Research methodology · Updated March 2026

Two numbers. Two studies.

The 88% adoption figure and the 5% scaled-value figure are not results from one study.

2026 context · added 1 October 2026

The 5% is a dated 2025 figure, not a current statistic.

BCG’s 2025 survey of 1,250 senior executives classified 5% of companies as ‘future-built’ and 35% as scaling (BCG, 30 Sep 2025). Its Applied AI Index 2026, based on 1,330 leaders, classifies 7.5% as future-built and a further 41% as scaling—nearly half creating value (BCG, 30 Sep 2026). The surveys differ year to year, so the change is indicative rather than a precise trend, and neither figure means ‘only 5% get any value’.

The defensible conclusion

Reported AI use is widespread. Repeatable enterprise value remains uncommon.

The 88% adoption figure and the 5% figure come from separate 2025 studies with different samples and definitions; this is a dated synthesis (updated March 2026). BCG’s 2026 Applied AI Index now classifies 7.5% of companies as future-built and a further 41% as scaling. Alyve compares them as a directional signal of the gap between widespread AI use and uncommon enterprise-scale value—not as one survey, one denominator or a subtraction exercise.

What each number measures.

88%

Respondents reporting AI use in at least one business function.

McKinsey & Company, The State of AI 2025. 2025. 1,993 respondents; global online survey across 105 countries.

Companies BCG classified in 2025 as ‘future-built’ (substantial value from AI). BCG’s 2026 index reports 7.5%.

Boston Consulting Group, The Widening AI Gap 2025. 2025. 1,250 senior executives; global senior-executive survey.

Do not subtract 5 from 88. The studies do not share a sample, denominator or measurement threshold.

The broader synthesis

A strategic signal, not a new statistic.

Alyve’s synthesis covers 17 sources, 22,000+ respondents and material dated 2023–2026. It was updated March 2026.

Alyve reviewed academic, institutional and corporate research that described enterprise AI use, deployment, maturity, barriers or realised value.

Measures were retained with their original definitions rather than converted into a common denominator. Confidence labels reflect source quality, sample disclosure and methodological clarity in the synthesis register.

The 88% and 5% measures were placed side by side to expose a strategic pattern: broad reported use does not establish deep, repeatable organisational value.

The comparison informs Mark Cameron’s Wide, Not Deep diagnosis and the Learning Velocity argument. It is interpretation, not a new pooled statistical estimate.

Alyve produced the synthesis. Mark Cameron developed the strategic interpretation and connected it to Learning Velocity and Wide, Not Deep.

A useful contrast

Why can adoption be 88% in one source and 20% in another?

McKinsey asks managers about any AI use in at least one function. OECD measures formal investment or dedicated tools under national statistical standards. Neither is necessarily wrong. They measure different things.

Read the comparison with care.

01The two headline figures come from different surveys, populations and definitions; they must not be directly subtracted.
02Most inputs rely on self-reported executive or manager responses and may be affected by interpretation, recall and optimism bias.
03The source set spans different countries, industries, respondent roles and collection periods, so it is not a representative sample of all organisations.
04‘Using AI’, ‘in production’, ‘at scale’, ‘measurable impact’ and ‘material value’ are distinct thresholds, not interchangeable labels.
05Some research is produced or commissioned by technology vendors, consultancies or investors. Those incentives are noted where the published register identifies them.
06The synthesis reports association and prevalence. It does not establish that AI adoption causes financial performance.
07The 5% figure is BCG’s 2025 ‘future-built’ classification. It does not mean only 5% of organisations get any value, and it is not a current or timeless statistic: BCG’s 2026 index (1,330 leaders) reports 7.5% future-built and a further 41% scaling.
08The synthesis metadata states 17 sources, but its currently published bibliography identifies 10. The remaining seven require bibliographic completion before the register can be described as complete.

Method: Executive survey

Caveat: The published register does not provide geography or sample size.

Source ↗

Questions about the synthesis.

Do the 88% and 5% figures come from one study?

No. The 88% figure comes from McKinsey’s 2025 State of AI research, while the 5% figure comes from BCG’s 2025 Widening AI Gap research. They use different samples and definitions.

Can 5% be subtracted from 88% to say that 83% of organisations failed?

No. The figures do not share a denominator or measurement threshold. Their value is directional: reported use is widespread, while BCG’s stricter ‘future-built’ threshold was met by 5% in 2025 (7.5% in 2026).

What does the synthesis establish?

It supports a strategic diagnosis, not a causal claim: access and experimentation have spread faster than repeatable enterprise value. Leaders should examine what has changed in decisions, work and outcomes—not only whether AI is being used.