The adoption
gap.
Why 89% of companies run AI, 6% can find it in earnings, and what separates the two at every scale from $1M to $1B+. The studio’s report — every figure classed delivered, modelled or external, never summed.
Published · Updated
AI adoption is saturated: 89% of organisations use it regularly, 37% can attribute any earnings impact, 6% more than five points — and both attribution numbers are flat against rising spend. The gap is not model quality. The companies capturing earnings impact are around three times more likely to have rebuilt the workflow around the AI rather than bolting AI onto the workflow they had.
The report maps the seven layers of every AI estate, follows the curve stage by stage from $1M to $1B+, sets out what changes as generation gets cheap and verification does not, and closes with the governed model, the European overlay — attribution, the AI Act, the funding routes — and a one-day estate audit as the starting move.
Key findings
-
89% of organisations use AI regularly in at least one function; 37% can attribute any earnings impact to it; 6% attribute more than five points of EBIT. Both attribution figures are flat year on year, against rising spend.
ExternalMcKinsey, The State of AI 2026
-
Above $1B revenue, 40% of organisations are scaling agents inside at least one function, up from 27%. Below $1B the figure is flat at 22%.
ExternalMcKinsey, The State of AI 2026
-
20% say AI operating costs now constrain deployment: token spend has become a budget line.
ExternalMcKinsey, The State of AI 2026
-
Failed programmes buy the visible interface layer and skip the data and integration layers underneath, then conclude the model was not good enough.
Method
-
The window in which disciplined remediation of ungoverned AI code is scarce, and therefore expensive, runs roughly 2026 to 2028.
Modelled
-
Under the EU AI Act, general application and the Article 50 transparency duties have applied since 2 August 2026; substantive high-risk obligations were deferred to December 2027 and 2028.
ExternalRegulation (EU) 2024/1689 and the 2026 omnibus agreement, as at September 2026
McKinsey State of AI 2026 · 1,719 respondents, 97 countries · attribution by executives, not audited accounts.
- 1
Executive Summary
Adoption is saturated; earnings impact is flat. The constraint was never the model — it is whether anyone will change the process the model sits inside.
- 2
The Interface Illusion
Why buyers purchase the visible layer and skip the decisive ones, with a worked example from the public record.
- 3
Why the Returns Don’t Arrive
The market in eight numbers, why saved hours never reach the accounts, and the diligence view of a failing estate.
- 4
The Curve by Scale
What gets deployed at each size of company, how it fails, and where the money is.
- 5
What Changes Underneath
The next three years: generation is cheap, verification is not. The gate moves upstream. A forecast, labelled as one.
- 6
The Governed Model
Why the category sells the wrong layer, the five guard-rails and the gate, and where the discipline has held.
- 7
The European Overlay
Making the impact attributable, the AI Act position at September 2026, and the funding routes most deployers never claim.
- 8
The Starting Move: The Estate Audit
Map the estate against the seven layers, answer three questions in writing. It takes a day, and the third question is usually the finding.
Read the full report.
You already have access.
This browser has already been given access to our papers. No need to enter your email again.