The handover is the moment a pilot leaves the team that built it and reaches the area that has to use it every day. We found no public study that follows the same companies through that stretch. What exists are snapshots from either side of it. They prove no cause, but they show where to look.
Plenty of use, little reported return
In McKinsey’s 2026 survey, 89% say their company regularly uses AI in at least one function. Among organizations using AI, 34% are piloting and 44% say they are at least scaling. 37% attribute some EBIT impact to AI, and about 6% reach 5% of EBIT or more, with value they call significant. All of it is self-reported, with no probability sample declared. The bases change between stages, and McKinsey sells transformation.
Use stays shallow
Among US employees, 52% use AI at least a few times a year, 30% a few times a week or more, and 15% every day. Where the organization already has AI, 14% strongly agree it changed how work gets done. It covers only the US and is self-reported. We found no public study showing use falling after launch.
In Brazil, the statistics agency IBGE measures something similar in industry, among companies with 100 people or more. In 2024, 41.9% of them used AI in at least one function. Among those using it, 85.6% to 90.8%, depending on the function, use it in a localized rather than comprehensive way. It covers only industry, and IBGE itself classifies the survey as experimental.
Nobody wrote down when to stop
Without a stopping criterion, a pilot can go on indefinitely. In a Gartner survey of CIOs, only 31% have criteria for a pilot not to reach production, and only 17% for abandoning something already in production. It is self-reported, Gartner sells advisory services, and no study proves that having the rule improves results.
The workflow stays the same
Among the highest-return companies in McKinsey’s survey, nearly 3/4 redesigned workflows because of AI. Among the rest, 1/4. The top group is small and defined by its own result, so this is association and says nothing about cause.
The firmest test dates from 2024. Among 25 attributes, workflow redesign carried the most statistical weight on EBIT reported from generative AI. The same study ranks tracking KPIs as the practice with the most weight among 12, and fewer than 1 in 5 do it.
Without a baseline, nobody can decide
In the Drexel and Precisely survey, 71% of leaders say AI is aligned with company goals, and 31% have metrics tied to KPIs. These are different questions, and they don’t subtract. All of it is self-reported, and Precisely sells data integrity.
Write this down before the pilot starts:
- The week 0 baseline: which metric, measured by whom, on which dashboard.
- The stopping criterion: what result, by what date, and who can end it.
- The owner in the receiving area, and the workflow step that goes away.
- The measure of use: frequency by function, instead of activated licenses.
No study tests this order directly. It is our inference, and no number here is a Nebuloz result.