Everyone's piloting. Almost nobody's shipped.

Depending on which survey you ask, somewhere between 11% and 44% – and the true figure is closer to the low end. Four major 2026 surveys approached the same question from different angles, and all four land in the same place: most organisations are still short of real, enterprise-scale AI use.

McKinsey's State of AI 2026 (n=1,719, fielded May–June 2026) found 88% of organisations regularly use AI in at least one business function, but only 44% report it is scaling across the enterprise – up from 38% a year earlier. Deloitte's State of AI in the Enterprise 2026 (n=3,235 leaders, 24 countries) found only 25% have moved 40% or more of their AI pilots into production. KPMG's Global AI Pulse Q1 2026 (n=2,110 C-suite and senior leaders, 20 countries) put the hardest number on it: just 11% qualify as true "AI leaders" scaling agents enterprise-wide.

Usage is universal. Value is not – and that gap is the whole story. McKinsey's 88% "regular use" figure and its 37% "any EBIT impact" figure are measuring two entirely different things, and the distance between them is where most AI budgets are currently being spent without a return.

That 6% figure – organisations attributing 5% or more of EBIT to AI, with significant impact – has stayed flat year over year even as usage has climbed sharply. Conviction is outrunning proof.

KPMG's Global AI Pulse Q1 2026 is the most finely sliced view available, splitting organisations across five stages of AI agent maturity:

The great majority of organisations sit in the exploring-to-deploying band, with only a minority reaching genuine multi-function scale. Separately, KPMG found 95% of organisations now have a formal AI strategy and plan to spend a weighted global average of $186 million on AI over the next 12 months – yet only 8% say they have established a working method for measuring AI's return on that investment, even though 64% report meaningful business value.

MIT NANDA's GenAI Divide report – based on 52 executive interviews, a 153-leader survey, and analysis of 300 public AI deployments – found the barrier is overwhelmingly organisational, not technical. The tools that succeed personally (ChatGPT and equivalents, used daily by over 40% of knowledge workers) are frequently judged unreliable the moment they're wired into enterprise workflows, because most pilots cannot retain feedback, adapt to context, or improve over time.

MIT calls this the "learning gap." It also documents a "shadow AI economy": in the large majority of firms surveyed, employees quietly use personal AI tools even where the sanctioned pilot has stalled or failed outright.

Every survey in this research converges on the same answer: something else. Deloitte's interview material describes a recurring pattern where organisations keep approving new pilots because pilots are cheap, visible, and politically easy, while the slower work of governance, integration, and workflow redesign gets deprioritised – producing what Deloitte calls "pilot fatigue": a growing inventory of disconnected trials rather than a smaller number of things that actually ship.

KPMG's own framing of its data is blunt: the risk for the 89% still stuck is not whether to accelerate deployment, but whether they do so while compounding the integration debt and governance deficits already constraining their returns.

It helps, but it doesn't solve the problem. McKinsey found 40% of $1 billion-plus revenue organisations are scaling AI agents somewhere in the enterprise, up sharply from 27% a year earlier. KPMG's sample – three-quarters of whom represent $1 billion-plus revenue organisations – still found only 11% had reached genuine enterprise-wide scaling, despite planning to spend a weighted average of $186 million on AI in the next year.

Scale buys options and budget. It does not automatically buy execution.

Yes – this is precisely the trough described by the AI J-Curve™: performance dips before it improves, because the organisation has adopted a new capability faster than it has redesigned the work, governance, and incentives around it.

The organisations pulling ahead in every survey cited here are not the ones with the most pilots. They are the ones furthest through the trough.

Consistently, it is talent and workflow design – not model choice or spend. Deloitte's research on "superscalers" (the small minority who have scaled 70% or more of their pilots) found their main differentiator was giving as much of the workforce as possible access to AI tools, positioned to support rather than replace what people already do well.

KPMG's data shows AI leaders compounding their advantage every quarter while everyone else accumulates integration debt. The practical takeaway from four surveys that all say the same thing: by late 2026, the constraint is very rarely the model. It is almost always the organisation.