inference data is less reliable than assumed), operational dependency (the AI system creates dependencies that were not in the original risk model), and cyber exposure (the AI deployment expands the attack surface in ways the security architecture did not anticipate). Standard DD checks for stated risk. The unstated risk is where the exposure is.
Gartner's research also cited "poor data quality" as a leading cause of PoC abandonment. But data quality problems are almost always known in advance. They appear in every technical assessment. The reason they persist into production is that the investment thesis was built on optimistic data assumptions — and no one had the mandate to challenge those assumptions. Signal quality evaluation is not a technical audit. It is a question about whether the analytical foundation of the investment decision is honest. The organisations that lose most on AI investments are typically the ones that knew their data was problematic and decided to proceed anyway, assuming the problem would be solved during deployment.
Gartner's fourth cited cause of PoC abandonment is "unclear business value." Unclear business value is almost always a symptom of unclear ownership. When no named individual is accountable for the business outcome — not the project, the outcome — the investment becomes a shared responsibility. In practice, shared responsibility for an AI outcome means that when execution becomes difficult and the business case comes under pressure, there is no single decision-maker with the authority and the incentive to hold the commitment. The investment drifts into purgatory. Standard DD identifies the project sponsor. Decision intelligence validates whether that person is genuinely accountable for the outcome.
What Rigorous Due Diligence Looks Like
The organisations BCG identified as generating tangible value from AI — the 26% — have not found a superior technology or a more experienced technical team. They have built a due diligence process that answers these five questions before the commitment is made.
That process looks different from standard DD in one critical way: it is not primarily a technology review. It is a commitment review. The technology questions are answered within it, but they are not the primary questions. The primary question is whether the organisation has the clarity, the capability, the risk architecture, the analytical integrity, and the ownership structure to hold this specific commitment through execution.
When the answer to all five questions is yes, the investment has a dramatically different probability profile. When any answer is "we'll figure it out," that is where the failure will live. Decision intelligence is the discipline of finding those gaps before capital is committed — not after it has been lost.