This chapter reports the results of a survey of artificial intelligence (AI) in enterprises in the state of São Paulo – the most economically important state of Brazil – using the same OECD/BCG/INSEAD survey questionnaire administered to 840 enterprises in the Group of Seven (G7) countries in 2022‑23. This chapter aims to present new data on AI use in firms in Brazil and compare these results to the findings from the OECD/BCG/INSEAD survey presented in Chapter 3. Overall, the use of AI is limited in the state of São Paulo, and there is a low incidence of enterprises developing AI systems internally. Benefits could come from examining the suitability of current funding and other support mechanisms.
The Adoption of Artificial Intelligence in Firms
New Evidence for Policymaking
Abstract
São Paulo is the most populous state in Brazil. In the most recent population census (2022), the state had 44.4 million residents, almost 22% of the population of Brazil.1 Economically, the state of São Paulo is also the largest, accounting for around 31% of Brazil’s gross domestic product (GDP).2 It also contributes significant shares of national production in high-tech sectors and hosts an innovation ecosystem that includes Brazil’s main universities and research centres. An important objective of the survey was to evaluate the technological maturity of enterprises in São Paulo, focusing on AI adoption, given that São Paulo is the most economically developed state in Brazil.
The survey was conducted by the Fundação Sistema Estadual de Análise de Dados (SEADE), the official statistics and data production organisation of the state of São Paulo, in partnership with the Regional Center for Studies on the Development of the Information Society (Cetic.br), from the Brazilian Network Information Center (NIC.br). With minor adjustments, the same survey instrument was used as that administered by the OECD/BCG/INSEAD to 840 enterprises adopting artificial intelligence (AI) across the Group of Seven (G7) countries, with the same target populations of medium and large-sized enterprises, in the same sectors of manufacturing and information and communication technology (ICT) services.
The survey results highlight that AI adoption in enterprises in São Paulo is dominated by companies procuring solutions externally. Around 70% of the data used to fuel AI applications is sourced internally. Just over half of the enterprises using AI (52%) employ some form of customised system from third parties or purchase off-the-shelf AI solutions (51%), indicating a limited level of internal development of AI applications. The most frequent AI application is in customer-facing services, reported by 49% of enterprises using AI, closely followed by process and control optimisation, cited by 44% of respondents. When considering obstacles to AI implementation, 44% of enterprises that use AI identify concerns about privacy and security as significant issues, followed closely (38%) by uncertainty about the return on investment (ROI).
The findings suggest that economic benefit could come from creating support instruments that encourage partnerships around projects for innovation in products and services using AI. Support instruments in Brazil specifically for AI are barely developed. With AI adoption in mind, benefits could be had from examining the suitability of current funding mechanisms and public programmes to support skills development and the provision of information services examined in the survey.
Following a brief section on the survey’s methodology, this chapter presents the survey’s main findings and a discussion on expanding the uptake of AI in Brazil, followed by a conclusion.
The data collection instrument was slightly adapted from the OECD/BCG/INSEAD survey questionnaire. After translating the questionnaire into Portuguese, specific wording adjustments were made to better suit the Brazilian context, incorporating insights from cognitive interviews. Overall, the adapted questionnaire is directly comparable with the survey implemented in G7 countries.
The target population comprised enterprises in the manufacturing and ICT sectors (only two ICT subsectors were considered, namely: ISIC 62: Computer programming, consultancy, and related activities; and ISIC 63: Information service activities. The manufacture of devices and components, such as semi-conductors, used in data and information processing and communication were not included under “ICT”). The survey also targeted two enterprise size classes: medium-sized (50 to 249 employees) and large-sized (more than 250 employees). Data were collected using computer-assisted telephone interviewing (CATI) and computer-assisted web interviewing (CAWI) techniques between February and July 2023.
The survey adopted a probabilistic approach, meaning that it aimed to obtain results that were statistically representative of the entire population of enterprises in the state of São Paulo. However, the sample design was not directly comparable to that adopted for the OECD/BCG/INSEAD survey in G7 countries, largely because the G7 survey is not directly generalisable to the respective population of enterprises in each country.
Sample design
The registry of enterprises for the state of São Paulo was obtained from the Brazilian Secretariat of the Federal Revenue website.3 Enterprises’ main activities were identified using the International Standard Industrial Classification of All Economic Activities (ISIC) classification. After this initial selection, the registry was forwarded to the Brazilian Institute of Geography and Statistics to distinguish between medium-sized enterprises (50 to 249 employees) and large enterprises (250 or more employees). This record was divided into four cohorts, as indicated in Table 6.1. Given the need to survey only those that use AI and knowing from prior research that rates of AI use in Brazil are low, all enterprises in the four strata were approached via telephone or email (without randomised selection) to increase the number of enterprises in the sample.
Source: SEADE survey data.
Initially, 4 557 enterprises were approached via telephone or email. Of those, it was determined that 2 561 of these enterprises were active in the state of São Paulo. From that group, 1 542 responded to the survey questionnaire (60% of the target population). A standard correction was made for the non-responses, such that the results presented below statistically represent the 2 561 enterprises.
From the total number of enterprises considered in the survey (i.e. 2 561), only 7% were users of some type of AI application. Accordingly, despite the large number of enterprises approached, the results obtained for the full OECD/BCG/INSEAD questionnaire are limited to a final set of 167 companies. Due to the small number of completed questionnaires obtained, the margins of error (both overall and by item in the questionnaire) are high and do not allow statistical analyses that are representative of all active users by sector (ICT services and manufacturing) and by enterprise size. This implies that the results can only be provided for the total number of enterprises, and it is not possible to obtain disaggregated results by, for example, medium-sized manufacturing or large-sized manufacturing enterprises.
This probabilistic approach, applied in a context of low AI adoption, presents notable operational challenges for survey implementation. Issues include respondents’ limited understanding of the concepts employed and, as the procedure described above illustrates, the need for extensive screening to find suitable respondents. As the number of enterprises utilising AI is expected to increase, survey implementation using a probabilistic approach may become more straightforward.
Uses of AI, types of application, and importance of AI to enterprises
Overall, 7% of the enterprises in the state of São Paulo use AI, a figure roughly aligned with various national-level surveys described in Chapter 2. Most of these are active users (6% of enterprises in the state). In contrast to the data obtained from G7 countries – where the sample consists of relatively advanced AI users – a vast majority of the enterprises surveyed in the state of São Paulo use only a few AI applications (58% with just one or two uses of AI compared to 4 in G7 countries) (Figure 6.1). Enterprises in the ICT services industry exhibit a higher average number of AI uses (27%) compared to manufacturers (5%). These results show that there is considerable room for expanding the use of AI in enterprises in São Paulo, transitioning from point solutions to more integrated adoption, such as incorporating customer relationship management systems.
Share of surveyed enterprises (%) using each number of AI applications
Source: SEADE survey data.
Among enterprises using AI, 49% use it in customer-facing services (Figure 6.2). The second most prevalent application of AI involves process control, automation, and optimisation of production (44%), including such uses as predictive maintenance and automated support for programmers. These results broadly align with the findings from G7 countries.
Only 23% of the surveyed enterprises use AI for research and development (R&D), considerably lower than in most G7 countries (Figure 6.2). Another 28% use AI for detecting defects and anomalies. The enterprises surveyed in G7 countries have embraced more advanced features of AI that demand greater capabilities and continuous learning. By comparison, enterprises in São Paulo are just beginning to unlock the benefits of AI, making more use of ready-made solutions, and with lower levels of internal development.
Share of enterprises using each application of AI (%)
Note: Percentages sum to more than 100 because enterprises may use more than one application of AI.
Source: SEADE survey data.
Regarding the importance assigned to AI applications, 47% of enterprises consider AI “very important”, and 32% consider AI to be “one among a number of important considerations” (Figure 6.3). A higher share of enterprises in São Paulo considers AI of minor importance to main business processes (20%) than in G7 countries (8%). This finding reflects the presence of less advanced AI users in the Brazil sample.
AI and data infrastructure
Regarding the type of databases used, enterprises in São Paulo have prioritised their own data resources to feed AI applications. As shown in Figure 6.4, 70% of enterprises obtain necessary data from internal sources, such as data from sensors for predictive maintenance of machines. Additionally, 53% of enterprises cite customers or product/service users as the primary sources of data or data acquisition. In G7 countries, 78% of enterprises, a similar share, states that they collect data internally, but a higher share, 75%, reports using data from customers and users.
Partnerships with external organisations that function as a data source, among other things, are significant but not as common as in the enterprises surveyed in G7 countries. Such partnerships include those with private enterprises (25%), research institutes (24%), government organisations (21%) and private data providers (20%). Overall, the results reflect that the use of internal and other proprietary data are more widespread in Brazil. In general, the use of external data (in addition to the use of internal data), which is more frequent in the G7 countries, is associated with greater data maturity.
Share of enterprises per attributed level of importance of AI (%)
Source: SEADE survey data.
Share of enterprises using selected sources of data for AI (%)
Source: SEADE survey data.
From the overall sample of enterprises that use some type of AI, 78% adopt data management solutions like remote servers, data lakes or data warehouses (Figure 6.5). This result also aligns with findings obtained among G7 countries, indicating that most enterprises are cognisant of the prerequisites for AI deployment.
Share of enterprises in each category of use or familiarity (%)
Source: SEADE survey data.
Practices and partnerships to adopt and develop AI
Approximately 52% of AI-using enterprises in the state of São Paulo turn to third-party customised systems, while 51% adopt AI by acquiring new software or hardware or hiring consultancy services. Additionally, 43% invest in their own R&D to develop AI (Figure 6.6). In G7 countries, a higher proportion of enterprises (70%) were found to engage in R&D in AI for their own use, followed by the development of customised systems and procurement of off-the-shelf software or hardware.
About 37% of enterprises that use AI cite collaboration with other enterprises that have capabilities in the field as a means of adopting or developing AI. This is low compared to G7 countries, where this figure is around or above 40% in all but one country, and, in some cases, above 50%. Employee training to help develop or apply AI was cited by 38% of enterprises, while in the G7 countries, this was mentioned by 75% of enterprises.
Only 17% of AI-using enterprises have established a senior management position or formed a dedicated team for AI. The limited existence of leadership positions in AI focused on team direction and decision making is discussed below in the section on human resources. The figures for São Paulo indicate a relatively low degree of emphasis on establishing leadership positions to develop AI. In contrast, the G7 survey revealed that around half of enterprises had created senior management roles or dedicated teams with responsibilities for AI.
Share of enterprises using each method to adopt or develop AI (%)
Note: Percentages sum to over 100 because enterprises may use multiple methods.
Source: SEADE survey data.
Figure 6.7 shows the incidence of partnerships between enterprises and other institutions to develop AI applications. Evident is the limited extent of partnerships with researchers. Undergraduate students, faculty, doctoral students and postdoctoral researchers are mentioned as collaborators in only 6% of cases, while partnerships with researchers outside of universities occur in only 5% of AI-using enterprises. Despite the evident importance of AI as an emerging technology, there is a significant gap in building collaborative partnerships. For instance, among G7 countries, over half of the surveyed enterprises have collaborated with university faculty, PhD or postdoctoral students. This indicates a more mature environment for fostering academic and business relationships, potentially catalysing innovation and company creation. Figure 6.7 also shows that the most frequent form of partnership occurs with partners not linked to academic or research organisations.
Share of enterprises engaged in each partnership type (%)
Source: SEADE survey data.
Human resources for AI
The level of awareness of the disruptive consequences of adopting AI can also be assessed using data on the organisational structure of enterprises. Brazil’s 2021 ICT Enterprises Survey showed that 13% of Brazilian companies use some type of AI. However, 39% of large companies do so. This significantly higher rate of adoption stems from large companies’ substantial investments in software and hardware, coupled with the greater availability of financial and human resources dedicated to experimentation with disruptive technologies (Brazilian Internet Steering Committee, 2022[1]).
Figure 6.8 presents data on the existence of positions related to AI in enterprises in the state of São Paulo. The most prevalent position relating to data management, processing and AI is that of data protection officer (DPO), which is present in 69% of enterprises. This role typically entails responsibility for defining policies, standards and practices to ensure data quality, security and compliance. The widespread presence of DPOs indicates a concern among enterprises in the state of São Paulo with data security.4 This situation may be associated with the Brazilian General Data Protection Law, enforced since 2020. In response to that legislation, enterprises have begun to emphasise internal data governance, instituting more robust processes for handling personal data.5
Share of enterprises with the associated role (%)
Note: Percentages sum to over 100 because enterprises may have more than one of the cited roles.
Source: SEADE survey data.
Managerial positions related to AI are still rare among enterprises in the state of São Paulo, even those that use AI. Only 21% of enterprises that apply some form of AI indicate the existence of a role like an AI risk manager or a position responsible for AI ethics, for trust and digital security, or an equivalent function (Figure 6.8). Only 18% of enterprises have AI project manager roles.
Regarding C-suite positions, 35% of enterprises using some form of AI have chief information officer and/or chief digital officer positions. In other words, just over one-third of enterprises that use AI have established leadership positions formally responsible for an effective information technology (IT) infrastructure or digital initiatives oriented to innovation. This finding indicates a relatively restricted use of AI.
As for the presence of senior staff responsible for using data for new strategic initiatives and business objectives, only 15% of enterprises have a chief analytics officer, chief data officer, and/or head of data science positions. This underscores the overall finding of a limited presence of professionals dedicated exclusively to data governance, an essential activity for developing AI applications.
In terms of the workforce, and relative to G7 countries, a pattern exists of lower demand for AI-related talent. Notably, 57% of enterprises in São Paulo report not opening specific positions for AI (Figure 6.9), while this proportion was only 20% among G7 countries. Furthermore, only 23% of enterprises in the state of São Paulo indicate hiring professionals for AI roles, compared to 67% among G7 countries. The somewhat incipient level of AI adoption among enterprises in the state of São Paulo suggests that there might not be significant problems with talent availability at present. However, this may not persist, particularly as more AI solutions enter the market.
Share of enterprises in each category (%)
Source: SEADE survey data.
Obstacles to adopting AI
Cloud computing provides critical infrastructure and cloud-based services to support AI applications and to make their development and deployment more accessible and scalable. The integration of these two technologies – infrastructure and services – is driving many advances in the field of AI and is present in a wide variety of industries. Among enterprises in the state of São Paulo, 70% indicate that they use cloud computing without difficulty. Only 7% of enterprises say they do not see any advantages in using cloud computing (Figure 6.10).
The cost of retooling systems was the most frequently indicated obstacle to cloud use among enterprises in G7 countries (cited by 60% of enterprises in manufacturing and 56% in ICT).
Share of enterprises experiencing each obstacle (%)
Note: Percentages sum to over 100 because enterprises may experience more than one obstacle.
Source: SEADE survey data.
This problem affects 36% of enterprises in the state of São Paulo. Regarding connectivity, 45% of enterprises in São Paulo cited concerns about network stability as a significant limiting factor in adopting cloud computing, a concern shared by a comparable fraction of enterprises in G7 countries.
For 44% of enterprises, the biggest obstacle to using AI relates to privacy, data protection or security (Figure 6.11). These results are connected to earlier observations that in enterprises based in São Paulo, there is a notable convergence between concerns about personal data protection and the utilisation of AI. Indeed, because much use of AI in enterprises in the state of São Paulo draws on internal data, it can be inferred that customer data are being utilised. This raises several questions regarding proper compliance with the law.
The difficulty in estimating the ROI in AI applications – a concern also described in Chapters 3 and 5 – was highlighted by 38% of respondents. In G7 countries, concerns regarding uncertain rates of ROI in AI were cited by 62% of enterprises in manufacturing and 56% in the ICT services industry.
Share of enterprises limited in using AI by each category (%)
Note: Percentages sum to over 100 because enterprises may experience more than one obstacle.
Source: SEADE survey data.
While Brazil’s public authorities have implemented many initiatives to support business innovation, none have been specifically tailored to AI to date. Enterprises were asked about the usefulness of three possible support mechanisms. Specifically, enterprises were queried on how helpful the following types of support could be to increase AI skills among staff:
partnerships with educational and vocational institutions
tax allowances or tax credits for training in AI
support to develop qualification frameworks for graduates in the field of AI.
As is the case in G7 countries, most enterprises indicate that one or another form of public support would help strengthen staff skills in AI. Figure 6.12 shows that 65% of enterprises that use some form of AI consider tax subsidies or tax credits for AI-related training as “very useful”. Some 64% also consider help to establish partnerships with educational and professional training institutions “very useful”. An only slightly lower incidence of support was shown for the development of qualification frameworks for graduates in the AI field, with 58% considering that this would be “very useful” and 34% deeming it “useful”.
Share of enterprises expressing agreement (%)
Source: SEADE survey data.
Enterprises were asked about how useful different types of mostly information services provided by the public sector could be to their use and development of AI:
information on and examples of business use cases in the firm’s industry
information on expected rates of ROI in AI
information on available and reliable technology vendors
information on available and reliable sources of private-sector advice and expertise
certification or accreditation schemes for AI solution providers
information on current or forthcoming regulations around data or AI.
As in G7 countries, a large majority stated that information services provided by the public sector would be “helpful” or even “very helpful” to their use of AI. For any of the services considered, no less than 78% of enterprises indicate that they would be at least “helpful”.
Figure 6.13 shows that 62% of the respondent enterprises consider that information on current or forthcoming regulations about data or AI would be “very useful”. Another 59% held that the dissemination of information about available and reliable sources of private-sector advice and expertise would be “very useful”. Some 54% also held that information about available and reliable technology providers would be “very useful”.
Share of enterprises expressing agreement (%)
Source: SEADE survey data.
With high levels of utility accorded to all of the selected public services, it is reasonable to argue that the role of the public sector in promoting new technologies in AI is important and should cover both the development of appropriate regulations and the provision of information to help equip managers to make better decisions in implementing AI.
Finally, regarding broader public sector initiatives to support the adoption of AI, investment in university education and professional training in AI is particularly important. Fully 75% of the enterprises that use some type of AI declare such initiatives “very useful” (Figure 6.14).
It is widely known that IT infrastructure and connectivity problems in some regions of Brazil require public sector incentives to be fully resolved.6 It is perhaps unsurprising then that 73% of enterprises cite upgrading IT infrastructure, such as high-speed broadband, as “very useful” for the adoption of AI.
Some 45% of enterprises maintain that collecting and publishing administrative databases would also be “very useful”. Despite the fundamental role data plays as an input for AI applications and the public sector’s efforts to make data available, this relatively low score indicates a pattern of intensive use of private data sources.
Share of enterprises expressing agreement (%)
Source: SEADE survey data.
The survey results indicate that the use of AI among large and medium-sized enterprises in the manufacturing and ICT services sectors in the state of São Paulo is still at an early stage of maturity. The data corroborates the findings of previous research in Brazil, such as the Brazilian Network Information Center, Regional Center for Studies on the Development of the Information Society and Brazilian Internet Steering Committee (2022[1]), which highlighted a low presence of AI across enterprises of all sizes and in all sectors of economic activity. The data also indicates that AI is mainly present in the processes most susceptible to automation. A particularly concerning point in comparison with G7 countries is the limited role of R&D in adopting AI.
Many opportunities exist for enterprises to promote the internal development of AI and to expand relationships with external partners. Beyond the enterprises themselves, the findings suggest that economic benefit could come from creating support instruments that encourage partnerships around projects for innovation in products and services using AI. Assessment of the suitability of current funding mechanisms and public programmes to support skills development and the provision of the information services examined in the survey would be worth pursuing.
There are some points to highlight that could contribute to improving the utility of the survey questionnaire. These concern adjustments to the questionnaire itself and changes to the data collection process. Regarding the questionnaire, one measure could be to expand the scope of research to enterprises that do not currently use AI but intend to do so or are in the initial steps of implementing AI for the first time. This would help to better understand the difficulties experienced in using AI and how these difficulties manifest in the different phases of implementation, such as in decision making around investments, the organisation and management of data, equipment acquisition and staff hiring. Such a shift to a broader set of themes on AI uptake would be particularly important in contexts where the overall use of AI in the corporate sector is low, which is the present reality in Brazil. In the current questionnaire, enterprises that do not use AI actively did not complete the survey.
Regarding data collection, in a future iteration of the survey, it could be helpful to identify specifically qualified persons in the responding enterprises to answer the questionnaire in advance. This is because the survey encompasses varied and specific topics (e.g. implementation obstacles, insights into the most helpful support services for the enterprise, and partnership arrangements). An alternative would be to consider having more than one respondent, as the topics addressed may be the responsibility of more than one team within the enterprise.
References
[1] Brazilian Internet Steering Committee (2022), Survey on the Use of Information and Communication Technologies in Brazilian Enterprises, https://cetic.br/media/docs/publicacoes/2/20221121122540/tic_empresas_2021_livro_eletronico.pdf.
← 1. For more information on the last census, see https://censo2022.seade.gov.br/.
← 2. Additional information on the GDP of São Paulo is available at https://pib.seade.gov.br/mensal/.
← 3. Every company in Brazil has a unique registration number. These are publicly available and provide basic information, such as the company’s address, whether it is active or not, and its size. This information can be accessed at https://solucoes.receita.fazenda.gov.br/servicos/cnpjreva/cnpjreva_solicitacao.asp.
← 4. According to the Brazilian Network Information Center, Regional Center for Studies on the Development of the Information Society and the Brazilian Internet Steering Committee (2022[1]), only 17% of Brazilian enterprises had appointed DPOs (41% of large enterprises, 29% of medium enterprises, and 15% of small enterprises). Although the Brazilian General Data Protection Law refers to the DPO as a person, there are no restrictions on the creation of interdepartmental data protection teams, or even hiring third-party agents.
← 5. During cognitive testing of the survey instrument, when asked about aspects of AI regulation, many respondents replied based on the procedures their enterprises were following to comply with the Brazilian General Data Protection Law. As of this writing, Brazil did not have specific regulations for AI.
← 6. Cetic.br conducted case studies on the deployment of the Industrial Internet of Things in manufacturing enterprises, revealing that challenges related to network stability and availability were identified as significant obstacles to increased sensor utilisation in companies' machines.