The End of the AI Pilot

Adoption solved itself. Value didn't.

McKinsey's 2025 State of AI survey — 1,993 respondents across 105 countries, fielded June–July 2025 — found that 88% of organizations now use AI in at least one business function, up from 78% in 2024. Generative AI use specifically jumped from 33% to 72% year over year. On paper, the adoption question is closed.

What isn't closed is the value question. The same survey found that nearly two-thirds of organizations have not yet begun scaling AI across the enterprise — they are still running experiments in isolated pockets, without deep integration into workflows. Only about 6% qualify as "AI high performers," defined as seeing more than 5% EBIT impact from AI. BCG's parallel research puts it even more starkly: 60% of companies are generating "hardly any material value" from AI despite real investment, and only 5% qualify as "future-built" — organizations systematically capturing substantial value.

Pilot purgatory, by the numbers

"Pilot purgatory" is the term analysts now use for a specific organizational state: a project has cleared feasibility testing but never advances to production — not cancelled, not resourced, not scaled, just perpetually pending. Three independent research streams converge on roughly the same scale of the problem, measured at different points in the same lifecycle:

• IDC, in its research with Lenovo, found that for every 33 AI proofs of concept an enterprise launches, only about 4 reach production — an 88% failure rate.
• MIT's NANDA Initiative put GenAI pilot failure even higher, at 95%, concluding the cause is "not primarily the model technology... but the integration into workflows, organisational alignment, and underlying data readiness."
• Gartner forecasts that 30% of generative AI projects will be abandoned outright after the proof-of-concept stage, and traces 85% of AI project failures to poor data quality.

And the trend line is worsening, not improving: S&P Global Market Intelligence found the share of companies abandoning most of their AI initiatives rose from 17% to 42% in a single year.

"It is not primarily the model technology that is failing, but the integration into workflows, organisational alignment, and underlying data readiness."

— MIT NANDA Initiative, on why enterprise GenAI pilots fail to scale

What separates the 6% who scale

The organizations that do convert pilots into enterprise value don't run better pilots — they run a different playbook. McKinsey's high performers set growth objectives (not just efficiency targets), fundamentally redesign workflows around AI rather than bolting it onto existing processes, deploy across functions rather than in isolated pockets, and invest at a different order of magnitude: 20%+ of digital budgets, against a much smaller share for everyone else. BCG's AI at Work 2025 survey of over 10,600 workers found the same pattern from the ground up: companies unlocking AI's full potential invest heavily in people — training and change management — and only 25% of frontline workers say their leaders provide enough guidance on AI in the first place.

That reframes the whole exercise. A pilot answers "does this work?" Capability answers "can we do this repeatedly, safely, and profitably at scale?" — a question about data governance, workflow design, measurement, and people, not about which model to buy next.

The data

Where CometX fits

Escaping pilot purgatory is a capability-building problem, and it maps onto three of CometX's five practices directly: