Artificial intelligence appears in almost every executive presentation, product roadmap and earnings call. A new study of S&P 500 filings suggests that operational adoption is much narrower than the language around it.
The researchers found that 11% of S&P 500 companies had deeply integrated AI into business processes by 2025. Another 10% used AI directly in producing goods or delivering services. Adoption had more than quadrupled from 5% in 2022, but technology companies accounted for roughly two-thirds of the deeply integrated group.
A stricter measure than tool access
The paper, posted to arXiv on July 9, uses annual SEC filings to distinguish material operational adoption from general statements about AI. The authors argue that 10-K disclosures provide a useful signal because public companies face legal constraints on materially misleading statements.
That method is more conservative than a survey asking whether employees use an AI tool. It is also imperfect. A company may deploy meaningful systems without describing them in sufficient detail, while filing language cannot reveal how well an implementation works. The study measures disclosed depth, not every instance of use or proven productivity.
The distinction nevertheless clarifies a persistent confusion. Giving thousands of employees access to a chatbot can happen quickly. Deep integration requires changes to data, software, workflow and accountability. Those changes are slower and more uneven across industries.
The productivity evidence is not settled
The researchers observed a J-shaped relationship between adoption stage and profitability: firms in the middle showed weaker performance before deeply integrated firms improved. They did not find differences in capital expenditure or productivity associated with adoption in the available data.
That should not be read as proof that AI produces no productivity gains. The study is observational, adoption may be too recent for results to appear and profitable technology firms may be better positioned to integrate AI in the first place. It does show why access metrics and vendor spending are insufficient measures of transformation.
Executives should ask a more demanding set of questions. Which process has changed? What data and decisions are now connected? Who owns the result? What failure rate is acceptable? What measurable outcome improved after the novelty wore off?
The competitive gap may not form between companies that use AI and those that do not. It may form between companies that add AI to existing work and those that rebuild a small number of important processes around it. The study suggests the second group remains rare—and therefore potentially consequential.
Sources for editorial review
Drafting note: This draft was prepared with AI assistance from the linked source material and requires author review, independent fact-checking and final editorial approval before publication.