Geyto Research · Mariano D'Angelo · Evidence review · · 3 min read

Why AI pilots do not scale. What the studies show

Almost every organization is trying AI and few are scaling it. What recent studies show about the distance between the pilot and everyday work.

In brief: almost every organization has already tried AI, and a minority manages to bring it into everyday work with results. Recent studies agree on where the gap opens: not so much in the quality of the models as in integration with the work, in redesigning processes and in the capacity to sustain the change. A pilot proves that the tool works. It does not prove that the work will change.

The gap in numbers

McKinsey's 2026 survey, with 1,719 participants in 97 countries, found that nearly nine in ten organizations use AI in at least one function. But 44% say they are scaling it across the enterprise, and 37% attribute any impact on earnings to it, almost the same share as the year before. The organizations McKinsey identifies as AI high performers are 6%.

Grant Thornton's 2026 survey shows the same gap from another angle: organizations with fully integrated AI report AI-driven revenue growth almost four times as often as those still piloting (58% versus 15%).

The 95% report

In 2025, a report from MIT NANDA, The GenAI Divide, became a reference because of one figure: 95% of organizations were getting no measurable return from their generative AI investments. The study combined a review of more than 300 public initiatives, interviews at 52 organizations and a survey of 153 executives. The authors themselves note that the figure is directional: it rests on what organizations report, not on their financial statements.

More useful than the figure is its explanation. For the authors, the main barrier is not model quality or regulation, but what they call a learning gap: most systems do not retain what they learn, do not adapt to each organization's context and do not improve with use. Pilots work under controlled conditions and stall when they have to fit into real processes.

The same report records another telling finding: only 40% of companies had bought an official subscription to a language model, while in more than 90% people were using personal AI tools for their work. AI was already in the work; the organization was not yet.

What the ones that scale do differently

In McKinsey's survey, nearly three quarters of high performers say they fundamentally redesigned how the work is done, compared with about a quarter of the rest. They are also twice as likely to have visible senior-leader commitment and a defined way of measuring AI's impact.

These are correlations, not proof of cause. But they point in the same direction as an idea more than twenty years old: M. Lynne Markus called technochange the projects in which technology is the means to change how an organization works, and showed that they fail when managed only as technical projects. Scaling AI looks more like that than like installing a tool.

What this says about adoption

A pilot answers a narrow question: whether the tool can work. Adoption answers another: whether the work changes and whether that change holds. Between the two lie things the pilot does not measure: which tasks change, which new hand-offs appear between people and systems, who reviews what AI produces, which data and systems have to be ready, and what capacity there is to sustain all of that where the work happens.

When the gap is large, both parts can be worked on: building capacity in the organization, and also adjusting the initiative, its scope, its sequence and what it asks for first.

What to do about it

  • Design the pilot with scale in mind: test the tool and also the change in the work.
  • Measure more than use: whether the way of working changed and whether it holds.
  • Look at the informal use that already exists; it says a lot about real demand.
  • Review the initiative, not only the people, when the pilot does not scale.

More in Using AI is not adopting it, in Why implementations stall after go-live and in The productivity paradox.

Geyto Research publishes essays, evidence reviews and market notes on AI and technology adoption in organizations.

Frequently asked questions

How many organizations are scaling AI?

In McKinsey's 2026 survey, nearly nine in ten organizations use AI in at least one function, but 44% say they are scaling it across the enterprise and 37% attribute any impact on earnings to it.

Is it true that 95% of AI pilots fail?

The figure comes from MIT's The GenAI Divide report (2025), which found that 95% of organizations were getting no measurable return from their generative AI investments. The authors themselves note that it is a directional estimate, based on interviews and surveys rather than financial reporting.

What do the organizations that scale do differently?

In McKinsey's survey, nearly three quarters of those getting the most value redesigned how the work is done, compared with about a quarter of the rest. They are also twice as likely to have visible senior-leader commitment and a defined way of measuring impact.

Sources

  1. Challapally, A., Pease, C., Raskar, R. and Chari, P. (2025). The GenAI Divide: State of AI in Business 2025. MIT NANDA. Summary in Computing.
  2. McKinsey & Company (2026). The state of AI. Survey of 1,719 participants in 97 countries, May–June 2026.
  3. Grant Thornton (2026). 2026 AI Impact Survey Report. Survey of 950 business leaders, February–March 2026.
  4. Markus, M. L. (2004). Technochange management: using IT to drive organizational change. Journal of Information Technology, 19(1), 4–20.

Share