Which usage targets can AI surveys justify?

An executive team reads that intensive AI users report benefits more often. A plausible idea follows: everyone should use AI more frequently. Between the observation and the instruction lies a decision that deserves its own reasoning. Which work should improve, and why would more frequent use help?

Marina Mogilko’s compilation of conversations with business leaders prompted this question. The publicly available episode, dated 18 August 2026, combines personal examples with figures about AI use. It provides a starting point for examining how a survey finding becomes an organisational instruction. [1]

As the founder of aiomics, I am interested in the conclusion a figure can actually support. An adoption rate can show whether an approved service reaches people. Deciding how to extend its use raises further questions: which service changes, for whom and under what conditions?

A percentage can describe different things

On 23 June 2026, BCG writes that 15% of CEOs obtain meaningful value from AI. It associates this group with at least eight hours a week developing their own AI capabilities. BCG also explicitly cautions that hours alone do not create value. [2]

The linked January survey covers 2,360 executives, including 640 CEOs, across 16 markets. Roughly 15% also identifies a leadership segment that BCG classifies by its outlook and approach. That classification is not an independent audit of profits across all companies. [3]

On 22 July, BCG reports nearly nine in ten CEOs seeing some cost or revenue benefits from AI in targeted areas. This concerns initial, localised benefits. These differently framed statements do not establish an increase from 15% to 90%. [4]

For a decision paper, I would therefore attach a complete sentence to the number: who answered which question, concerning what period and what kind of benefit? Without that sentence, it remains unclear whether we are discussing an expectation, a personal assessment or a verifiable economic result.

What a reported benefit measures

Gallup’s May 2026 US survey finds positive productivity assessments among 45% of AI users reporting one or two purposes, compared with 90% reporting seven or more. These are self-reported assessments of impact. Gallup expressly cautions against interpreting the association causally. [5]

A simple calculation clarifies the unit. If nine out of ten people say a change helped, we know the proportion giving that answer. We still do not know how many additional cases the group completed. That would require a different measurement. Both questions may be interesting together; their results have different meanings.

Nor does a requirement to use AI for seven purposes follow. For a local investigation, I would first ask whether an additional use fits an upcoming task. Someone could use AI for one valuable purpose every day. Another person could try many uses without producing a usable result. These are possible cases that a simple count does not distinguish.

Which workforce is in the denominator?

Consider a wholly fictional medical technology company with 120 employees. Forty have been trained and authorised to use an AI tool for a particular research task. During the week under review, 32 of them completed that task with the tool. The other 80 employees are currently outside its intended user group.

Dividing 32 by 40 gives 80%. Dividing 32 by 120 gives approximately 27%. Both calculations are correct. The first describes use within the intended group. The second describes reach across the workforce. Neither establishes whether the research became more complete or a decision improved.

Suppose the executive team now sets a usage target of 70% across the workforce. That would mean 84 people. It would extend the existing user group. Suitable tasks, access and preparation for that extension remain to be established. A seemingly small change to a metric contains a larger organisational decision.

In a reporting table, I would therefore spell out numerator and denominator: 32 of 40 authorised people, during this week, for this research task. The reach of 32 among 120 employees can sit alongside it. Both remain visible. This makes it harder to inflate a limited deployment through a favourable denominator or dismiss it prematurely through an unsuitable comparison.

A short check before the next target

I propose recording four things for any survey figure used to justify an organisational decision. This is my own proposed method; its effectiveness has not been evaluated here.

First, the exact finding. Record the question, response option, reference population, fieldwork period and source. A shortened headline will rarely supply all of these.

Second, the intended action. Is the figure meant to justify additional training, wider authorisation, a purchase or a usage-frequency target? Write down the action in concrete terms.

Third, the missing assumption. Explain why that particular action might improve the intended result in your organisation. For example, investigate whether limited practice constrains use or whether the available application is poorly suited to the task.

Fourth, the observation for the next decision. State what would help you assess whether the assumption holds. The linked pilot evaluation guide provides a separate procedure for assessing a bounded trial. The question here comes earlier: the reasoning for beginning that particular trial.

A carefully read survey can raise a useful question about your own company. For the next meeting, select one figure and translate it into these four entries. Wherever the transition from finding to action remains unclear, there is a question for the next investigation.