Enterprise AI adoption is spreading rapidly among consumers but delivering little measurable improvement to the finances of most large companies, according to McKinsey’s 2026 State of AI survey. While 88% of organisations use AI in at least one business function, only 6% report an earnings impact of more than 5%.
The analysis behind the figures argues that the central obstacle is not a lack of powerful technology, but the failure of businesses to redesign their operations around it. The result is described as an “Innovation Paradox”: faster technological progress has coincided with weaker productivity growth.
Consumer uptake has far outpaced corporate adoption. Smartphone use reached 80% of Americans within seven years, while enterprise resource-planning systems took 33 years to reach 57% of US firms. Consumer AI use has reached 53% of American adults less than three years after ChatGPT’s launch, but only 10% of businesses are using it in productive deployments.
There is also a sharp divide among companies that have adopted AI. Only 21% of adopters have redesigned workflows from end to end, yet the small group of high-performing organisations is three times more likely to have done so.
That split is reflected in financial markets. Thirty-six AI companies in the S&P 500 now account for 45% of the index’s market capitalisation. Over three years, the index returned 76%, but the return fell to 32% when those AI-linked companies were excluded.
Why AI adoption is failing to transform businesses
The analysis contrasts the current technology cycle with the spread of electricity and telecommunications in the early twentieth century. Those innovations reached consumers and businesses at broadly similar speeds and forced companies to reorganise how they worked.
Electric motors encouraged factories to move away from rigid layouts towards more flexible production, while telephones reduced the cost of coordinating activities. New management structures followed, including the multidivisional model associated with General Motors and the scientific-management approach linked to Frederick Taylor.
Digital technologies have entered a different kind of economy, in which services make up a much larger share of output. Hospitals, schools, government agencies and construction firms do not have the same supply-chain connections as industrial manufacturers, making it harder for improvements in one part of the economy to spread through other sectors.
These areas are also heavily dependent on human judgement and interaction. The analysis points to economist William Baumol’s observation that sectors such as healthcare, education, public administration and construction can remain resistant to productivity gains. Together, they account for roughly half of the GDP of advanced economies.
The report argues that AI could nevertheless be significant in such industries because it is designed not merely to process or transmit information, but to assist with reasoning, decisions and creation. Those capabilities could be relevant in fields where diagnosis, evaluation and personalisation are central.
However, organisational complexity can prevent technology from being used effectively. A proposed measure called the Combined Overhead Ratio combines government spending, corporate selling, general and administrative costs, and regulatory compliance.
In the US, the ratio is said to have passed 47% of GDP around 2000. The analysis links levels above that threshold with annual GDP growth that has not sustainably exceeded 2.5%, compared with roughly twice that rate when overhead was below 35% in the 1960s.
Government spending is now put at 36% of GDP, while corporate SG&A costs have doubled since the 1980s. The Competitive Enterprise Institute estimates that US companies spend more than $2 trillion a year on compliance.
An assessment of S&P 500 companies between 1985 and 2025 found that 73% of sector-by-decade observations fell into what the analysis calls an “overhead trap”, where SG&A costs rose as revenue growth declined.
Jack Dorsey, the founder of Block, has described corporate hierarchy as an “obsolete information-routing protocol”. The analysis uses the phrase to argue that many businesses are adding AI to existing structures without changing the processes that govern how decisions are made.
Businesses face resistance to organisational change
Examples from consumer businesses and healthcare illustrate the problem. In one case, digitisation produced strong online growth but failed to spread through the much larger offline operation because independently owned distributors would not hand over customer data to an integrated platform.
Healthcare has faced a different set of barriers. Digital tools can be introduced to doctors relatively easily, but creating fully connected clinical workflows is more difficult because of data governance, compliance requirements and the influence medical professionals can exert over management.
A 2026 survey by Writer and Workplace Intelligence of 2,400 global leaders found that 79% of organisations faced difficulties adopting AI, an increase of more than 10 percentage points from the previous year. Some 54% of C-suite executives said adoption was “tearing their company apart”.
The analysis concludes that the speed of consumer adoption is a poor guide to the economic impact of new technology. AI is advancing more quickly than previous innovations, but the institutional changes needed to use it effectively are likely to take years or decades.
Its wider effect will depend on whether AI becomes embedded in core business operations and whether that transformation reaches beyond a small group of successful companies into the major sectors that account for most employment and economic output.
