AI is moving from experimentation into everyday organizational decisions. In 2025, 88% of surveyed organizations reported regular AI use in at least one business function, while 60% of managers using algorithmic-management tools said the tools improved their own decision making. At the same time, worker surveys show strong resistance to handing dismissal and promotion decisions entirely to AI, and responsible-AI concerns remain substantial.
Table of contents
- How widely organizations use AI in business decisions
- Where algorithmic management is used
- What managers say about decision quality and skills
- Worker trust and high-stakes employment decisions
- Risks, transparency, and responsible-AI evidence
- How employers expect work decisions to change
How widely organizations use AI in business decisions
The McKinsey report The state of AI in 2025: Agents, innovation, and transformation found that 88% of surveyed organizations reported regular AI use in at least one business function in 2025. That was up from 78% in 2024, a 10-percentage-point year-over-year increase in the reported share. These are survey measurements, not population-wide estimates.
AI use is also spreading across multiple functions rather than remaining isolated in one team. More than two-thirds of respondents said their organizations used AI in more than one business function, and one-half said they used it in three or more functions. About one-third said their organizations had begun scaling AI programs across the enterprise.
AI agents were already part of many organizations’ plans, but scaling remained less common. In the 2025 survey, 62% of respondents said their organizations were at least experimenting with AI agents. Within that group, 23% said their organizations were scaling an agentic AI system in at least one enterprise function, while 39% said they had begun experimenting but were not scaling. No individual business function had more than 10% of respondents reporting scaled AI-agent use.
Reported financial impact was still limited for most organizations. Thirty-nine percent of respondents attributed some enterprise-level EBIT impact to AI, and most of those respondents reported that less than 5% of EBIT was attributable to AI. The result suggests that adoption is broad, while material enterprise-wide financial effects are not yet equally broad.
Where algorithmic management is used
Algorithmic-management data provides a useful workplace decision-support proxy, but it has an important limitation: the OECD report How widespread is algorithmic management in workplaces? measures software that automates or supports managerial tasks and notes that these tools are not necessarily AI-powered.
In the OECD’s 2025 survey of more than 6,000 mid-level managers in France, Germany, Italy, Japan, Spain, and the United States, 90% of U.S. firms had adopted at least one tool to instruct, monitor, or evaluate workers. The average adoption rate across the surveyed European countries was 79%. Adoption was 81% in France, 78% in Germany, 76% in Italy, and 78% in Spain. Japan was lower at 40%.
| Country or group | Adoption of algorithmic-management tools |
|---|---|
| United States | 90% |
| France | 81% |
| Germany | 78% |
| Spain | 78% |
| Italy | 76% |
| Japan | 40% |
| Surveyed European-country average | 79% |
The U.S. figure also reflected broader tool coverage. More than three-quarters of U.S. managers said their firms provided at least 10 of the 15 algorithmic-management tools measured. U.S. firms had an average 90% adoption rate across instruction, monitoring, and evaluation tool categories.
Across the surveyed European countries, instruction tools had a 69% adoption rate, ranging from 64% in Italy to 74% in France. Monitoring tools had a 67% adoption rate, ranging from 64% in Germany to 71% in France. Worker-performance evaluation tools were used by 35% of European firms, with country rates ranging from 27% in Italy to 40% in France.
Japan showed a different balance among tool categories. Monitoring tools had a 32% adoption rate, compared with 25% for instruction tools and 11% for evaluation tools. Across all countries in the survey, tools used to sanction poor performance had a 14% adoption rate, compared with 23% for tools rewarding good performance.
What managers say about decision quality and skills
The OECD findings connect algorithmic-management adoption with perceived changes in managerial work. Sixty percent of managers using these tools said they improved the quality of their own decision making. This is a reported managerial perception, not an experimental measure of accuracy or fairness.
The tools were also associated with higher perceived skill requirements. Sixty percent of managers reported an increased need for analytical skills because of algorithmic-management tools, while 3% reported a decreased need for analytical skills. Social skills were reported as increasingly important by 32% of managers.
Governance was common but varied in form and coverage. Nearly 90% of managers said their firms had at least one governance measure for algorithmic-management tools. Guidelines were implemented by between 46% and 91% of firms, depending on country. The range indicates that the presence of governance does not mean the same level of formal guidance everywhere.
These figures describe a pattern of augmentation rather than simple replacement. Managers often reported better decision quality and a greater need for analytical skills at the same time. The survey does not establish that the tools caused those outcomes, and it does not show whether reported improvements were consistent across different types of decisions.
Worker trust and high-stakes employment decisions
The OECD report The impact of AI on the workplace: Main findings from the OECD AI surveys of employers and workers reports worker and employer survey results collected in 2022. They should not be read as measurements of worker opinion in 2026.
Workers were particularly cautious about delegating employment decisions to AI. Fifty-seven percent of surveyed workers supported banning AI that would decide which workers were dismissed. A further 25% said such use should be allowed only with restrictions. Promotion decisions also drew strong opposition: 47% supported banning AI that would make promotion decisions.
Recruitment views differed by industry. Forty percent of finance workers supported banning AI that would make recruitment decisions, compared with 41% of manufacturing workers. The near-identical figures show substantial concern in both sectors, while the survey does not establish why respondents held those views.
Data collection was another source of concern. Forty-nine percent of finance workers and 39% of manufacturing workers said their company’s AI collected data about them or their work. Among workers reporting AI-related data collection, 62% in finance and 56% in manufacturing felt increased pressure to perform. Meanwhile, 58% of finance workers and 54% of manufacturing workers in that group felt too much data was being collected.
Employers saw both possible benefits and harms for workers with disabilities. They said AI would help workers with disabilities in 46% of finance firms and 50% of manufacturing firms. They also said AI would harm workers with disabilities in 8% of finance firms and 12% of manufacturing firms. These are employer-reported expectations from the 2022 survey, not observed outcome rates.
Risks, transparency, and responsible-AI evidence
The Stanford AI Index 2025 section on Responsible AI records both rising incident counts and expanding research attention. The AI Incidents Database recorded 233 reported AI-related incidents in 2024, and reported incidents increased 56.4% from 2023 to 2024.
Leaders surveyed in 2025 identified several responsible-AI risks. Inaccuracy was cited by 64%, regulatory compliance by 63%, and cybersecurity by 60%. These percentages measure concern among surveyed leaders; they do not estimate the probability that any particular AI decision will fail.
Transparency indicators also changed. Restricted tokens in actively maintained C4 domains rose from 5–7% to 20–33% between 2023 and 2024. The average Foundation Model Transparency Index score rose from 37% in October 2023 to 58% in May 2024. The two measures describe different aspects of information access and model transparency, so they should not be treated as interchangeable.
Research activity increased as well. Responsible-AI papers accepted at leading AI conferences rose from 992 in 2023 to 1,278 in 2024, an increase of 28.8%. More research does not by itself prove that deployed systems are safer, but it shows that responsible AI remains an expanding area of technical and policy attention.
How employers expect work decisions to change
The World Economic Forum’s Future of Jobs Report 2025, workforce-strategies chapter describes employer expectations for 2030. These figures are forecasts of what surveyed employers expect, not observed outcomes.
Overall, 86% of surveyed employers expected AI and information-processing technologies to transform their business by 2030. Expectations were even higher in specific industries: 97% of surveyed financial-services employers and 95% of surveyed electronics employers expected transformation by that time.
Employers also anticipated changes in how people work with AI. Seventy-seven percent planned to reskill or upskill existing workers to work more effectively alongside AI by 2030. At the same time, 41% expected to downsize their workforce as AI capabilities expand. Because these are employer expectations, the figures do not indicate how many jobs will actually be eliminated or how many workers will receive training.
Taken together, the statistics point to a decision environment with three simultaneous features: widespread organizational adoption, continued human concern about high-stakes delegation, and growing pressure for governance and skills. The strongest evidence supports measuring where AI or algorithmic tools are used, what decisions they influence, and how workers and managers experience those systems over time.