Methodology

The largest source is the Oliver Wyman Forum Global Voices survey. This report uses two waves fielded in 2026: April (N=12,883) and August (N=12,837). Both waves covered the same 15 markets: the United States, Canada, Mexico, Brazil, the United Kingdom, France, Germany, Italy, Spain, South Africa, Saudi Arabia, the United Arab Emirates, India, China, and Hong Kong (reported separately from mainland China). Each market had roughly 900 respondents, except Saudi Arabia and the UAE at roughly 500 each.

Respondents were adults age 18 to 75, surveyed online in local languages, with quotas on age and gender. The survey focused on the employed population; about 96% were employed, so findings for “consumers” or “all respondents” reflect a predominantly working population rather than the general public.

The surveys were sourced from a panel of 67 million people worldwide. To support representative distributions, our respondent pool generally mirrored the demographics of each country, including age, income, education level, political affiliation, and gender. In a small number of waves, certain demographic segments may have been modestly over- or underrepresented due to sampling constraints; where this occurred, results were still intended to be broadly representative at the country level. We also applied standard quality controls to remove low-quality responses.

Each wave combined fixed questions on demographics, employment, and attitudes with modules on AI in the workplace, governance and risk, shopping, and sovereignty. Many questions were asked only of relevant subsets, such as employed respondents, workplace AI users, or white-collar employees by job level. Some analyses also excluded “not sure” responses. Bases therefore varied across exhibits, and the same question could produce different shares depending on the population analyzed.

Throughout the report, we generally used global findings across the 15 markets rather than presenting country-by-country results. Aggregation may smooth over meaningful local differences in attitudes, adoption, regulation, or market context. We reviewed the major findings for directional consistency across markets.

In most cases, while magnitudes differed, the directional story held across the markets analyzed. Even so, this is global research rather than a comprehensive assessment of every local market, and some country-specific differences may not be fully captured.

CEO findings come from the Oliver Wyman Forum x NYSE CEO Agenda 2026, an online survey of 415 CEOs and CEO-equivalents fielded January to March 2026. Respondents were recruited through the NYSE’s network, the Oliver Wyman Forum’s network, and specialist research partners. By region, 56% of respondents were based in North America, 32% in Europe, 9% in Asia-Pacific, and 4% elsewhere. By revenue, 45% led companies with under $1 billion, 31% with $1 billion to $5 billion, and 24% with over $5 billion. Public companies made up 63% of the sample and private companies 36%, across 10 industry groups.

Year-on-year comparisons draw on the 2025 edition (165 CEOs, March to April 2025). That sample was predominantly North American and used different question wording, so these comparisons are directional only. The report also draws on the Oliver Wyman 2026 Parallel Surveys of CEOs and Tech Leaders on Agentic AI and IT (200 respondents, July to August 2026).

Several survey findings rely on segments we constructed for analysis. These include “AI superworkers” (workplace AI users who say AI at least doubles the work they can complete), companies that employees describe as ahead of or behind competitors on AI adoption, and CEOs grouped as AI ROI leaders or laggards based on the financial impact they report.

Beyond survey data, the Oliver Wyman Forum Sovereign AI Index ranks the 100 largest economies across six dimensions of AI capability, drawing on sources including the Oxford Insights Government AI Readiness Index. Three proprietary analyses of AI shopping inform the findings on agentic commerce: simulated recommendations, a product-page audit, and simulated purchase journeys. These are illustrative samples, and AI outputs can vary with model, prompt, and timing. The 2030 estimate of AI-influenced e-commerce is a directional scenario based on survey adoption data and market growth assumptions. Conceptual frameworks and illustrative exhibits labeled “Oliver Wyman analysis” represent our perspective and are not empirical estimates.

Finally, parts of this report are informed by conversations with senior executives and by secondary research used to contextualize and triangulate our findings. Secondary sources include Challenger, Gray & Christmas layoff reporting, Stanford Digital Economy Lab research using ADP payroll data, the UK Competition and Markets Authority’s cloud services market investigation, and public reporting on market developments.

Survey responses were subject to standard quality checks. Results were self-reported and descriptive unless otherwise noted, and percentages may not sum to 100 because of rounding or multi-select questions. Readers with questions about bases, markets, or definitions should please reach out.