This chapter draws together diverse types of information on the diffusion of artificial intelligence (AI) at national, sector and firm levels. It concentrates on AI in manufacturing and information and communication technology, the same sectors examined in the OECD/Boston Consulting Group/INSEAD survey conducted in 2022-23. This review provides an evidence base against which to assess the survey findings (presented in Chapters 1, 3 and 6). Most of the prior evidence indicates that, at least prior to the advent of generative AI, the adoption of AI in firms is an exception rather than the norm. Single-digit adoption rates for entire sectors are common in many countries. A universal finding is that adoption is highest in larger firms. The chapter also shows discrepancies in adoption rates across countries. More work is needed to understand the reasons for these divergences, which, among other things, are likely to reflect methodological issues in measurement.
The Adoption of Artificial Intelligence in Firms
Abstract
Prior survey evidence has compared internal and external barriers to AI adoption, and shows that despite many commonalities, national differences also exist. An issue that merits further examination, given its policy ramifications, is why firms in many studies (but not all) indicate that cost is a barrier to adoption, and demand more public financial support for adoption. The enterprise-level interviews discussed in chapter 5 examine this topic further.
Prior studies also underscore the role of digital readiness as a condition for adopting AI. An adoption hierarchy exists whereby digital-intensive firms tend to apply AI in deeper ways. A six-stage model of the readiness journey – used to inform a survey exercise in the United States – is outlined in this chapter. This chapter also presents evidence rarely commented elsewhere on the importance of competition in the AI vendor market. More competition among AI vendors could induce technology providers to maintain high levels of operational quality, lower prices for customers, develop specialisations, serve previously underserved industries, and make technology available to a wider range of markets. Relative to the United States, the European Union may be lagging in terms of numbers of AI vendors. Data on venture capital (VC) investment also show that AI vendor firms’ participation in VC deals is significantly lower in Europe than in the United States.
This chapter begins by reviewing data on the extent of AI adoption and diffusion among enterprises in the Group of Seven (G7) countries, the European Union and Brazil. It next assesses the barriers to greater AI adoption and diffusion in these countries.
This section examines the available literature on AI adoption in enterprises across several countries based on data provided by national statistical agencies, the US federal government, and the European Union. Overall, it finds significant diversity in the state of AI deployment across countries. Overall, however, the extent of AI uptake across firms is relatively limited. A further and consistent finding is that larger enterprises significantly outpace smaller ones in deploying AI technologies.
Canada
Canada’s 2019 Survey of Innovation and Business Strategy addressed a sample of firms and industrial non-profit organisations with at least 20 employees and CAD 250 000 (Canadian dollars) in revenue (Canada, Statistics, 2019[1]). Responding firms were sorted into 1 of 14 sectors (by the North American Industry Classification System [NAICS] code) and 1 of 3 size classes (20‑99 employees, 100‑249 employees, and 250+ employees). Questions covered the period 2017‑19. The questionnaire was designed so that definitions of the innovation concepts used were consistent with those used by the OECD and Eurostat. Participation in the survey was mandatory under Canada’s Statistics Act.
The three leading industries in Canadian AI adoption were: 1) information and culture (18% of all firms); 2) finance and insurance (21%); and 3) professional, scientific and technical services (21%). The largest Canadian firms have the highest rates of AI uptake, as in other countries.
Firms in leading AI adopter industries report a much greater shortage of computer science, information technology (IT) and general data science and analytics skills than in other industries. Shortages reported in the manufacturing sector were nearly identical to those for all industries. Excluding finance and insurance, the greatest shortage was of computer scientists, followed by persons with skills in general data science and analytics, and then IT.
Japan
Japan’s Ministry of Internal Affairs and Communications (MIC) compiled its Communication Usage Trend Survey (CUTS) in 2020 (MIC, 2021[2]). The survey aimed to gauge the development of information and communication networks and trends in information and communication technology (ICT) adoption. The 2020 CUTS received responses from 6 017 companies. Surprisingly, the share of all firms (with at least 100 employees) using AI and Internet of Things (IoT) fell from 14% to 12% from 2019 to 2020 (see Figure 2.1). It is striking that the share of firms reporting that they do not know whether they will adopt these technologies halved between 2018 and 2020 (from 16% to 8%). The share of firms reporting that they have not and do not plan to adopt AI and/or IoT increased from 63% in 2018 to 70% in 2020.
Source: MIC (2021[2]), 2020 Communication Usage Trend Survey, Ministry of Internal Affairs and Communications, http://www.soumu.go.jp/johotsusintokei/statistics/pdf/HR202000_002.pdf.
Manufacturing, information and communications, finance and insurance, and real estate have above-average AI adoption rates in Japan. Unsurprisingly, the finance and insurance industry leads all industries in both the share of firms using AI and IoT and those that do not use these technologies but plan to (see Figure 2.2). Overall, the survey indicates that 12.4% of Japanese firms use AI and/or IoT. Again, a clear positive relationship exists between the likelihood of using AI and IoT and firm size. The use rate in the 100‑299-employee size class is half the rate of the group with 1 000 to 1 999 employees (10% and 22%, respectively). Similarly, the rate of AI use in the group of firms with 1 000 to 1 999 employees is less than half of that in firms with 2 000 or more employees (at 22% and 48%, respectively) (see Figure 2.3).
Some 69% of Japanese firms use cloud computing to some degree (up from 58% in 2018). Between 2018 and 2020, the share of firms not using cloud services and not planning to fell from 21% to 16%. Unsurprisingly, usage of cloud services is most common among Japanese firms in ICT (92%), followed by firms in the real estate and finance and insurance industries (86% and 81%, respectively). Japanese manufacturing has a slightly below-average share of firms using cloud computing services at 68% compared to the 69% economy-wide average. However, manufacturing also has an above-average share of firms not using cloud computing but planning to (12% compared to the average of 10%). More than half of Japanese firms report a shortage of ICT-related human resources (e.g. computer programming or data science skills). Only 15% of firms report that their current workforce has enough of these skills. The largest ICT skills shortage in Japan is for network operators. Across all Japanese industries, over 60% of firms reporting a shortage of human resources in ICT report a shortage of network operators. In all industries except for finance and insurance, the second-most prevalent shortage is of systems engineers, followed by data scientists.
Source: MIC (2021[2]), 2020 Communication Usage Trend Survey, Ministry of Internal Affairs and Communications, http://www.soumu.go.jp/johotsusintokei/statistics/pdf/HR202000_002.pdf.
Source: MIC (2021[2]), 2020 Communication Usage Trend Survey, Ministry of Internal Affairs and Communications, http://www.soumu.go.jp/johotsusintokei/statistics/pdf/HR202000_002.pdf.
United Kingdom
A 2022 survey conducted for the United Kingdom’s Department for Digital, Culture, Media, and Sports queried 2 019 businesses in England, Scotland and Wales about their current and planned adoption of AI (Evans and Heimann, 2022[3]). Participants were asked about their usage or planned use of the following technologies: 1) robotic processes automation; 2) machine learning; 3) natural language processing and generation; 4) data management and analysis; 5) computer vision and image processing; and 6) hardware related to AI.
As in other countries, AI adoption rates are strongly associated with firm size. The survey found that 15% of small firms adopted AI, compared to 34% of medium-sized firms and 68% of large firms. Compared to those adopting AI technology and those planning to do so in the future, the share of firms in the piloting stage was quite small. Only 2% of small firms, 5% of medium-sized and 9% of large firms were piloting AI (see Figure 2.4). Larger firms that have adopted AI technology are also more likely to have adopted multiple AI technologies. Notable in this survey is the high adoption rate among medium-sized and large firms compared with survey findings in other countries. In general, the size of such disparities between a number of cross-country surveys suggests that different methodologies or sampling frames limit comparability (see the discussion below). In such cases, the most useful insights from the national survey relate to inter- and intra-sectoral findings (such as, for example, differences in adoption by firm size and age).
Source: Evans A. and Heimann A, (2022[3]), AI Activity in UK Businesses Report, Capital Economics and DCMS, January 2022, http://www.gov.uk/government/publications/ai-activity-in-uk-businesses.
According to the same survey, data management and analysis is the most common AI-related technology adopted in the United Kingdom for all firm sizes, used by more than half of AI-adopting firms in each size class. In fact, the order of the five technologies by share of AI-adopting firms using them is the same for all three size classes (small, medium, large), with natural language processing (NLP) and natural language generation being the second-most-common use, followed by machine learning (ML), computer vision and image processing/generation, and then hardware.
From a sectoral perspective, the United Kingdom’s legal sector has the highest adoption rate and share of firms planning to use AI technology. This is followed closely by the IT and telecommunications sector, with approximately three in ten IT and telecommunications firms having adopted AI. Manufacturing has the sixth-highest adoption rate, with fewer than one in five firms being current adopters and only 14% planning to adopt AI technology.
In total, AI-adopting firms in the United Kingdom invest the equivalent of around 9% of turnover on AI. Some 73% of those expenditures (6.6% of turnover) go toward AI-related labour, and the other 27% (2.4% of turnover) go to AI technologies. Interestingly, AI expenditures as a share of turnover are not strictly related to firm size. The study’s authors suggest that medium-sized AI-adopting firms in the United Kingdom spend so much more of their turnover on AI because they are more likely to develop AI technologies in-house than small and large AI adopters.
United States
Findings from the National Science Foundation
At writing, the National Science Foundation’s (NSF) 2020 Annual Business Survey (ABS) (National Science Foundation, 2021[4]) is the most current and comprehensive assessment of AI adoption in US industry. The technology module of the survey contained three detailed questions regarding: 1) the availability of information in digital format (digitalisation); 2) expenditures on cloud computing services; and 3) the use of several advanced “business technologies”, including a number typically categorised as AI, including: augmented reality, ML, machine vision, NLP, voice recognition software, robotics and automated vehicles (McElheran et al., 2021[5]).
Overall, the US ABS data show that AI adoption across almost all US industries remains very low. An analysis of the data by Zolas et al. (2020[6]) finds that across AI-related technologies for all firms in the US economy, the aggregate AI adoption rate was 6.6%. The ABS data show that the adoption of the mentioned AI technologies ranges between 83.2% to 85.8%. Some 89% of US manufacturers report not using AI at all. In fact, in key manufacturing industries such as machinery, electronic products and transportation equipment, fewer than 12.4% of companies report using AI as a production technology in any capacity. Reported use of AI by US companies in non-manufacturing sectors was likewise low. For instance, only 3.2% of US enterprises in professional, scientific, and technical services reported using AI. Only 2.2% of companies in the finance and insurance sector used AI. For firms in healthcare and social assistance, and educational services, the adoption rates were 5.7% and 2.0%, respectively.
While overall adoption of AI is low, the US ABS data show that larger companies are the leading AI adopters. The 2019 ABS surveyed 850 000 nationally representative firms on the use of AI as a production technology, receiving 590 000 responses for the period 2016‑20 (NSF, 2019[7]). More than 25% of the largest companies use AI tools to create high-quality goods and services, compared with only 3‑4% of small and medium-sized enterprises (SMEs) (see Figure 2.5). The results show that smaller US enterprises are lagging in the utilisation of advanced technologies, which is concerning when companies with under 500 employees contribute around 43% of US gross domestic product (GDP) (Kobe and Richard, 2018[8]).
Several reasons help explain why larger firms adopt AI more than smaller firms. For one, ICT adoption is higher in larger firms, and AI adoption relies on advanced use of ICT. For another, because large firms tend to serve large markets, they can better amortise the fixed costs associated with employing AI production technologies over more sales, lowering the unit costs of production. Furthermore, because a share of the talent needed to harness AI is foreign-born, larger companies can better afford the time, fees, and personnel resources inherent in the US visa process to attract AI workers. Larger firms also offer higher wages and more benefits, increasing the pool of top AI talent these firms can access. Finally, because vendors of AI systems benefit from supplying companies with a large consumer base, vendors may focus on creating relationships and contracts with larger firms, helping these firms better understand the value that AI systems can bring to their businesses. Zolas et al. (2020[6]) also argue that, until recently, with the greater use of cloud computing (a problematic topic for some of the enterprises surveyed by OECD/BCG/INSEAD, as discussed in Chapter 3), the extensive computing power required for large-scale AI applications was beyond the means of most firms, making AI more feasible for larger firms. Beyond firm size, Zolas et al. (2020[6]) also noted a relationship between firm age and AI use. For small firms (here, meaning those with fewer than 50 employees), use rates tended to decline with age, with the oldest firms having the lowest adoption rates, suggesting that it may be the “new, young, born-on-the-web firms” that are the main AI users. However, for larger firms (here, meaning those with over 50 employees), use rates exhibited the opposite pattern: as firm age increased, usage rates also increased, with the highest usage rates found in the oldest and largest firms. Overall, firm size appears to be a significant predictor of firms’ AI use (Fleming, 2023[9]).
Note: Numbers in parentheses represent the number of employees per company.
Source: NSF (2019[7]), Annual Business Survey: 2019 (Data Year 2018), National Science Foundation, https://ncses.nsf.gov/pubs/nsf22315.
Findings from the US Patent and Trademark Office
The US Patent and Trademark Office (USPTO) (2020[10]). studied the volume, nature and evolution of AI and its component technologies using US patents from 1976 through 2018. The report describes an AI patent landscape over that period. The numbers presented in the USPTO study provide insights into the diffusion and adoption of AI in firms, even though patenting activity includes more than just firm-based innovations. When the USPTO examines a patent application, it reviews its technical content and assigns the patent to a specific technological grouping that has more than 600 subclasses covering a vast array of technologies. Key findings from the USPTO research include that, in the 16 years from 2002 to 2018, annual AI patent applications increased by more than 100%, rising from 30 000 to more than 60 000 annually, and that over the same period, the share of all patent applications that contain AI grew from 9% to nearly 16%. Moreover, the USPTO found an increasing “diffusion of AI across patent technology subclasses,” essentially referring to increases in patenting of discrete applications of AI such as NLP and ML. The USPTO found that, in 1976, patents containing AI appeared in about 10% of the subclasses, but by 2018, they had quadrupled to spread to more than 42% of all patent technology subclasses. These figures suggest a broad and deepening engagement with AI technology within the business sector, not only in terms of quantity but also in the diversity and sophistication of applications.
The USPTO report also offers data on the diffusion of AI patents across geography, finding that while the leading metro areas like the San Francisco Bay Area, southern California, and the Northeast Corridor still lead, the data since 2001 show that AI technologies are diffusing widely across US states and counties. Lastly, the report notes that several of the leading AI-patent-receiving firms today are not just the AI-technology producers (like IBM or Microsoft) but firms such as Bank of America, Boeing, and General Electric that are developing their own unique AI tools for their specific needs. A separate study suggests that US enterprises’ AI adoption has accelerated recently, in part as a response to the coronavirus (COVID‑19) pandemic.1
Europe
The most statistically representative data on AI adoption in European countries is the Eurostat database, with a sample size of 142 000 representatives of enterprises and a median 80% response rate (Eurostat, 2022[11]). Eurostat data yield insights on the level of technology use in companies and the state of workforce preparedness for AI. France, with about 47% of firms with very low digital intensity, not only underperforms Italy and Spain (each at 39%) and Germany (38%) but also the EU average (41%) (see Figure 2.6, Panel A), which could pose challenges for the shift to more productive operations using AI. Unsurprisingly, the ICT sector is much more digitised than European economies overall (Panel C), as IT requires digitisation to function effectively. The telecommunications sector enables 5G mobile networks that can enhance the benefits from AI and accelerate adoption. France’s telecommunications sector, with 10.7% of firms having very low levels of digitisation, is lagging major economies such as Germany (5.1%) and Italy (9.8%), as well as the EU average (8.8%). Only 22% of firms in the European Union have high or very high values for digital intensity.
Percentage of enterprises with ten or more employees
Note: The Digital Intensity Index (DII) is a composite indicator derived from the Survey on ICT Usage and E-commerce in Enterprises. The DII is one of the key performance indicators in the context of the Digital Decade, which sets out Europe’s ambition on digital, laying out a vision for the digital transformation and concrete targets for 2030 in the four cardinal points: skills, infrastructures, digital transformation of businesses and public services. The 2030 target of the Digital Compass is that more than 90% of EU SMEs should reach at least a basic level of digital intensity. The indicator is useful to describe the extent to which EU enterprises are digitalised. It measures the use of different technologies by enterprises. The different definitions of the DII can be found at https://circabc.europa.eu/ui/group/89577311-0f9b-4fc0-b8c2-2aaa7d3ccb91/library/84b390d2-6a83-4dae-8aba-37c18557eb5b/details.
Source: Eurostat, Digital intensity by NACE Rev2 activity, ISOC_E_DIIN2, Percentage of enterprises, https://ec.europa.eu/eurostat/databrowser/view/ISOC_E_DIIN2, (accessed on March 2024).
In terms of buying cloud computing services, Italy performs well (see Figure 2.7, Panel A). Its share of enterprises buying at least one type of cloud computing service (61%) is higher than that of the European Union (45%) and Germany (46%). France underperformed in 2023 with 27% of enterprises. For the ICT sector, Italian enterprises demonstrate a strong propensity for embracing cloud technologies services, with the highest share at 84%; Germany closely follows, with 82% of enterprises buying at least one cloud service. France (68%) and Spain (74%) underperform the EU27 average, which stands at 79% (Panel C).
Percentage of enterprises with ten or more employees and self-employed persons
Source: Eurostat, Cloud computing services by NACE Rev.2 activity, https://ec.europa.eu/eurostat/databrowser/view/isoc_cicce_usen2, (accessed on March 2024).
In 2023, Denmark and Finland led the European Union with the highest share of enterprises utilising at least one AI technology, both standing at about 15% (see Figure 2.8).
Percentage of enterprises with ten or more employees
Source: Eurostat (2025[12]), Artificial Intelligence by size class of enterprise, https://doi.org/10.2908/ISOC_EB_AI.
On the other hand, Italy and France have relatively lower shares, with only 5% and 6% respectively. German enterprises showcase a higher usage of AI technologies at 12%, positioning it above the EU average of 8%. A separate study covering all firms in Germany found that only 5.8% used AI in 2019 (Rammer, Czarnitzki and Fernández, 2021[13]).
A survey commissioned by the European Commission (Gossé, Hoffreumon and van Zeebroeck, 2020[14]). found, unsurprisingly, that IT is the sector with the highest rate of AI adoption, with 63% of enterprises in this sector adopting at least one AI technology. IT is followed by education (49%) and manufacturing and human health (both with 47%). Transport is the sector with the lowest AI adoption rate (36% of enterprises). Though only 40% of enterprises in Europe’s finance and insurance sector have adopted AI technologies, 27% plan to do so in the future, which is a higher share than for any of the other sectors considered.
The same survey suggests that among European enterprises that have adopted AI technologies, purchasing ready-made systems is the most common procurement method; nearly three in five adopting enterprises have opted for this method. This is followed by hiring external providers to develop the technology, though only 38% of enterprises have used this method. In-house development or modification of AI systems is much less common (about one in five enterprises).
Other survey evidence from national official sources
Other survey evidence from national official sources also covers Denmark, France and Korea (Montagnier P. and Ek I., 2021[15]). The findings from these national surveys are shown in Table 2.1, broken down by size category of firm. The table shows large differences across countries, ranging from a low of just 1.5 % of all firms in Korea in 2017 to 11.4 % in France in 2018. The possible reasons for such large differences are commented on further below. The data in Table 2.1 also coincide with findings from recent OECD work that used a novel statistical approach to analyse the diffusion of AI in firms across ten countries (Calvino and Fontanelli, 2023[16]). This OECD study used an assemblage of micro-data from firm-level surveys across 11 countries to examine the prevalence and impact of AI in different sectors. It found that AI is mostly used in the ICT and Professional Service sectors and is more widespread across large and somewhat younger firms. The study also showed that companies using AI tend to be more successful, especially the larger ones. It pointed out that complementary assets, including ICT skills, high-speed digital infrastructure, and using other digital technologies, are crucial for companies to get the most out of AI.
Survey findings from recent years
Notes: 1. Statistics Denmark, www.dst.dk/en. 2. INSEE, www.insee.fr. 3. Ministry of Science and ICT, www.msit.go.kr/eng/index.do. 4. Ministry of Science and ICT www.msit.go.kr/eng/index.do. 5. Based on the establishment level, not on the firm level.
Source: Montagnier and Ek, (2021[17]), "AI measurement in ICT usage surveys: A review," OECD Digital Economy Papers 308, OECD Publishing, https://doi.org/10.1787/72cce754-en.
Possible reasons why survey results differ across countries
A need exists to better understand survey comparability across countries. Some surveys have yielded results that appear counter-intuitive and hard to explain in terms other than methodological. Montagnier and Ek (2021[15]) describe possible sources of difference in survey results. These could include, first, the coverage of the surveys, both in terms of the target population (e.g. business units surveyed can be enterprises or establishments) and in terms of industries and the size of firms included in different sample strata. For example, in Korea (see Table 2.1), the unit surveyed was “establishment”, not “enterprise”. Second, it may be due to differences in the questionnaires, the heterogeneity of AI definitions and the different nature, wordings and scope of the questions.
Some of the factors behind weak adoption of AI in enterprises include, broadly speaking, a lack of digital readiness, uncertainty about use cases and return on investment (ROI), concerns about access to skills and talent, and concerns about the cost of AI technology.
Kazakova et al. (Kazakova et al., 2020[18]) reported that the most common internal barriers to AI adoption for European companies were: 1) difficulties hiring staff with the right skills (57% of surveyed establishments reporting this as a barrier); 2) cost of adoption (52%); and 3) cost of adapting operational processes (49%). External barriers are less common, but the most reported were: 1) lack of public/external financing (36%); 2) liability for potential damages (33%); and 3) data standardisation (33%). In addition to being the most frequently reported, internal obstacles were the most significant deterrents to adoption. Of the internal obstacles, lack of internal skills, cost of adoption, lack of internal data, and IT infrastructure were considered the most important.
Establishments in the United Kingdom reported fewer barriers to adoption than the European average, and these barriers were more likely to be external (than for the average country in the analysis). Meanwhile, France and Germany reported more barriers than the European average, and for both countries, the share of internal barriers was roughly in line with the European average.
Table 2.2 shows the most frequently reported internal and external barriers for companies in the European Union, as well as for France, Germany, Italy, and the United Kingdom individually (the survey period preceded the departure of the United Kingdom from the European Union). The most common barriers are internal to firms. The most cited barrier in France was the cost of adapting existing processes; the most commonly cited barrier in Germany was the difficulty of hiring staff with the necessary skills (with more than three in four establishments reporting this as a barrier). The most cited barrier in both Italy and the United Kingdom was the cost of adoption (though less than half of establishments in the United Kingdom report this as a barrier).
Source: Kazakova et al., (2020[18]), European Enterprise Survey on the Use of Technologies Based on Artificial Intelligence, Publications Office of the European Union, Luxembourg, https://data.europa.eu/doi/10.2759/759368.
Lack of digital readiness
AI solutions require access to data. However, many firms have not fully digitised key business functions. For example, in the United States, Zolas et al. (2020[6]) find that, overall, “the lowest rates of [digital information] adoption are observed in production and supply chain activities” (although this is impacted by the lower count of manufacturers in their survey). As much as 36% of customer feedback, 38% of information on production activities, and 42% of supply chain-related information is not digitised. These shortfalls restrict the ability to generate value using AI.
Zolas et al. (2020[6]) find that “technology adoption exhibits a hierarchical pattern, with the most sophisticated technologies adopted most often only when more basic applications were as well.” One of the most important technologies many companies will adopt before AI is cloud computing, a platform that makes it easier and cheaper for businesses to innovate with AI by helping them to operate and maintain the IT infrastructure and services they need (Bill Whyman, 2021[19]). However, the ABS found that 60.7% of US companies (including 64.6% of companies in manufacturing industries and 60.6% in non-manufacturing industries) had not yet adopted cloud computing (NSF, 2019[7]).
As shown in Figure 2.9, manufacturers generally progress along six stages of digitalisation (Schuh et al., 2017[20]). Stages 1 and 2 refer to basic digitalisation – “computerisation” and “connectivity” –getting data into computers, integrating companies’ various technology systems, and (for manufacturers) connecting key production equipment into an integrated, enterprise-wide IT system. Companies can then progress to advanced monitoring, being able to see what is happening in real time across the business, from production equipment on the factory floor to parts as they move through the supply chain, customers’ use of a firm’s products and digital services. The firm can possess an always-up-to-date digital model of its factories.
Source: Suchuh et al., (2017[20]), Acatech – National Academy of Science and Engineering, original graphic courtesy FIR e. V. at RWTH Aachen University.
Next, the most sophisticated companies progress to using technologies that permit an understanding of why what is happening is happening. For example, for a manufacturer, this might entail using AI to facilitate root-cause analysis for the failure of a part or a piece of production equipment. Lastly, companies can move to a phase of “Predictive Capacity,” or of being prepared for what may happen (e.g. predicting machine failure in advance or proactively readjusting the flow of inventory to stores) and then, ideally, to “Adaptability,” which refers to self-optimising organisations or factories in which autonomous responses can be achieved, all the way to machines capable of detecting and even fixing their own error modes.
From March to June 2019, the Manufacturers Alliance for Productivity and Innovation (MAPI, now renamed as the Manufacturers Alliance) and the Information Technology and Innovation Foundation (ITIF) surveyed 200 US manufacturers (generally with sales of between USD 500 million and USD 10 billion), receiving 60 usable results (Atkinson and Ezell, 2019[21]). This six-stage model described above was used to assess companies’ progress in digitalising manufacturing. Over half of the respondents indicated their companies were only at the initial stages of digitalisation of manufacturing (see Figure 2.10).
Source: Atkinson R. and Ezell S, (2019[21]), The Manufacturing Evolution: How AI Will Transform Manufacturing and the Workforce of the Future, Information Technology and Innovation Foundation, Washington, DC, https://itif.org/publications/2019/08/06/manufacturing-evolution-how-ai-will-transform-manufacturing-and-workforce.
Around two-fifths of the surveyed firms had progressed to the stage of having strong visibility into their manufacturing operations, while less than 10% had achieved either transparency or predictability in manufacturing operations. No respondents reported being at a stage of self-optimisation. Firms at the initial stage of digitalisation – just integrating their IT systems and data sets – which account for half of medium-sized manufacturers in the United States, will find it hard to run AI tools like expert or predictive systems.
The National Science Foundation (2019[7]) reported that the most significant factor adversely affecting AI adoption and utilisation was the cost of the technology, with 12.5% of US manufacturers and 7.3% of non-manufacturers reporting that cost was prohibitive. The next most significant factors adversely affecting AI adoption among manufacturers were lack of capital (2.8%), concerns regarding the technology’s maturity, and lack of access to talent (2.4% of respondents each). For non-manufacturers, apart from cost, concerns about the technology’s maturity (2.0%) and lack of access to capital (1.4%) were the most important. Overall, 45.9% of respondents reported that AI was not applicable to their business, while 42.4% reported that no factors adversely impacted their adoption of AI.
Enterprises responding to the MAPI/ITIF survey reported that the most significant barrier to deploying AI solutions was a lack of data resources (58% of respondents). The second-most significant barrier was uncertainty about how to use AI solutions to solve specific manufacturing challenges (52%). The third most important barrier was a lack of interoperability between equipment, which precluded the data integration necessary to support AI applications (47%). Other significant concerns included a lack of workplace digital skills, uncertainty about the ROI, and a lack of buy-in from senior executives (see Figure 2.11).
Percentage of respondents
Source: Atkinson R. and Ezell S, (2019[21]), The Manufacturing Evolution: How AI Will Transform Manufacturing and the Workforce of the Future, Information Technology and Innovation Foundation, Washington, DC, https://itif.org/publications/2019/08/06/manufacturing-evolution-how-ai-will-transform-manufacturing-and-workforce.
In December 2019, the tech research and outreach company O’Reilly Media surveyed nearly 1 400 business leaders worldwide, with most responses gathered from North America, followed by Western Europe and Asia. It found that the primary reasons businesses choose not to adopt AI were that: 1) “Company culture does not yet recognise needs for AI”; 2) “Difficulties in identifying appropriate business use cases”; 3) “Lack of skilled people/difficulty hiring the required roles”; and 4) “Lack of data or data quality issues” (Magoulas and Swoyer, 2020[22]).
Lack of organisational readiness
Beyond technological costs, what surfaces from the studies cited above is that many firms are uncertain about how to deploy AI tools and what the ROI of using AI might be. Lack of senior management buy-in in many companies also appears problematic. Successfully adopting AI will sometimes require enterprises to develop effective internal change management strategies. For those that do, the rewards could be significant. For instance, one study of companies with revenues over USD 3 billion found that organisations that take steps to embrace digital transformation generate an average of USD 100 million more in operating income each year than those that do not (Arkan, 2018[23]). Yet despite such potential gains, many executives fail to pursue digital transformation projects, partly due to challenges in changing corporate culture or adopting new ways of working. Overcoming such challenges will be vital to capturing the promise of AI.
This highlights a key point made by Daugherty and Wilson (2018[24]) that the enterprises that will do best in gaining from AI are not those that merely apply AI tools to existing processes but the ones that apply AI tools to fully reimagine and reinvent their processes, especially as concerns the creation of collaborative teams of humans working alongside machines. For instance, BMW (Bayerische Motoren Werke AG) determined that human-robot interactions in an automotive factory are about 85% more productive than either humans or robots working on their own (Knight, 2014[25]). As Markus Schaefer, head of production planning at Mercedes-Benz, observed, “When we have people and machines co-operating, such as a person guiding a part-automatic robot, we’re much more flexible and can produce many products on one production line. The variety is too much for the machines to take on.” (Knight, 2014[25]).
For companies, this leads to another essential point: implementing AI requires effective change management practices. However, a 2019 McKinsey study found that “only 8% of firms engage in core practices that support widespread AI adoption” (Fountaine, McCarthy and Saleh, 2019[26]). The MAPI/ITIF survey queried executives in medium-sized manufacturers on specific change management strategies to enable AI transformation. It found that, among other things, only 8% of respondents had “developed internal retraining programs to upskill existing workers with needed AI/other digital skills.” Furthermore, only 8% had “developed a communications process to explain the implications of AI applications and solutions to employees, customers, and partners.”
Lack of access to skills
Surveys assessing barriers to AI adoption often find concerns about the availability of human capital, especially the ability of existing workers to adapt to make effective use of AI tools and systems. For instance, a survey of 1 200 C-level executives found that only 25% considered their workforce ready for AI adoption (Shook and Knickrehm, 2019[27]). However, surprisingly, only 3% reported that their organisations had plans to significantly increase investment in training programmes over the next three years. One possible explanation may be that employees are more ready for the AI transformation than their employers think. In fact, 68% of highly skilled workers and nearly half of lower-skilled workers were enthusiastic about AI’s potential impact on their work, while 67% of workers considered it important to develop their own skills to work with intelligent machines. As AI becomes more prevalent, investing in people becomes more important (Atkinson and Ezell, 2019[21]).
The competition for top-level data or computer scientists (i.e. professionals who code algorithms and develop AI/ML systems, as opposed to workers who would use them) is acute. Dividing US companies into three groups – “seasoned,” “skilled,” or “starters”, based on their number of AI production deployments undertaken – Deloitte found, in a December 2020 survey, that 41% of “seasoned” companies, 47% of “skilled” companies, and 58% of “starter” companies were experiencing “moderate to major” skills gaps in AI (Jarvis, 2020[28]). Similarly, in a 2020 survey of about 1 000 executives, 39% said they were not using AI because of a lack of technical expertise (McKendrick, 2020[29]). Other research focusing on the US AI workforce found that some skill types are scarcer than others, particularly scientists in computer and information research (Gehlhaus and Rahkovsky, 2021[30]).
Lack of vendors of AI solutions
Another factor that could significantly affect AI adoption is the availability of AI provider enterprises. Businesses in different industries may demand unique AI solutions that require specialised expertise from AI vendors. Consequently, even though specific innovation ecosystems may have many high-quality AI service companies, oligopolies or even monopolies can emerge in some industries and particular areas, leading to higher prices and fewer opportunities for businesses to increase productivity and competitiveness. Competition among AI vendors, however, could induce technology providers to maintain high operational quality, lower prices and retain customers, change specialisation and serve previously unaffected industries, and make technology more available to a wide range of markets.
Canada, the United Kingdom, and the United States all have high levels of local competition. According to Statista (2024[31]) (a German company specialising in market and consumer data), there are more than five AI enterprises for every million people in these three countries. Canada has more than seven AI firms per million people. The Canadian performance is perhaps unsurprising as it ranks fourth in the Global AI Index and produced the highest number of AI patents per capita among G7 countries between 2015 and 2018. The European Union is lagging, as dominant countries in the European Union – like France, Germany, and Spain – have fewer than two AI firms per million people. Asian countries face the greatest challenges in terms of the level of AI competition (with India securing only one AI vendor for every ten million people).
VC investment data from CB Insights show that AI vendor firms’ participation in VC deals, measured per million of the population, is more than 50% higher in the United States than in Canada (CB Insights, 2021[32]). Europe underperforms the United States and Canada in terms of competition, which hinders the potential to converge on the North American performance. Asia and Latin America considerably lag North America and Europe.
These findings also align with the OECD analysis of global investments by venture capitalists in private companies focused on AI (Tricot, 2021[33]). This work found VC investments in AI to be growing at a dramatic pace. The United States and the People’s Republic of China (hereafter, “China”) were seen to be leading this wave of investments that tend to concentrate on a few key industries. The data showed that the European Union, the United Kingdom, and Japan increased investments but lagged the two dominant players. The study analysed VC investments in 8 300 AI firms worldwide, covering 20 549 transactions between 2012 and 2020.
Figure 2.12 compares VC investments in AI across several countries from 2012 to an estimated value for 2023. The United States demonstrates a leading position in AI investments, with a noticeable peak in 2018 before continuing the same trendline experienced in prior years. China exhibited a surge in investment in 2017, temporarily surpassing the United States. The EU27 aggregate shows a gradual increase in investment since 2021, reaching levels slightly below those of China in 2022‑23. Other countries maintain comparatively low and stable levels of investment throughout the period.
Note: 2023 value is an estimate.
Source: OECD AI Policy Observatory, https://oecd.ai (accessed on 8 February 2024). Visualisations powered by JSI using data from Preqin.
AI represents a transformative technology for the 21st century. Countries, industries and enterprises that develop strong competencies in this general-purpose technology will enjoy productivity and competitive advantages. Yet the available survey evidence shows that most enterprises, especially smaller ones, are only at the earliest stages of AI adoption. The literature prior to the OECD (2022-23[34]) indicates that challenges exist in understanding business models and use cases, affording the technology, adopting effective change management practices, and acquiring or retraining skilled workers capable of fully taking advantage of AI technologies.
References
[23] Arkan, C. (2018), The Workforce of the Future: Insights around disruption, transformation and the role of AI // 13, Microsoft, http://download.microsoft.com/download/0/8/5/0856D32D-EA50-4226-8173-8D03BA2CEFCF/Microsoft_Insights_on_Workforce_Transformation_EN_US.pdf.
[21] Atkinson, R. and S. Ezell (2019), The Manufacturing Evolution: How AI Will Transform Manufacturing and the Workforce of the Future, Information Technology and Innovation Foundation, Washington, DC, https://itif.org/publications/2019/08/06/manufacturing-evolution-how-ai-will-transform-manufacturing-and-workforce.
[19] Bill Whyman (2021), Secrets From Cloud Computing’s First Stage: An Action Agenda for Government and Industry, https://itif.org/publications/2021/06/01/secrets-cloud-computings-first-stage-action-agenda-government-and-industry/.
[16] Calvino, F. and L. Fontanelli (2023), “A portrait of AI adopters across countries: Firm characteristics, assets’ complementarities and productivity”, OECD Science, Technology and Industry Working Papers, No. 2023/02, OECD Publishing, Paris, https://doi.org/10.1787/0fb79bb9-en.
[1] Canada, Statistics (2019), Survey of Innovation and Business Strategy, http://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&SDDS=5171&lang=en&db=imdb&adm=8&dis=2.
[32] CB Insights (2021), State of AI Global 2021, Research Artificial Intelligence, http://www.cbinsights.com/research/artificial-intelligence.
[24] Daugherty, P. and H. Wilson (2018), Human + Machine: Reimaging Work in the Age of AI, Harvard University Press, Boston, https://store.hbr.org/product/human-machine-updated-and-expanded-reimagining-work-in-the-age-of-ai/10724?srsltid=AfmBOoo3cipasn7DMFTq4cXJT4oUqi0tSRfwKd52bgecd4Xrlex2eMcm.
[12] Eurostat (2025), “Artificial Intelligence by size class of enterprise”, Artificial intelligence by size class of enterprise, https://doi.org/10.2908/ISOC_EB_AI (accessed on 6 March 2023).
[11] Eurostat (2022), “ISOC_EB_AI, Percentage of enterprises”, Artificial intelligence, https://ec.europa.eu/eurostat/databrowser/view/isoc_eb_ai/default/table?lang=en.
[3] Evans, A. and A. Heimann (2022), AI Activity in UK Businesses Report, Capital Economics and DCMS, January 2022, http://www.gov.uk/government/publications/ai-activity-in-uk-businesses.
[9] Fleming, M. (2023), Business Experts Press, New York Breakthrough: A Growth Revolution, Business Experts Press, New York.
[26] Fountaine, T., B. McCarthy and T. Saleh (2019), “Building the AI-powered organization”, https://hbr.org/2019/07/building-the-ai-powered-organization.
[30] Gehlhaus, D. and I. Rahkovsky (2021), AI Workforce: Labor Market Dynamics, Center for Security and Emerging Technology, https://cset.georgetown.edu/wp-content/uploads/CSET-US-AI-Workforce-Labor-Market-Dynamics.pdf.
[14] Gossé, J., C. Hoffreumon and N. van Zeebroeck (2020), European Enterprise Survey on the Use of Technologies Based on Artificial Intelligence, European Commission, https://op.europa.eu/en/publication-detail/-/publication/f089bbae-f0b0-11ea-991b-01aa75ed71a1.
[20] Herbert Utz Verlag, M. (ed.) (2017), Industrie 4.0 Maturity Index, Managing the Digital Transformation of Companies, acatech STUDY, https://www.acatech.de/wp-content/uploads/2018/03/acatech_STUDIE_Maturity_Index_eng_WEB-1.pdf.
[28] Jarvis, D. (2020), “The AI talent shortage isn’t over yet”, Deloitte Insights, http://www2.deloitte.com/us/en/insights/industry/technology/ai-talent-challenges-shortage.html.
[18] Kazakova, S. et al. (2020), European Enterprise Survey on the Use of Technologies Based on Artificial Intelligence, Publications Office of the European Union, Luxembourg European Enterprise, https://data.europa.eu/doi/10.2759/759368.
[25] Knight, W. (2014), “How human-robot teamwork will upend manufacturing”, MIT Technology Review, http://www.technologyreview.com/s/530696/how-human-robot-teamwork-will-upend-manufacturing/.
[8] Kobe, K. and S. Richard (2018), Small Business GDP: 1998-2014, US Small Business Administration, Office of Advocacy, https://advocacy.sba.gov/wp-content/uploads/2018/12/Small-Business-GDP-1998-2014.pdf.
[22] Magoulas, R. and S. Swoyer (2020), AI Adoption in the Enterprise 2020, http://www.oreilly.com/radar/ai-adoption-in-the-enterprise-2020/.
[5] McElheran, K. et al. (2021), “AI Adoption in America: Who, What, and Where”, https://mackinstitute.wharton.upenn.edu/wp-content/uploads/2022/03/McElheran-Kristina-et-al._AI-Adoption-in-America.pdf.
[35] McKendrick, J. (2021), “AI adoption skyrocketed over the last 18 months”, https://hbr.org/2021/09/ai-adoption-skyrocketed-over-the-last-18-months.
[29] McKendrick, J. (2020), Artificial intelligence skills shortages re-emerge from hiatus, https://www.zdnet.com/article/artificial-intelligence-skills-shortages-re-emerge/.
[2] MIC (2021), 2020 Communication Usage Trend Survey, http://www.soumu.go.jp/johotsusintokei/statistics/pdf/HR202000_002.pdf.
[15] Montagnier P. and Ek I. (2021), “AI measurement in ICT usage surveys: A review”, OECD Digital Economy Papers, No. 308, OECD Publishing, Paris, https://doi.org/10.1787/72cce754-en.
[17] Montagnier, P. and I. Ek (2021), “AI measurement in ICT usage surveys: A review”, OECD Digital Economy Papers, No. 308, OECD Publishing, Paris, https://doi.org/10.1787/72cce754-en.
[4] National Science Foundation (2021), Annual Business Survey: 2021, https://ncses.nsf.gov/surveys/annual-business-survey/2021.
[7] NSF (2019), “Annual Business Survey: 2019 (Data Year 2018)”, National Science Foundation, https://ncses.nsf.gov/pubs/nsf22315.
[34] OECD (2022-23), OECD/BCG/INSEAD Survey of AI-Adopting Enterprises.
[13] Rammer, C., D. Czarnitzki and G. Fernández (2021), Artificial Intelligence and Industrial Innovation: Evidence from Firm-Level Data.
[27] Shook, E. and M. Knickrehm (2019), Reworking the Revolution, Accenture, http://www.academia.edu/38048342/Reworking_the_Revolution_2019.
[31] Statista (2024), Number of artificial intelligence (AI) companies worldwide as of June 2018, by country, https://www.statista.com/statistics/941054/number-of-ai-companies-worldwide-by-country/.
[33] Tricot, R. (2021), “Venture capital investments in artificial intelligence: Analysing trends in VC in AI companies from 2012 through 2020”, OECD Digital Economy Papers, No. 319, OECD Publishing, Paris, https://doi.org/10.1787/f97beae7-en.
[10] USPTO (2020), Inventing AI: Tracking the Diffusion of Artificial Intelligence with Patents 2020, United States Patent and Trademark Office, http://www.uspto.gov/sites/default/files/documents/OCE-DH-AI.pdf.
[6] Zolas, N. et al. (2020), Advanced Technologies Adoption and Use by U.S. Firms: Evidence from the Annual Business Survey, National Bureau of Economic Research, Cambridge, MA, https://doi.org/10.3386/w28290.
← 1. A PricewaterhouseCoopers (PwC) study found that 52% of surveyed companies accelerated their plans to adopt AI because of the COVID‑19 crisis (McKendrick, 2021[35]).