This session examined how rapidly artificial intelligence is spreading across firms, why adoption remains uneven, and how boards, management practices, organisational design and firm resources shape whether AI supports efficiency, productivity and more exploratory innovation.
The session brought together international survey evidence and three firm-level studies to explore both the pace and the character of AI adoption. The opening evidence showed that reported use is already widespread and rising quickly, but that realised employment and productivity effects remain modest compared with firms’ expectations for the next few years. Across the papers, the central message was that AI is not a stand-alone technology: its value depends on complementary capabilities in governance, management, data, skills and organisational coordination. Adoption is also highly uneven, with larger, more productive and better-resourced firms generally better placed to invest, experiment and capture returns.
The opening presentation drew on repeated business surveys across a growing group of countries, coordinated largely through central banks and covering almost 9,000 responses. It reported that around 78% of firms were using at least one form of AI, up from roughly 69% six months earlier. Use was concentrated among larger, more productive and higher-paying firms, and was also more common in younger firms and those with younger directors. AI appeared to be diffusing faster than earlier general-purpose technologies, although adoption levels and intensity varied across countries. US firms were reported to spend substantially more on AI per employee than UK firms, while senior executives were increasingly using AI directly.
The evidence suggested that employment effects to date have been small, but firms expected modest reductions in employment over the next three years, concentrated in sectors such as finance, insurance, professional services and science. Realised productivity effects were also limited so far, while expected gains were considerably larger. Indicative calculations pointed to a potential contribution of around 0.4 percentage points a year to total factor productivity over the next three years, subject to important assumptions. The presentation stressed the need for frequently updated data because adoption is moving too quickly for older surveys to provide a reliable picture.
The first paper examined whether the technological experience and networks of company directors help explain differences in AI adoption among US firms. It combined board-level information, firm accounts and AI-related job-posting data for approximately 3,000 firms between 2010 and 2018. The authors constructed an AI board index based on four dimensions: directors’ exposure to technology firms, the breadth of that experience, experience in technical functions such as engineering or technology leadership, and experience in leading technology firms.
The results showed a strong association between the board index and AI-related hiring, including outside the technology sector. Breadth of experience and exposure to frontier technology firms appeared particularly important. Preliminary analysis of unexpected director departures was presented as a possible route towards stronger causal identification, with the loss of technically experienced directors associated with reduced AI adoption. Discussion highlighted the need to distinguish executive from non-executive directors, examine resignations as well as deaths, test alternative definitions of frontier firms, and consider whether board diversity and governance status affect the relationship.
The second paper used the Office for National Statistics Management and Expectations Survey to investigate why AI adoption remains patchy across UK businesses. It compared AI with other technologies, including cloud computing, robotics, specialised software and specialised equipment, and linked 2023 adoption responses to earlier information on management practices. The analysis focused on firms for which a technology was relevant, distinguishing those already using it in business processes from those that had not yet adopted it.
Better-managed firms were more likely to adopt AI, and the relationship between management practices and adoption appeared stronger for AI than for most other technologies. Target setting, key performance indicators and monitoring practices were especially important, while decentralised product development was also positively associated with adoption. The paper suggested that AI may require iterative implementation, data readiness and coordination across an organisation rather than a one-off installation. Discussion raised the importance of formal AI policies, unrecorded employee-led use, independent governance and oversight, data management capabilities, and the possibility that the findings reflect AI’s current stage of diffusion as well as its technological characteristics.
The third paper examined how firms’ development of AI capabilities affects both the volume and direction of innovation. Using AI-related job postings as a measure of organisational commitment and patent data to track inventive activity, the study distinguished exploitation of familiar technological domains from exploration into areas that were new to the firm. The pre-generative-AI setting provided a cleaner view of how predictive AI was being incorporated into firms’ innovation processes.
Firms developing AI capabilities produced more innovation overall, but the increase was concentrated in familiar areas, consistent with AI acting as a “streetlight” that improves search where firms already possess data and expertise. The organisational location of AI skills mattered: adoption through managerial roles was associated mainly with efficiency and exploitation, whereas AI capabilities in scientific and engineering roles were more closely linked to exploratory innovation. Larger firms benefited more strongly, reflecting their greater data, skills and financial resources, while volatile market conditions encouraged firms to use AI for forecasting, control and risk reduction. The findings raised concerns that AI may reinforce existing knowledge paths and shift innovative capacity towards already dominant firms unless organisational design and policy deliberately support experimentation.