Proof-of-Concept Purgatory: The $30 Billion Question Organisations Aren't Asking

In 2024, enterprise organisations collectively invested approximately $30 billion in generative AI pilots. MIT's Project NANDA — State of AI in Business 2025, analysing those 2024 deployments — found that 95% delivered zero measurable P&L impact. Not zero ROI — zero measurable result of any kind. The pilots ran. The technology often worked. The investment was written off.

S&P Global's 2025 survey found that 42% of companies had abandoned most of their AI initiatives — a figure that had been 17% the year before. Gartner had already signalled this in July 2024, predicting that at least 30% of generative AI projects would be abandoned after proof of concept by year end, citing poor data quality, inadequate risk controls, escalating costs, and unclear business value.

The industry has named the phenomenon: pilot purgatory. The framing positions it as a scaling problem — pilots that cannot reach production. My view is different. Pilot purgatory is not a scaling problem. It is a commitment architecture problem. The question that needed answering was never asked — and it cannot be answered by running the pilot.

The Acceleration Paradox

The data presents a paradox that the standard framing cannot explain. AI investment is accelerating — spending surged to record levels in 2024. Adoption is widening — 88% of enterprises now have active AI deployments, per McKinsey's 2025 research. Yet the abandonment rate is also accelerating. More investment and more adoption are producing more failures, not fewer.

If pilot purgatory were a technology maturity problem, this trajectory would not make sense. Technology matures with adoption. Failure rates should fall as the ecosystem develops. The fact that they are rising tells us the problem is not in the technology. It is in the decision-making process that approves the pilot in the first place.

Gartner's Rita Sallam, Distinguished VP Analyst, put it directly: "After last year's hype, executives are impatient to see returns on GenAI investments, yet organisations are struggling to prove and realise value." The impatience is real. But impatience with a PoC process is not the same as impatience with a technology. It is impatience with a commitment architecture that was never designed to produce the answer the organisation needed.

The Question The Pilot Cannot Answer

A proof of concept is designed to answer one question: can the technology work in this context? It is the right question. It is not the only question. And it is not the question that determines whether the investment will produce value.

The question the pilot cannot answer is: Is this organisation committed to what success requires?

That question has three dimensions that must be resolved before the pilot begins:

• Outcome clarity: What does success look like in concrete business terms — not technology terms? If the answer is "we'll know it when we see it," the pilot should not run.
• Commitment reality: What does success require from the organisation — in workflow change, data infrastructure, talent, governance? Is the organisation actually prepared to provide it? If the preparation is contingent on the pilot succeeding, the logic is circular and the commitment is not real.
• Decision authority: Who owns the outcome — not the project, the outcome? If there is no named leader accountable for the business result, the pilot is an experiment without a sponsor. It will not survive the transition to production when the complexity increases and the enthusiasm fades.

"The pilot succeeds when the technology works. The investment succeeds when the organisation was ready for it to work — and had committed to acting on what it learned."

The Real Cost of Purgatory

The financial cost of pilot purgatory is direct and quantifiable. Thirty billion dollars in 2024 generative AI pilots, 95% of which produced no measurable return. That is an extraordinary concentration of capital destruction in a single year, and the trajectory in 2025 — with abandonment rates rising — suggests the problem is not self-correcting through experience.

But the strategic cost is larger than the financial cost. Every failed pilot creates organisational debt. It produces scepticism about AI investment among the leadership team that approved the budget. It creates risk aversion among the technical teams that ran the project. It generates a narrative — "AI doesn't work here" — that is factually wrong but organisationally durable.

The organisations suffering most from pilot purgatory are not those with the worst technology. They are those with the most pilots and the least commitment architecture. They are running more experiments precisely because they are avoiding the harder question of what they are actually committed to.

What Exits The Loop

BCG's research found that 26% of organisations have developed what it calls "the necessary set of capabilities to move beyond proofs of concept and generate tangible value from AI." What that cohort has in common is not superior data infrastructure or more experienced technical teams, although those factors matter. What they have is a decision architecture that precedes the pilot.

Before the PoC is approved, three decisions have been made: what success looks like, what it requires, and whether the organisation is committed to both. The pilot then runs against a defined commitment rather than an open aspiration. When it succeeds, there is a ready path to production. When it reveals a gap, the organisation can decide whether to close the gap or not pursue the commitment. Either outcome is productive. Neither outcome is purgatory.

If any of these answers is "to be determined," the investment case has not been made. The pilot is not the place to make it. The commitment architecture must come before the experiment — not emerge from it.