The Adoption of Artificial Intelligence in Firms

4. The goals and practices of institutions supporting the diffusion of artificial intelligence in firms

Abstract

This chapter identifies the main mechanisms used by 19 institutions in the Group of Seven (G7) countries, plus Singapore, to help firms adopt AI. It finds that technology extension services can help firms define the business problem to be solved and develop proofs-of-concept that demonstrate how AI can help. In addition, grants for business research and development and applied public research remove part of the risk associated with AI investments. Business advisory services also provide non-technical assistance that can raise managers’ understanding of their firms’ AI readiness and the specific opportunities and challenges that AI entails. Furthermore, networking, and collaborative platforms help build AI ecosystems of public and private actors. In addition, on-the-job training can help address bottlenecks around AI skills. Finally, information services and open-source code provide helpful resources for firms seeking to strengthen their AI capabilities. The chapter seeks to draw lessons for designing and implementing such services.

Introduction

Several technical features of artificial intelligence (AI) and broader characteristics of the markets faced by AI adopters have made its application more challenging than other digital technologies. Consequently, the uptake of AI in manufacturing firms, especially in small and medium-sized enterprises (SMEs), has been relatively low to date (see Chapter 2). Institutions supporting the diffusion and adoption of AI can help to address this issue. Many governments have ambitious national strategies that seek to achieve higher rates of AI uptake than would occur without active support for diffusion.

The OECD/Boston Consulting Group/INSEAD survey conducted among 840 enterprises in the Group of Seven (G7) countries in 2022‑23 examines how the policy environment supports (or can support) enterprises attempting to adopt and/or develop AI applications. This chapter affords complementary evidence on the types of public support being provided for AI uptake and the experiences of firms using such support. The literature on AI adoption has not yet explored the role of institutions in technology diffusion in depth.

Institutions for technology diffusion are public or quasi-public bodies that facilitate the spread and use of knowledge and methods that assist firms in adopting technologies (OECD, 2017[1]). Diffusion institutions have a unique perspective on the AI adoption challenge, as they interact daily with a wide range of firms and organisations seeking to develop and implement AI applications.

This chapter presents evidence gathered through desk research, structured interviews and written contributions from 19 institutions promoting AI adoption in firms in the G7 countries plus Singapore.

The interviewed institutions confirmed the role of the obstacles to AI adoption described in prior literature. However, they lay greater emphasis on some specific issues. They highlight uncertainties about the return on investment (ROI) as a significant barrier to firms adopting AI. Managers frequently struggle to grasp how AI can address real-world challenges in the workplace. They also tend to underestimate the implications of deploying AI solutions, which can require considerable changes in business culture and practices across many (if not all) segments of the firm. Many firms fail to apply AI due to limited access to AI skills and insufficient data maturity. Furthermore, regulatory uncertainties can deter enterprises from making efforts toward adoption.

This chapter identifies seven main mechanisms that diffusion institutions use to assist firms in overcoming the challenges of adopting AI:

1. Technology extension services can help firms narrow down and describe business problems to be solved and develop proofs-of-concept that demonstrate how AI can help.

2. Grants for business research and development (R&D) and applied public research mitigate some of the risks associated with AI expenditures.

3. Business advisory services provide non-technical support to managers, helping them improve their understanding of their firm’s AI readiness, opportunities, and challenges.

4. Grants for applied public research which can help promote high-risk research and/or the development and implementation of technologies close to commercialisation.

5. Networking and collaborative platforms aid in developing public and private AI ecosystems, increasing demonstration effects, and facilitating knowledge transfer.

6. On-the-job training can assist firms in solving constraints around AI skills.

7. Information services and open-source code provide helpful resources for firms to raise their AI capabilities.

To optimise their service offering, diffusion institutions frequently blend these mechanisms.

The role of diffusion institutions in promoting AI adoption

Rationales for the existence of diffusion institutions

AI’s economic and societal benefits will only materialise if the technology is responsibly designed, widely diffused, and adopted. Besides the firm-level barriers to the adoption of AI referred to in Chapter 2, certain market and systemic conditions may also lead to socially suboptimal adoption. Research and technology policies have generally emphasised support for basic science and R&D more than technology diffusion and adoption. There are several reasons why a greater focus on technology diffusion could be socially beneficial.

Governments may have strategic economic goals that require rates of AI uptake to be faster than would occur without active support for diffusion. For example, labour productivity has stagnated in OECD countries for decades. Technological upgrading in firms is essential to offset this stagnation. Increasing the average value of output per hour worked is more urgent still in the current context in which OECD populations are ageing rapidly and, as a consequence, the share of the population in work is falling while public spending on health and social care is rising rapidly. Even if there is a direct outlay of public resources to facilitate the uptake of AI and other productivity-enhancing technologies, the wider economic benefits of increased productivity may exceed the associated costs.1 Achieving a more balanced pattern of economic activity at the subnational level (e.g. regions) is another case where an overarching economic and social policy priority could justify spending on institutions to accelerate the uptake of AI and other productivity-enhancing technologies.

As described in Chapter 2, relatively few firms, particularly SMEs, have adopted AI. Consequently, the wider business community may see relatively few examples of successful use cases. In this context, diffusion institutions can help increase spillovers of useful information. They can provide information on examples of successful uses of AI while documenting implementation methods, business models, risks and other details that other companies might replicate. Useful information need not be limited to successful applications of AI: even if a firm’s attempt to adopt AI fails or an AI start-up collapses, valuable information is still created for others to learn from (e.g. pitfalls that should be avoided). However, entrepreneurs and businesses that generate this socially beneficial information receive no reward for doing so. Institutions that facilitate the diffusion of AI can aid the spread of economic and technological information about all aspects of AI in business.

Another problem that can arise in markets for relatively new technologies such as AI is the lack of specialised services supply. In the early stages of a technology’s market penetration, providers of complementary business and advisory services and specialised software often target large firms. This may happen due to the greater complexity and cost of working with large numbers of SMEs and because larger firms are better able than SMEs to bear the associated risks and costs. To address this, one step that diffusion institutions can take is to help SMEs understand the services and information provided by AI suppliers and help lower search costs for SMEs trying to identify services of reliable quality and relevance to their specific needs.

Finally, although less directly relevant to AI adoption in firms, diffusion institutions can also generate knowledge that informs policy making. Since these institutions work directly with firms, they have primary information about the needs, obstacles, and opportunities facing the private sector, as well as what policy settings best work for AI adoption. This information can be channelled into the policy-making process to improve the pertinence and quality of policies to support AI adoption.

Mechanisms used by diffusion institutions

Between April and August 2022, the OECD undertook 18 structured interviews with institutions in G7 countries that work to accelerate the diffusion of digital and other technologies in the business sector, including AI (for the sake of clarity and brevity, these institutions are hereafter referred to as “AI diffusion institutions”). The interviews explored how these institutions promote AI adoption. Public bodies and publicly funded non-profit organisations were identified through the OECD AI Observatory,2 the Science, Technology and Innovation Policy Compass3 database, and conversations with industry experts and delegates to the OECD’s Committee for Scientific and Technological Policy. AI Singapore was also invited to participate, as this institution has employed a number of novel approaches to diffusion, from which lessons might be drawn that are helpful to other countries.

The interviews first aimed to characterise each diffusion institution and how it supports AI adoption, complementing information already available online. They then explored each institution’s experiences and understanding of the main barriers to AI adoption in firms and other organisations, including awareness of what AI can do, staff skills and difficulties estimating the ROI. Finally, interviewees were asked to describe the institutions’ views on the most effective forms of support to overcome barriers to AI adoption. Annex I includes the core list of questions used to structure the interview. The questions were adjusted to the specificities of each institution using the information found online. Table 4.1 lists the participating institutions and sets out the mechanisms they use to promote AI adoption.

Table 4.1. Institutions interviewed for this study and the diffusion mechanisms they use

Country	Institution	Technology extension services	Grants for business R&D	Business advisory services	Grants for applied public research	Networking and collaborative platforms	On-the-job training	Information services and open-source code
Canada	Vector Institute	X				X
Canada	Scale AI	X	X			X
Canada	National Research Council-Waterloo Collaboration on AI, IoT and Cybersecurity	X	X		X	X	X
Canada	Forum IA Québec	X
France	Ministry of Ecology, “AI and Green Transition” programme						X
France	Cap Digital	X	X		X
Germany	Fraunhofer Institute for Industrial Engineering IAO	X		X				X
Germany	German Research Centre for Artificial Intelligence (DFKI)		X			X	X
Germany	Plattform Lernende Systeme				X	X
Germany	Mobility Data Space							X
Italy	Artificial Intelligence Research and Innovation Centre (AIR)		X
Italy	Siena Artificial Intelligence Hub (SAIHub)
Japan	New Energy and Industrial Technology Development Organization (NEDO)	X	X			X
United Kingdom	National Health Service (NHS) AI Lab			X
United Kingdom	Digital Catapult				X
United Kingdom	techUK			X
United States	Manufacturing Extension Partnership (MEP)	X
United States	Digital Manufacturing and Cybersecurity Institute (MxD)	X				X
Singapore	AI Singapore (AISG)		X	X	X	X	X	X

Each of the following sections focuses on a given mechanism used by diffusion institutions, i.e. technology extension services, grants for business R&D, business advisory services, funding for applied research, networking and collaborative platforms, on-the-job training, and information services and open-source mechanisms (Box 4.1). These sections also introduce the interviewed diffusion institutions, describing how they use the various mechanisms.

Box 4.1. Mechanisms used by technology diffusion institutions to promote AI adoption

• Technology extension services: Technology assistance provided by technology or research organisations with expertise in AI research and its applications. Rather than aiming for new technological breakthroughs, these services adapt established AI solutions for firms and other organisations. The services can be funded through public support, contractual services paid by firms, or a mix of public and private sources.

• Grants for business R&D: A direct allocation of funding for companies to invest in R&D to develop an AI application. Grants can be allocated to individual firms of different sizes (e.g. start-ups, SMEs, large firms) or a consortium involving research organisations. Beneficiaries often make a matching contribution and co-finance their R&D projects.

• Business advisory services: Assistance that promotes innovation and entrepreneurship by supporting business processes. These can include readiness assessment, market studies, fundraising and business design coaching, and support for business R&D grant applications, among other types of assistance. While these services do not foster technology transfer or the implementation of AI solutions per se, they can provide valuable guidance for businesses wishing to adopt AI.

• Grants for applied public research: A direct allocation to universities or other public research organisations seeking to finance AI research projects. Such projects conduct experimental work to enlarge the technological frontier. They involve industry actors that sometimes co-finance research expenditures and provide other in-kind resources (e.g. staff time, data used to train AI). Depending on the project, the pathway to research commercialisation can vary in duration, from months to more than ten years.

• Networking and collaborative platforms: Associations that gather a set of actors within the AI research and innovation system, often sharing common geographical locations. These platforms include industry players such as entrepreneurs, investors and companies, as well as public sector entities such as universities, research institutes and funding agencies. They can help a range of actors monitor the state-of-the-art in commercial AI technologies and their applications, match buyers and sellers of AI solutions, and support the fundraising efforts of AI start-ups, among other types of support.

• On-the-job training: Courses or instruction offered to employees and provided in parallel to their work, with the aim of deepening AI-related knowledge and skills. Different courses are offered to technical employees (e.g. data engineering, machine learning [ML]) and business executives (e.g. developing AI business plans and strategies). While not a substitute for tertiary education, on-the-job training helps employees contextualise coursework with the specific challenges and requirements of work.

• Information services and open-source code: Infrastructures and related resources that develop and maintain datasets or software that firms can use to develop AI applications. Firms, industry experts and other stakeholders participate in building the infrastructures and resources to ensure their relevance and applicability in business use cases.

Source: Adapted from EC/OECD (EC/OECD, 2023[2]), EC/OECD Science, Technology and Innovation Policy (STIP) Survey, https://stip.oecd.org/assets/downloads/STIPCompassTaxonomies.pdf; OECD (2017[1]), “The next production revolution and institutions for technology diffusion”, https://doi.org/10.1787/9789264271036-11-en; interviews with AI diffusion institutions.

Technology extension services

Technology extension services are commonly offered by research technology organisations (RTOs). These organisations are primarily concerned with developing and transferring research and technology to the private sector and society at large (Sanz-Menéndez et al., 2011[3]). Unlike other research organisations, which are chiefly driven by scientific research, RTOs are often established to provide scientific and technological solutions to the wider economy in their specific fields of competence. They are generally publicly funded (through block funding and competitive research grants) as well as industry-commissioned projects.4

An example of an RTO of this sort is Germany’s Fraunhofer Institute for Industrial Engineering IAO,5 which works with large companies and SMEs to realise the promise of AI and other emerging technologies, including blockchain, autonomous vehicles and Internet of Things (IoT) platforms. This Fraunhofer Institute transfers AI research through contractual services supporting product engineering, process development, and systems development and implementation. Over the past two years, the Institute has run around 200 projects.6 Many of these are carried out in collaboration with the Fraunhofer Institute for Manufacturing Engineering and Automation IPA7 in their jointly managed AI Innovation Centre.8 The AI Innovation Centre receives funding from the Ministry of Economic Affairs, Labour and Tourism of Baden-Württemberg and provides expert advice on getting started in AI and robotics9 free of charge for companies.

Based in Toronto, the Vector Institute10 is an autonomous not-for-profit corporation committed to AI research, specialising in ML and deep learning. It was established in 2017 with financial support from the Government of Canada, the Government of Ontario, and private sector actors in partnership with the University of Toronto. Its mission is to work with firms, research organisations, start-ups, incubators and accelerators to advance AI research and steer its application, adoption and commercialisation across Canada. The Vector Institute offers technology extension services delivered through industry sponsorships. Based on the success of the Vector Institute’s industry programmes, it is now adapting them to help SMEs through its FastLane Program,11 which gives SMEs access to talent, training and networking opportunities. The FastLane Program is federally funded and does not require payment from SMEs to join.

The Hollings Manufacturing Extension Partnership (MEP) provides technology extension services to SMEs based in the United States. These services include technology scouting and transfer, supplier scouting, business-to-business network pilots, technology-driven market intelligence and co-operative R&D (Sargent, 2019[4]). The MEP programme is housed at the US Department of Commerce’s National Institute of Standards and Technology (NIST). The programme has MEP centres in all US states and Puerto Rico and more than 1 450 trusted advisors and experts at approximately 430 MEP service locations.

Non-profit institutions, higher education institutions, US states and territories, and local and tribal governments can compete to establish an MEP Centre. The federal government may provide up to 50% of the funding necessary to create and operate a given centre. To be eligible, the MEP Centre must secure at least half of the financing through non-federal sources, such as state governments or service fees. After an SME submits a manufacturing problem statement to its local MEP Centre, staff there undertake assessments involving factory visits. In this process, they evaluate business use cases and may identify opportunities where AI could help. In implementing projects, MEP works with firms to develop AI solutions using its in-house experts or subcontractors, including research organisations and industry actors.

MEP clients, or any US manufacturer, can also apply to the MEP-Assisted Technology and Technical Resource (MATTR) service12 to work with NIST research laboratories and user facilities. This service gives companies access to further technical expertise and engineering capabilities in advanced manufacturing, collaborative robotics, cybersecurity, and information and communication technologies, among other fields, often at no cost.

The Italian Artificial Intelligence Research and Innovation Centre (AIRI)13 is an interdepartmental centre explicitly created to conduct industrial research and AI technology transfer. Its staff comprises professors, researchers and postgraduate students from the University of Modena and Reggio Emilia’s departments of Engineering, Economics, and Physics, Informatics and Mathematics. AIRI conducts basic and applied research. The latter can be considered a form of technology extension service, as it exclusively involves companies with high technology readiness and close-to-market projects that could have significant business impact.14 The Emilia-Romagna region provides subsidies equivalent to about 30% of project costs; the remainder is funded by beneficiary companies.

The United Kingdom’s National Health Service (NHS) created the NHS AI Lab15 in 2019 to help deliver the promises of AI in the health sector. The NHS AI Lab gathers government, health and care providers, academics, and technology companies to create opportunities for collaboration and technology co-creation with a view to addressing challenges in developing and implementing AI systems. It also aims to develop the use of synthetic data for AI applications in healthcare. It uses various diffusion mechanisms, including a technology extension service called “AI Skunkworks”. In the AI Skunkworks programme, NHS AI Lab experts and external solution providers work with healthcare providers (e.g. hospitals and health centres) on data-rich problems in care delivery to develop proofs-of-concept and demonstrate if and how AI can tackle them.

Grants for business R&D

Set up by United Kingdom Research and Innovation (UKRI) in 2013 and part of the Catapult Network, the Digital Catapult16 is a partly publicly funded research technology organisation that encourages early adoption of innovative digital technologies. It supports projects that might subsequently be replicated by UK businesses more broadly. Digital Catapult focuses on accelerating the adoption of virtual and augmented reality, 5G and IoT, blockchain and AI technology as individual technologies – and also in combination in emerging complex systems, such as the Metaverse and digital twins.17 One example of a supply-side programme delivered by Digital Catapult that supports AI start-ups is the Made Smarter Technology Accelerator (MSTA).18 MSTA is a matching-fund programme that invites medium and large manufacturing companies to define and scope business challenges that advanced digital technologies such as AI can solve. Digital Catapult then solicits solutions to these challenges from the United Kingdom’s technology and digital start-up ecosystem – validating their suitability and supporting the process of collaboration.

Based in Canada, Scale AI19 is a consortium of enterprises, research centres, academic bodies and high-potential start-ups dedicated to the diffusion of AI technologies. In its mission statement, it pledges to support: 1) investments in developing AI applications across supply chains; 2) the commercialisation of AI-powered solutions; 3) the AI start-up ecosystem; 4) AI skills in the workforce; and 5) the collaborative development of AI applications. It is one of Canada’s Global Innovation Clusters, supported by Innovation, Science and Economic Development (ISED) Canada.20 Scale AI provides grants for industry-led projects21 in demand forecasting, automated in-plant logistics and real-time data integration. Government funding is matched by contributions from the private sector, with Scale AI reimbursing up to 50% of expenses for approved projects (of which there are about 30 per year).

Based in the United States, the Digital Manufacturing and Cybersecurity Institute (MxD)22 aims to transform US factories by fully equipping them with the digital tools and expertise they need to reduce costs, grow and improve their operations, and become globally competitive. It is part of Manufacturing USA,23 a network of regional institutes with diverse technological focuses. MxD has invested over USD 120 million (US dollars) (about EUR 120 million [euros]) in more than 85 applied R&D projects24 in the areas of design, product development, systems engineering, future factories, agile and resilient supply chains, and cybersecurity. The development and implementation of AI solutions are often embedded in many of these projects.25 MxD awards up to USD 75 000 (about EUR 75 000) to teams composed of at least one post-secondary academic institution. MxD prioritises groups that include at least one industrial partner, which has to provide matched funding and/or in-kind contributions.

Japan’s New Energy and Industrial Technology Development Organization (NEDO)26 is a public agency supporting R&D that addresses energy and global environmental problems and develops new advanced technologies. It does not conduct its own research but formulates technology strategies and programmes and, as part of its R&D project management activities, establishes implementation frameworks combining the capabilities of public-private actors in industry, academia, and government. NEDO supports basic and applied research on high-risk, innovative technologies. It has two main R&D programmes supporting AI diffusion in firms:

• Development of Integrated Core Technologies for Next-Generation AI and Robots: Spanning the years 2018‑23, and with a 2022 budget of JPY 1.40 billion (Japanese yen) (about EUR 9.6 million), the programme assists R&D and technology demonstration in areas such as business analysis, data gathering and processing, AI model development and impact assessment. It also supports AI projects involving business inventory optimisation, decision making and improved efficiency.27

• Realisation of a Smart Society by Applying Artificial Intelligence Technologies: This programme, spanning the years 2018‑22, and with a 2022 budget of JPY 1.375 billion (about EUR 9.4 million), funded AI R&D and technology demonstration using data-acquisition sensor technologies (IoT) in three strategic sectors: health, medical care and welfare, and mobility.28

The French Ministry of Ecology is launching an AI and Green Transition Programme.29 This initiative seeks to support AI demonstrators in reducing carbon emissions and addressing other environmental challenges in public services and the public sector. Eligible projects must be led by regional governments, municipalities, and other parts of the public administration and must involve local companies or research organisations. Projects must have a budget of at least EUR 1 million for developing AI systems capable of making recommendations, forecasting or decision making. Examples of expected proposals include raising energy efficiency in buildings, visual image analysis to detect unauthorised waste disposal and optimising public transportation services. The programme will support between 50‑70% of applied R&D expenses and 25‑45% of experimental development expenses, depending on the size of the beneficiary firm.30 Projects involving collaboration with one or more SMEs or research organisations receive more public funding.

Two diffusion institutions providing technology extension services also manage business R&D grants. Fraunhofer IAO organises and co-ordinates consortia for collaborative research projects – involving industrial and research partner organisations – funded by the German government and the European Union. For instance, its AI Innovation Seeds31 programme gathers a group of 5‑12 firms to explore new AI approaches to address challenges of common interest, with funding from the Ministry of Economic Affairs, Labour and Tourism of Baden-Württemberg. The NHS AI Lab runs an AI in Health and Care Award,32 which has committed around GBP 90 million (about EUR 108 million) in over 70 awards for companies to accelerate the testing and evaluation of strategic AI technologies for healthcare. The award aims to speed up real-world applications by helping build an evidence base demonstrating their effectiveness and safety.

AI Singapore33 is a national AI programme launched by Singapore’s National Research Foundation and hosted by the National University of Singapore. One of the programme’s primary missions is to support AI adoption in firms, and to this end, it manages a suite of diffusion mechanisms.34 Its 100 Experiments (100E)35 flagship initiative provides business R&D grants to solve firms’ AI challenges and help them build their own AI teams. A company can apply by submitting a problem statement that cannot be readily tackled using third-party applications and that could be solved within 9 to 18 months by Singapore’s ecosystem of AI researchers and engineers. AI Singapore provides selected projects with up to SGD 250 000 (Singapore dollars) (about EUR 179 000) for Singapore’s universities and research institutes to work with companies. Beneficiaries provide matching funds and in-kind contributions (i.e. staffing in AI disciplines, engineering resources, etc.). AI Singapore assigns its staff to an engineering team that joins the project and develops an AI minimum viable model. Such staff typically includes full-time AI engineers, data scientists and apprentices from the AI Apprenticeship Programme (see below).

Business advisory services

Besides offering technical expertise, diffusion institutions can provide non-technical guidance to managers and executives to support AI adoption. Cap Digital,36 for example, is a French association of companies specialising in digital technologies (including AI). It provides fundraising support and business coaching services to help its members learn how to pitch to different audiences and seize opportunities in foreign markets. Cap Digital also provides R&D support services to member firms interested in applying to regional, national and European calls for grant proposals and tenders. It helps companies prepare applications by providing expert assessments and supporting partner search (such as other companies and research institutions) to form grant consortia. Cap Digital is funded by a mix of public and private sources, including the French national government and regional councils of the Paris Region and Hauts-de-France, as well as through membership and professional services fees.37

Diffusion institutions can offer business advisory services in combination with other AI diffusion mechanisms. Fraunhofer IAO, for example, complements its technology extension services with business guidance. Business and public sector bodies can commission feasibility studies, market and trend studies, strategy development and organisational design. Diffusion institutions can also combine grants for business R&D with business advisory services to address weaknesses in business technology upgrading efforts and financial constraints (OECD, 2017[1]). For example, Digital Catapult’s start-up acceleration programme “FutureScope” delivers an AI initiative called the Machine Intelligence Garage,38 which provides access to computational resources to early-stage start-ups with high growth potential to help develop and test new products and services. Such resources include cloud credits for partners such as Amazon Web Services and Google Cloud, as well as specialised and independent hardware support and advice from partners such as Graphcore and Nvidia. The programme gives practical guidance to help AI start-ups build sustainable and ethical solutions. It also provides access to fundraising opportunities, as well as technical support from industry leaders who have partnered with the programme.

Scale AI also has an AI Acceleration39 programme that supports Canada’s SME and start-up ecosystem. It does not fund individual companies but instead gives financial support to organisations that fund firms, including incubators, accelerators, innovation centres, corporate labs and open innovation initiatives. Eligible organisations also provide support services such as coaching, mentorship, customer and business development assistance, intellectual property and commercialisation assistance, product expansion and fundraising support. These organisations can receive up to CAD 50 000 (Canadian dollars) (about EUR 37 000) from Scale AI for each supported start-up working to build applied AI products and services for supply chains. Applicable costs eligible for reimbursement include wages and salaries for activities and expenses related to equipment, labs, facilities, supplies and materials.40

AI Singapore has a framework that helps companies and other organisations assess their existing capabilities and opportunities to adopt AI. Further, it helps them identify obstacles that need to be overcome to achieve more advanced readiness (Table 4.2). Under this framework, AI Singapore provides business advisory services delivered through workshops and six-week AI solutions development projects.

Table 4.2. AI Singapore’s AI Readiness Index

	AI-unaware (Less than 1)	AI-aware (1 to 1.9)	AI-ready (2 to 2.5)	AI-competent (More than 2.5)
General capabilities	Might hear about AI but is unaware of applications	Savvy consumers of AI solutions. Capable of identifying use cases for AI applications	Capable of integrating pre-trained AI models into products or business processes	Capable of developing customised AI solutions for specific business needs
General characteristics	Wait for vendors to convince use cases and business value of AI	Identified potential use cases and seeks AI solutions from vendors	Evaluated the viability of pre-trained AI models	Developed a roadmap for AI implementation
AI adoption suitability	Consume ready-made, end-to-end AI solutions	Integrate pre-trained AI models and solutions for common AI applications	Integrates pre-trained AI models and solutions for common AI applications	Develop customised AI model for unique business needs

Note: Numbers under each category represent average score ranges.

Source: AI Singapore (2024[5]), AI Readiness Index (AIRI), https://aisingapore.org/airi.

Grants for applied public research

In addition to providing financial support to companies, diffusion institutions can fund AI research conducted in universities or public research institutes. This research is generally performed in collaboration with industry actors and can support high-risk research or the development and implementation of technologies close to commercialisation. Canada’s National Research Council (NRC) - University of Waterloo Collaboration on AI, IoT and Cybersecurity41 is an example of such a programme. The university works with companies in Ontario to develop promising AI technologies that do not have a clear path to commercialisation. The average commercialisation timeframe usually ranges between five to ten years, depending on the project.42 The German Research Centre for Artificial Intelligence (DFKI),43 funded by the Federal Ministry of Education and Research, conducts “human-centric” AI research in the search for technology and application breakthroughs. It hosts public-private research partnerships with software, automotive and manufacturing companies. A quarter of DFKI activities in a given year involve work with industry actors. Commercialisation of the research it supports is typically ten years away at least.44

Besides supporting applied AI R&D in businesses, NEDO also operates two programmes supporting applied research led by universities and public research institutes:

• Technology Development Project on Next-Generation Artificial Intelligence Evolving Together with Humans spans the period 2020‑24 and, in 2022, had a budget of JPY 2.68 billion (about EUR 18 million). This programme supports the development of interactive AI systems that work together with humans. More specifically, the programme will support research that: 1) facilitates human understanding of AI decisions and decision-making processes; and 2) develops mechanisms for human inputs to improve the inference accuracy of AI.45

• Development of AI-Based Innovative Remote Technologies spans the period 2021‑24, and in 2022, had a budget of JPY 500 million (about EUR 3.5 million). This programme funds R&D on AI for extended reality systems that comprehensively and accurately depict remote environments and transmit information visually, aurally and through haptics.46

AI Singapore manages two programmes supporting close-to-market public research:

• The AI Grand Challenge47 initiative funds research in collaboration with the public sector that aims to solve national challenges. For example, the “AI in Health Grand Challenge” supported research to enhance primary healthcare and disease management in Singapore. The “AI in Education Grand Challenge,” co-organised with the Ministry of Education, will follow a similar model to support mother-tongue language learning for primary-level students in Singapore.

• The AI Technology Challenge48 aims to develop innovative AI solutions that can be adopted in government and business sectors that are strategically relevant to Singapore. Funded research projects are conducted in collaboration with a government office or an industry partner.

AI Singapore’s Technology Offers49 promotes science-industry collaborations to create new products or services to support the commercialisation of the research results obtained from these programmes. Its catalogue offers more than 15 AI solutions developed from past research for firms to adopt. These cover a range of business sectors, including healthcare, biochemistry, manufacturing and transportation.

Networking and collaborative platforms

Diffusion institutions can bring together firms, higher education and research institutions, and other public and private entities to facilitate collaboration around AI diffusion. They provide networking services to match the supply and demand for AI technologies and applications and promote a collective pool of knowledge to increase participants’ productivity, innovation and competitiveness.

Networking and collaborative platforms sometimes have a regional focus. Based in Italy, the Siena Artificial Intelligence Hub (SAIHub)50 aims to gather AI SMEs, large companies and research actors in the Tuscany region. To attract talent, it creates partnerships with the University of Siena to propose scholarships and cash prizes51 for students who, after completing the master’s degree course or doctorate, begin their professional activity at one of the more than 30 companies of the SAIHub Network. The hub also promotes AI services and solutions offered by member SMEs.52 Based in Montreal, Forum IA Québec53 aims to support the region’s AI ecosystem. To this end, it offers several informational and other resources to help firms adopt AI. For instance, its open directory54 of AI actors includes information on consulting firms, solution providers, research and technology transfer institutions and venture capital funds. The directory also includes a collection of AI use cases and funding opportunities. Forum IA Québec also conducts assessments of the performance of the region’s AI ecosystem to inform policy decisions.

Other platforms operate at a national level. Plattform Lernende Systeme,55 for example, brings together AI specialists from science, industry, government and civic organisations to promote adoption and inform policy makers and other stakeholders. It was set up by the German Federal Ministry of Education and Research and managed by the German National Academy of Science and Engineering (acatech). The Plattform hosts a variety of working groups in areas such as the future of work, healthcare and mobility to examine the prospects, challenges and prerequisites for developing socially responsible AI. It has introduced a national Map on AI56 that includes information on more than 1 100 use cases and hundreds of research institutions, knowledge intermediaries and study programmes. It also provides various forms of business intelligence, including market and technology analyses and thematic reports. Based in London, techUK57 is a trade association that gathers individuals, businesses, government, and stakeholders to deliver the promises of digital technologies. It includes AI as one of its six technology focus areas and has about 500 member SMEs (technology providers). The association is active in: 1) analysing and formulating policy proposals for the adoption of digital technologies; 2) promoting the use of technologies in business sectors such as financial services, defence, manufacturing, utilities and consumer electronics; and 3) monitoring emerging trends in technologies and innovation (e.g. digital twins).

Some of the interviewed institutions use other diffusion mechanisms to foster opportunities for networking and establishing collaborations:

• The Vector Institute offers services that help companies identify and hire AI talent through its FastLane programme. It hosts the “Digital Talent Hub” online platform that links employers with skilled AI talent seeking employment. The Institute organises recruiting events, executive networking events and research symposia.

• The MEP’s Supplier Scouting Service58 helps manufacturing SMEs connect to suppliers with the right technical and production capabilities. The service operates on a national, regional and local level to connect suppliers with purchasers higher up the supply chain, including larger companies and government agencies.

• Enterprises, research centres, education institutes, start-ups and other actors in the Canadian AI ecosystem can join Scale AI as associates at no cost. This membership allows them to benefit from networking opportunities, including matchmaking and informational events.59

• Cap Digital, mentioned in the previous section, organises more than 100 events each year with more than 1 000 members to provide networking opportunities and promote their innovative technologies and services.

• AI Singapore’s The Epoch60 web portal aims to be a digital platform supporting the country’s AI ecosystem of students, teachers, apprentices, professionals and SMEs. The site, open to all and free of charge, seeks to create networking opportunities, host exchanges around learning and applying AI in the workplace, and publish community-contributed articles and job opportunities.

On-the-job training

While training services are not a mechanism central to the interviewed diffusion institutions, many recognise that firms often struggle to adopt AI due to the lack of skilled staff. This deficiency can limit what companies can gain through AI technology extension services or grants for business R&D. To tackle this obstacle to adoption, some institutions propose courses or training for professionals as a complementary diffusion mechanism:

• The Vector Institute offers training courses61 to raise management and technical staff skills and improve awareness of AI applications. Some courses invite business leaders to analyse real-world AI use cases and identify opportunities and challenges underpinning successful adoption. The Vector Institute likewise hosts applied and research internships62 that allow participants to work alongside AI engineers and researchers.

• As part of its statutory activities, the MEP facilitates training (offering courses in-house), supports new or existing apprenticeships, and provides access to information and experts that can help address workforce needs and skills gaps (Sargent, 2019[4]). Examples of AI training courses targeting SMEs include those offered by North Carolina MEP and South Carolina MEP.63

• Scale AI provides training support for working professionals by covering half their registration fees for more than 180 accredited courses proposed by partner training programmes.64 It also offers grants for companies to develop tailored on-the-job training courses65 for their employees, covering up to 85% of CAD 100 000 (about EUR 74 000) in eligible expenses.

• MxD’s Virtual Training Centre66 is an online platform that assists manufacturers and workers in skills development. It gives access to more than 1 000 free and paid courses on cutting-edge technologies (including 48 centred on AI) offered by Google, Microsoft, Amazon Web Services and other leading technology companies.

• Through its AI Apprenticeship Programme (AIAP)®,67 AI Singapore aims to nurture the country’s AI talent and expand job opportunities in AI-related fields. To apply, candidates must demonstrate a baseline skill set in data science and intermediate programming. Selected apprentices follow a two-month mentoring and self-directed learning scheme. Afterwards, they are assigned to AI Singapore projects (including 100E, mentioned above) for seven months to work on industry projects and thereby gain practical knowledge in building and deploying AI models. During this time, apprentices receive a training allowance ranging between SGD 3 500 and SGD 5 500 (EUR 2 500 and EUR 4 000). AI Singapore has also launched several training and instruction programmes68 at various levels that address different audiences, including courses for students, educators, and workers. Some of these courses seek to increase the skills of prospective applicants to the AI Apprenticeship Programme.

Information services and open-source code

The AI diffusion mechanisms covered in prior sections aim to raise AI capabilities in firms. However, diffusion institutions can also support the development of standalone data platforms and open-source code that firms and other organisations can readily use to develop AI applications. For example, Mobility Data Space69 is an online marketplace for automotive, transportation, logistics and many other types of data relevant to the mobility sector. It works as a data matchmaking service by providing a digital infrastructure for secure, peer-to-peer or one-to-many transactions. The data are exchanged for various purposes, including for companies to develop autonomous driving and mobility AI.70 Data sellers can establish a price tag and define terms and conditions, such as the data’s intended use, via technical and legal data usage policies.

Diffusion institutions frequently combine information services and open-source code with other mechanisms. For instance, the premise of Digital Catapult’s Machine Intelligence Garage,71 mentioned earlier, is to provide access to computational resources to assist early-stage AI start-ups. AI Singapore compiles and publishes various open-source tools developed through AI Apprenticeships and other programmes in its AI Ready Bricks72 collection. Code is freely accessible to other AI engineers and companies and comes with information on use cases, tutorial videos, and other resources to help with reuse.

After developing proofs-of-concept through its AI Skunkworks programme, the NHS AI Lab publishes a working version of the related software code in a publicly accessible GitHub repository73 under an open-source licence. This resource allows healthcare providers and companies supplying AI solutions to learn about the approaches undertaken with full technical details and reuse and build upon the code in their applications.

Key barriers to AI adoption identified by diffusion institutions

In their day-to-day activities promoting AI, diffusion institutions have identified a series of obstacles that hinder adoption. These are described in the following sections.

Digitalisation is a necessary first step

Before adopting AI, firms must embrace digital technologies that systematically gather data from business processes and customer and supplier interactions. AI applications need an accurate digital representation of business processes to make accurate predictions and prescriptions. However, many firms lag in adopting digital technologies. According to Sarah Gagnon-Turcotte, director of Forum IA Québec, “Adopting AI is the final step in organisations’ digital technology adoption pipeline of 5‑10 years, typically starting with Enterprise Resource Planning.”

Insufficient understanding of AI

Typically, companies that approach diffusion institutions have implemented some degree of digitalisation and have at least a superficial understanding of what AI is and what it can do for them. They generally recognise that AI can play an important role in their core business processes. Manufacturing and information and communication technology (ICT) firms have a general (though often limited) awareness of AI’s potential benefits and applications. However, for other business sectors, use cases and applications are less clearly established, and as a consequence, investing in AI is perceived as too risky. In addition, firms that are capable of and interested in adopting AI tend to be concentrated geographically in regional clusters.

Even when firms have some familiarity with AI, managers often do not have a sufficient grasp of what AI is, what adoption entails, or what their businesses can gain from it. Managers are often confronted with the “black box” problem of AI, i.e. opacity in how the AI makes decisions or recommendations. They usually have a plug-and-play conception of adoption, i.e. they expect AI to be a commodity technology they can easily integrate into core business processes. Furthermore, technicians, whose training is based on understanding mechanisms and their workings, often believe that AI is unnecessary or does not offer value to their business. Given this insufficient understanding, managers and technicians can mistrust AI’s predictions, recommendations or (even more so) decisions derived from the data.

ROI is difficult to estimate

Companies must invest considerable time and resources to adopt each use case and tailor the AI application to their specific needs and conditions. Successful off-the-shelf solutions that firms can obtain from third parties are rare. To illustrate this point, Scale AI’s Julien Billot compares AI today with “where the Internet was in 1995”. Back then, setting up a website involved hiring specialised software engineers. Today, there are plenty of solutions for non-experts to build complex online portals.

Similarly, only a few third-party AI applications are currently available for firms. These applications allow companies to add AI features into Software-as-a-Service for generic use cases. “Companies try to buy AI applications as off-the-shelf licence-based solutions, only to find these do not work or only yield limited results,” Valter Fraccaro and Riccardo Valletti from SAIHub point out. AI applications are generally ad hoc and subject to the firm’s specific work environments and processes. Companies use diverse software and systems to manage business operations, production lines, services, accounting systems and other functions that need to be integrated when developing AI solutions. AI projects involve an important degree of experimentation, where the ROI is inherently uncertain. This happens even for well-established use cases.

AI applications need a proven record of success, especially regarding economic impacts, to convince firms to invest. Firms (particularly SMEs) that engage with diffusion institutions are often uncertain about what they can gain financially from implementing AI. They can find it challenging to define and delimit the business case for adoption. Companies often try to tackle complex problems with AI, making it difficult to estimate an ROI that might materialise several years later. Finding reliable estimates of the ROI can also be difficult, even when applications are narrowly defined. For instance, an ROI estimate in a predictive maintenance application relies on how well the counterfactual can be calculated. An AI system can, for example, alert users to the possibility of a machine breaking down, prompting a firm to service the machine for maintenance. But it is difficult to determine if this intervention was truly necessary and that the firm has indeed avoided a breakdown (with its ensuing costs). An historical record of breakdowns could help to estimate the ROI of such an AI system, but such data may not be readily available.

While it can sometimes be relatively straightforward to estimate cost savings and efficiency gains, it can be more challenging to calculate the ROI for new AI-enabled products, services or business models. Service providers selling AI solutions also face ROI-related problems, as the right revenue model can be unclear (e.g. subscription, licence, or charging per task as some cloud computing companies do). Different companies can use AI in many ways, making it hard for service providers to decide how to charge for it. Service providers are often uncertain about how companies will use their services. For example, some companies might use AI often and benefit from a subscription model. Other customers might prefer a pay-per-task model if their usage is sporadic.

A lack of access to AI skills

Identifying, scoping and implementing AI applications requires a mix of technical and domain expertise involving employees with MSc or PhD diplomas. The presence of AI-skilled staff is often a baseline criterion for venture capital funds to invest in firms developing AI applications. However, as described in prior literature (see Chapter 2), diffusion institutions confirm that access to AI talent can be highly constrained, especially for SMEs. Smaller firms compete for limited AI specialists and data engineers with postgraduate education with large multinationals, including tech giants such as Amazon, Google and Microsoft, which can offer more attractive salaries and work conditions. SMEs also have more limited access to on-the-job training opportunities that can help staff build AI skills. Countries also compete for talent at the postgraduate level, e.g. by offering higher PhD salaries. Diffusion institutions often express regret that there are not enough AI-skilled students and graduates.

Insufficient data maturity

Besides AI skills, diffusion institutions note a recurrent problem: firms often do not have the necessary data streams to develop AI applications. As mentioned above, data are essential to create, test, evaluate and validate AI models. However, companies approaching diffusion institutions for support often do not have sufficient data in terms of quantity, quality, cleanness and structure. They frequently lack an adequate understanding of what information needs to be gathered systematically. Consequently, they may not have the necessary data collection mechanisms in place or, if they do, struggle to assess how appropriate their data are for a given AI use case. Collecting high-quality data comes at a cost (which also needs to be factored into the ROI estimation). For instance, some manufacturing firms (particularly SMEs) may struggle to afford to install sensors in every factory, production line and machine they operate.

In addition to collecting the necessary data, firms also face data management challenges. They often have to be able to integrate data from different sources such as software, machines, business areas within the firm and data provided by third parties. Data sources can vary in periodicity (e.g. weekly, daily, hourly), type (e.g. quantitative or qualitative) and format (e.g. Excel spreadsheets, MySQL databases). Data can be unstructured, unlabelled and disorganised, making it challenging to integrate. Preparing data to build an AI model takes expertise and considerable time. According to AI Singapore’s Laurence Liew, “Companies tend to underestimate efforts in getting data ready for AI applications.” Companies can also struggle to access the necessary computing resources and cloud services. Deep learning, for example, is computationally expensive.

The data needed for AI applications can sometimes be outsourced, as automotive and transport sector companies do via the Mobility Data Space. However, data transfer and exchange also have obstacles. For instance, companies may have concerns about inadvertently selling personal data collected from customers (risk of data leaks74). For fear of losing the value of the data they collect, companies are sometimes unwilling to sell it or enter collaborative projects that exploit it. Data security (i.e. avoiding data breaches) and regulatory compliance are other emerging concerns when discussing data transactions. Some enterprises are reticent to publish data or the results from work with technology extension services or from funding for applied research. This aversion can make partnering with research institutions that manage such diffusion mechanisms difficult, as academics want to publish their research. Companies need to take this, and other researchers’ needs and interests into account.

Uncertainties around regulatory compliance

Diffusion institutions also confirm that firms struggle to navigate complex regulatory and ethical landscapes. Unaware of the applicable regulations and legislation, they often fear possibly unknown legal risks and inadvertently becoming non-compliant. Companies, and particularly SMEs, are often overwhelmed when dealing with existing and emerging regulations, such as the European Union’s General Data Protection Regulation, as well as data acts and acts pertaining to AI. Managers often have questions about quality assurance requirements for AI applications and where liability lies if damage occurs to customers using AI-enabled products or services.

Insufficient knowledge of privacy regulations can make companies overprotective of data, rendering it challenging to explore and develop AI applications. Some diffusion institutions mentioned examples of drawn-out data-sharing requests for AI proofs-of-concept, which can take up to 12 months. Such a long wait time is excessive for start-ups and SMEs acting as AI service providers. The “black box” problem also means that biases in training data, breaches in regulation or ethical implications can go undetected unless the firm has specialised data engineers and staff trained in regulatory and ethical issues. Firms may hesitate to introduce new AI-enabled services or products, especially in public sectors like healthcare and education, due to uncertainties regarding how well the public will accept them.

Difficulties scaling up AI applications

Many companies run proofs-of-concept without later deploying them as full-scale AI solutions, even if they were successful. An AI pilot application should be a milestone demonstrating the potential benefits of adoption. However, firms often run pilots without a strong vision or business plan to scale them up and integrate them with core business processes. According to diffusion institutions, firms often do not realise (or are unprepared to make) the shifts in organisational structure, business processes and culture needed to adopt AI solutions. Compared to adopting other digital technologies, adopting AI in core business processes can require a significantly larger company-level transformation, involving changes to business operations across various departments that managers lacking AI literacy often fail to foresee. “Large firms often misconceive AI as an add-on rather than a revolution,” says Sarah Gagnon-Turcotte from Forum IA Québec. AI systems are not a one-off project. It is often the case that companies also fail to understand the extent of continuing investments required for AI quality management. Keeping AI models performing well over time requires constant assessment, retraining (with the most recent data) and redeployment.

Companies can also suffer from a “Not Invented Here” syndrome75 when third parties are involved in developing AI applications. If insufficiently engaged or reassured, employees may refuse to co-operate, feeling anxiety about their jobs being made redundant by the AI system.76 For example, one diffusion institution observed a case of internal pushback against a third-party application used in wholesale trade. In this instance, a new AI solution successfully predicted the demand for a company’s product portfolio and performed better than its current system for projecting demand. However, deployment stagnated as the firm’s information technology (IT) department defended its in-house system, raising conflicting interests in the company.

Effective forms of support for AI adoption identified by diffusion institutions

Many of the approaches used to support the adoption of AI described in this chapter start with an early assessment of firms’ digital and AI capabilities, e.g. addressed in the eligibility criteria for grants for business R&D, in technical visits that are a part of technology extension services and in workshops providing business advice. AI Singapore uses self-assessment tools to help companies evaluate their capabilities and determine the support they need. AI diffusion institutions usually select to work with firms with the right initial capabilities and where AI is or can be part of the company’s core business. For companies that are not sufficiently digitally mature, many governments have a separate suite of policy instruments offering dedicated support for digitalisation.77

Various diffusion institutions report that they select only projects with a clear path to increases in performance, product or service quality, or cost reductions. They explain that this makes the achievement of tangible impacts in proofs-of-concept more likely, which helps to convince firms to scale up investments. Conversely, other institutions consider that firms should not obsess about the ROI from the outset and instead value experimentation that may lead to breakthroughs. In addition, other impacts might be sought besides raising productivity. Charles Huot from Cap Digital illustrates this point with the example of wind turbines equipped with cameras and AI to protect avian wildlife by autonomously reducing blade speed.

Diffusion institutions agree that catalogues of applications, use cases and success stories can help firms understand the possible gains from AI. Such catalogues help establish a proven record of success. They can also document experiences other businesses can learn from, such as what did not go well initially and how obstacles were overcome. In particular, case studies that measure the economic impact of investments (e.g. in terms of sales increases or cost reductions) can help tackle concerns around the ROI. In this way, catalogues can help managers better grasp the opportunities, challenges and limitations presented by AI. AI solution providers could also refer to such catalogues to better understand industry needs and adjust their service offerings. Some diffusion institutions included in this chapter are compiling such catalogues (Table 4.3).

Table 4.3. Sample catalogues of AI applications, use cases and experiences

Institution and web link	Short description
Digital Manufacturing and Cybersecurity Institute (MxD) https://www.mxdusa.org/projects/	A selection of past and ongoing projects funded by the Institute, including information on participants, problem statements, proposed solutions and impacts
Plattform Lernende Systeme https://www.plattform-lernende-systeme.de/map-on-ai-map.html	A map covering AI applications that have been developed in Germany
Forum IA Québec https://vitrine.ia.quebec/en/studies	A database of case studies of AI projects in the Quebec region
Fraunhofer IAO / IPA https://www.ki-fortschrittszentrum.de/de/projekte.html	A database of over 70 case studies of publicly funded projects providing expert advice
NHS AI Lab https://nhsx.github.io/skunkworks/	A repository of projects supported by the NHS AI Lab

Source: Authors.

Many factors driving the success of support for AI diffusion are specific to the type of mechanism used by diffusion institutions, as described in the following sections.

Technology extension services

A recurrent point across the interviews is that companies should avoid thinking of AI as a technology in search of a solution. Instead, they should focus on delineating and describing their business problem and then assess the added value that AI might bring. Understanding the relevant opportunities offered by AI and properly framing the business problem provides the groundwork for determining what the company can gain from adoption, what data need to be collected, how data should be managed and how the AI model needs to be built. “As was the case for electricity in the early 20^th century, AI won’t be adopted for its own sake but for the innovations it enables,” says Sarah Gagnon-Turcotte from Forum IA Québec.

In implementing technology extension services, the interviews suggest that diffusion institutions should work with firms following a sequence of steps:

1. establishing one or more business cases describing how to apply AI (for instance, clarifying how autonomous forecasting, decision support or decision making would help)

2. scoping possible AI solutions and assessing data maturity (for example, is the business gathering and processing the correct data?)

3. developing pathways to implementation.

Recommendations for each of these steps are presented below:

1. As mentioned earlier, use case analysis is a helpful tool to advance a base understanding of AI in firms. However, diffusion institutions need to actively link past experiences to firms’ specific needs and culture. To establish a business case for AI adoption, diffusion institutions need to obtain as much operational data from firms as possible.

2. The staff of diffusion institutions need to spend time at the firm to assess its digital maturity and simulate what an AI solution would do. Proof-of-concepts should start by tackling more straightforward problems using readily available data. Staff should also estimate the ROI for a more extensive implementation project and help firms decide whether to invest in it. To this end, diffusion institutions highlight the need to have an economist join data engineers and other technical experts in technology extension projects.

3. An implementation roadmap should comprehensively describe what deploying the proof-of-concept as a fully integrated AI solution across the organisation entails. There are often significant impacts across various business processes and departments (e.g. accounting, purchasing and production). The roadmap should also describe how to ensure AI models perform well over time. The vision for implementation should be co-developed with staff from the outset to secure their co‑operation.

Technology extension services reportedly work best when beneficiaries assign their own staff and contribute in-kind resources. Projects can also involve other types of actors (e.g. universities and research institutes). These collaborative projects can be valuable, particularly in pre-commercial stages involving AI applications that have not yet been introduced in any market. In various instances, diffusion institutions can offer technology extension services to multiple firms facing shared business problems.

Business advisory services

According to interviewees, business advisory services can be particularly effective in three main ways. First, they can help firms make initial estimations of the ROI using scenario analysis without necessarily going into the technicalities of AI. For instance, advisors can help managers estimate the downtime of machines or production lines and the financial savings to be made using predictive maintenance. Second, diffusion institutions can help raise awareness and understanding of any public support for AI adoption offered at national and international levels (e.g. EU calls). Firms are often unaware of such support, including funding opportunities. Firms can also benefit from advisory services that help them prepare applications for assistance and maximise their chances of obtaining a favourable response. Third, diffusion institutions can offer business advisory workshops to raise AI literacy in managers. They can also provide advice on ethics and regulation.

Networking and collaborative platforms

Similarities often exist in companies’ business problems and how they use AI to solve them. Seminars and conferences can facilitate exchanges between business executives and help raise understanding of the opportunities AI presents and the types of transformation that firms need to make. Seminars and conferences also facilitate networking between managers, researchers, trade associations, diffusion institutions, AI solution providers and other actors. Such events can help AI reach business sectors where adoption tends to be lower. Furthermore, they can systematically gather the views of stakeholders in order to inform and shape policies and regulations for AI.

Grants for business R&D and applied public research

Financial support reduces the risks entailed in developing proofs-of-concept and exploring theoretical applications. Some grant schemes ask firms to indicate the expected ROI or cost reduction in their funding application as part of their allocation criteria. Diffusion institutions can provide guidance in this connection. For example, they can help businesses estimate savings and sales projections. They can also keep track of such estimates and see if they materialise over time. Financial support can also help firms build a digital infrastructure for collecting, managing, and processing data for AI, e.g. support for deploying IoT technologies. Some business sectors, such as adtech and fintech, already use AI intensively. However, when used to help acquire third-party AI applications, grants can encourage other business sectors to work with AI solution providers. According to the interviewed diffusion institutions, grants that deliver the best outcomes require beneficiaries to match public support with their own resources (financial or in-kind). Similarly, publicly funded research projects produce the best results when companies assign their staff to the research team.

On-the-job training

Training courses are essential for existing employees to gain the technical knowledge required for AI adoption. Tools for self-assessment of digital maturity, like AI Singapore’s AI Readiness Index, can also be used as training tools for managers, venture capitalists and solution providers to learn to identify use cases and design business models for AI solutions. Managers and technicians can also be trained in information governance, regulations and ethical issues. Such training can help tackle compliance and AI assurance concerns that often stop firms from using their data or prevent them from engaging with AI altogether. While on-the-job training can help firms address the scarcity of workforce skills in AI in the short term, various diffusion institutions are of the view that countries need to step up efforts in embedding AI across tertiary education.

Information services and open-source code

Open-source tools make AI methods and resources accessible to a broad audience beyond AI specialists and computer scientists. Statisticians, data engineers, physicists, and other professionals with varied backgrounds can work more readily with such tools than by developing algorithms from scratch. Diffusion institutions use open-source resources together with other mechanisms, such as on-the-job training and technology extension services (AI Singapore and NHS AI Lab).

Publicly funded infrastructures that subsidise computing resources (e.g. hardware or computation credits on the cloud) and provide real or synthetic training data for free or at a low cost can be particularly helpful for SMEs. Such resources also need to be combined with other forms of support, such as business advice. For example, Digital Catapult’s Machine Intelligence Garage, described earlier, gives SMEs access to computational resources in combination with mentorship and fundraising opportunities. By verifying the parties’ identities and ensuring the integrity of data transfer, digital platforms and online marketplaces provide a trustworthy channel for secure data transfers. According to Moritz Stober (acatech – National Academy of Science and Engineering), firms tend to underestimate the opportunities to establish data partnerships to tackle common problems, especially those involving competitors.

Conclusion

This chapter reports findings from structured interviews with 19 institutions across the G7 countries and Singapore that work to increase the uptake of AI. It has not sought to establish a comprehensive picture of the entire institutional landscape supporting AI adoption across these countries. Rather, it samples and characterises diffusion mechanisms and gathers institutions’ views on the main barriers to the adoption of AI in firms and the most effective forms of support for adoption. Future work could use this chapter’s findings to survey a larger number of institutions supporting AI adoption.

The diffusion institutions interviewed for this chapter confirm the obstacles identified in prior literature. They emphasise uncertainty over the ROI as a critical obstacle for firms considering adopting AI. Managers often struggle to grasp how AI can solve real problems in the workplace. They also tend to underestimate the implications of deploying AI solutions, which often involve significant changes in business culture and processes across many (if not all) parts of the firm. A lack of AI skills and data maturity are fundamental barriers to implementing AI. Moreover, uncertainties about regulation can prevent firms from taking steps towards adoption.

This chapter identified seven main mechanisms that diffusion institutions use to help firms overcome the challenges of adoption.

1. Technology extension services can help firms delineate and define the business problem to be solved and develop proofs-of-concept that demonstrate how AI can help.

2. Grants for business R&D remove part of the risk associated with AI investments.

3. Business advisory services provide non-technical assistance that can raise managers’ understanding of their firms’ AI readiness and the specific opportunities and challenges that AI entails.

4. Grants for applied public research which can help promote high-risk research and/or the development and implementation of technologies close to commercialisation.

5. Networking and collaborative platforms help build AI ecosystems of public and private actors, creating demonstration effects and opportunities for knowledge transfer.

6. On-the-job training can help address bottlenecks around AI skills.

7. Information services and open-source code provide helpful resources for firms seeking to strengthen their AI capabilities.

Diffusion institutions often combine these mechanisms to optimise their impact (see Box 4.2).

Box 4.2. Synergies across AI Singapore’s programmes

AI Singapore’s “AI for industry” training courses are structured to prepare professionals for the AI Apprenticeship Programme. Apprentices, in turn, are embedded in other programmes (e.g. 100E, AI Ready Bricks). The results from R&D supported by AI Singapore are available for firms to adopt and extend through collaboration and licensing opportunities.

AI Singapore selects highly motivated and self-directed individuals who have already independently acquired data science and AI skills. This selection of candidates who already possess skills but lack real-world experience allows them to give the apprentices two months of deep skilling, after which they can put them to work on a hands-on project within a team. In AI Singapore’s 100E programme, companies invest up to SGD 180 000 (30% cash, 70% in-kind) to implement a project with visible ROI and deployment as the objective. AI Singapore matches the contribution in kind with AI engineering resources (for a total of SGD 360 000), ensuring commitment on both sides. The programme is designed for companies to derive tangible benefits, including AI models that can be deployed nearly immediately into production for an immediate ROI.

While they would like to help every Singaporean and every Singaporean company, they cannot do so. So, they use the AI Readiness Index to identify and work only with AI-ready companies. For companies that are AI-unaware or AI-aware or individuals who are not ready for the apprenticeship, they point them to other programmes. This way, they avoid over-taxing their team in trying to help everyone or every company that comes to them.

Source: Liew (2024[6]).

While diffusion institutions have developed sophisticated mechanisms tailored to AI technology adoption, they can only support a small fraction of the population of firms that could benefit from AI. The number of firms these institutions serve tends to range between 10 and 400 annually, depending on the diffusion mechanism.78 The relatively limited scale of the public AI diffusion mechanisms considered here contrasts with the ambitious scope of national strategies and aspirations for the widespread adoption of AI that governments from the G7 countries and beyond have laid out.

Policy makers also have to consider the additionality of impacts from their spending – that is, evaluating outcomes compared to what would have occurred without public support. As noted earlier, firms often self-select into programmes offered by diffusion institutions, raising the question of whether these firms might have eventually adopted AI through other channels. Currently, there is no evidence that this possibility has been tested rigorously. Further research could measure the additionality achieved by diffusion institutions, including the magnitude of demonstration (or spillover) effects. One approach might involve identifying control groups of firms similar to those receiving support at the time when successful candidates are accepted into programmes. A more elaborate method could involve randomised control trials. While such research may be costly, the expense might be justified if policy makers intend to expand these diffusion programmes significantly.

References

[5] AI Singapore (2024), AI Readiness Index (AIRI), https://aisingapore.org/airi.

[2] EC/OECD (2023), EC/OECD Science, Technology and Innovation Policy (STIP) Survey, edition, https://stip.oecd.org/assets/downloads/STIPCompassTaxonomies.pdf.

[7] Larrue, P. and O. Strauka (2022), “The contribution of RTOs to socio-economic recovery, resilience and transitions”, OECD Science, Technology and Industry Policy Papers, No. 129, OECD Publishing, Paris, https://doi.org/10.1787/ae93dc1d-en.

[6] Liew, L. (2024), , https://aisingapore.org/innovation/100e/.

[8] MEP (2020), MEP Annual Report FY 2020, Manufacturing Extension Partnership.

[1] OECD (2017), “The next production revolution and institutions for technology diffusion”, in The Next Production Revolution: Implications for Governments and Business, OECD Publishing, Paris, https://doi.org/10.1787/9789264271036-11-en.

[3] Sanz-Menéndez, L. et al. (2011), Policy Brief - Public Research Organisations, https://www.researchgate.net/publication/287595871_Policy_Brief_-_public_research_organisations.

[4] Sargent, J. (2019), The Manufacturing Extension Partnership program, Congressional Research Service, https://catalog.libraries.psu.edu/catalog/30866457.

Notes

← 1.  For instance, the US-based Hollings Manufacturing Extension Partnership (MEP), one of the diffusion institutions studied in this chapter, estimates that for every dollar of public investment, the programme generates USD 26.20 in new sales growth and USD 34.50 in new investment in the supported firms (MEP, 2020[8]).

← 2.  The OECD AI Observatory can be accessed at https://oecd.ai/.

← 3.  The Science, Technology and Innovation Policy Compass can be accessed at https://stip.oecd.org.

← 4.  For a survey on the activities, governance and funding modalities of RTOs, see Larrue and Strauka (2022[7]).

← 5.  See Fraunhofer Institute for Industrial Engineering IAO at www.iao.fraunhofer.de/en/about-us/fraunhofer-iao.html.

← 6.  Author's communication with Fraunhofer IAO staff, 28 April 2022.

← 7.  More information about the Fraunhofer Institute for Manufacturing Engineering and Automation IPA can be found at www.ipa.fraunhofer.de/en/about-us/institute-profile.html.

← 8.  The AI Innovation Centre’s homepage is located at www.ki-fortschrittszentrum.de/en.html.

← 9.  More information about this service is available at www.iao.fraunhofer.de/en/press-and-media/latest-news/expert-advice-on-getting-started-in-AI-and-robotics.html.

← 10.  See Vector Institute’s About page at https://vectorinstitute.ai/about.

← 11.  More information about the FastLane Program is available at https://vectorinstitute.ai/fastlane-program.

← 12.  See MEP-Assisted Technology and Technical Resource (MATTR) at www.nist.gov/mep/mattr.

← 13.  AIRI’s homepage is located at www.airi.unimore.it.

← 14.  Author's communication with AIRI staff, 2 August 2022.

← 15.  See NHS AI Lab’s website at https://transform.england.nhs.uk/ai-lab.

← 16.  See Digital Catapult’s About page at www.digicatapult.org.uk/about.

← 17.  Author's communication with Digital Catapult staff, 10 May 2022.

← 18.  More information about the accelerator is available at https://accelerator.madesmarter.uk.

← 19.  Scale AI’s website is located at www.scaleai.ca.

← 20.  Other Global Innovation Clusters focus on digital technologies, protein industries, advanced manufacturing and the ocean economy. See https://ised-isde.canada.ca/site/global-innovation-clusters/en.

← 21.  More information on these projects can be found at www.scaleai.ca/projects.

← 22.  See MxD’s website at www.mxdusa.org.

← 23.  Manufacturing USA’s portal can be accessed at www.manufacturingusa.com.

← 24.  More information on these projects can be found at www.mxdusa.org/projects.

← 25.  Author's communication with MxD staff, 27 May 2022.

← 26.  See NEDO’s website at www.nedo.go.jp/english.

← 27.  See the programme page at www.nedo.go.jp/english/activities/activities_ZZJP_100138.html.

← 28.  See the programme page at www.nedo.go.jp/english/activities/activities_ZZJP_100137.html.

← 29.  The Ministry provides more information at https://greentechinnovation.fr/les-acteurs-de-lia.

← 30.  Smaller firms have a higher share of public support.

← 31.  More information on this programme is available at https://s.fhg.de/KI-Fortschrittszentrum-AI-Innovation-Seed.

← 32.  See AI in Health and Care Award at https://transform.england.nhs.uk/ai-lab/ai-lab-programmes/ai-health-and-care-award.

← 33.  AI Singapore’s website is available at https://aisingapore.org.

← 34.  Besides AI diffusion, AI Singapore also supports basic research that can advance strategic technologies and address societal challenges (see, e.g. AI Governance Research Grant Call, AI Research Grant Call, and AI Kickstarter Grant Call).

← 35.  More information about this programme is available at https://aisingapore.org/industryinnovation/100e.

← 36.  See Cap Digital’s About page at www.capdigital.com/en/who-we-are/our-mission .

← 37.  Author's communication with Cap Digital staff, 11 April 2022.

← 38.  See Machine Intelligence Garage at https://migarage.digicatapult.org.uk.

← 39.  See Scale AI’s AI Acceleration programme at www.scaleai.ca/acceleration.

← 40.  See www.scaleai.ca/acceleration/the-programs-were-investing-in.

← 41.  More information on this partnership can be found at https://uwaterloo.ca/news/university-relations/waterloo-and-nrc-reaffirm-partnership-future-facing.

← 42.  Author's communication with NRC staff, 12 August 2022.

← 43.  See DFKI’s website at www.dfki.de/en/web.

← 44.  Author's communication with DFKI staff, 26 July 2022.

← 45.  See the programme website at www.nedo.go.jp/english/activities/ZZCD_100016.html.

← 46.  See the programme website at www.nedo.go.jp/english/activities/activities_ZZJP_100194.html.

← 47.  See the programme website at https://aisingapore.org/grand-challenges.

← 48.  See the programme website at https://aisingapore.org/technology/technology-challenges.

← 49.  See the programme website at https://aisingapore.org/innovation/technology-offers.

← 50.  SAIHub’s About page is available at https://saihub.org/service/missionevision.

← 51.  These prizes are described at www.quinewssiena.it/siena-programma-stayhub-assegnati-premi-studenti.htm (in Italian).

← 52.  Author's communication with SAIHub staff, 2 May 2022.

← 53.  See Forum IA Québec’s website at https://forumia.quebec/en.

← 54.  This directory is available at https://vitrine.ia.quebec/en/directory.

← 55.  Plattform Lernende Systeme’s homepage is at www.plattform-lernende-systeme.de/home-en.html.

← 56.  The Map can be accessed at www.plattform-lernende-systeme.de/map-on-ai-map.html.

← 57.  See techUK’s About page www.techuk.org/who-we-are/about-us.html.

← 58.  More information on this service can be accessed at www.nist.gov/mep/supplier-scouting.

← 59.  See Scale AI’s events at www.scaleai.ca/events.

← 60.  The web portal is available at https://epoch.aisingapore.org.

← 61.  See Vector Institute’s training courses at https://vectorinstitute.ai/programs-courses.

← 62.  For more information on these internships, see https://vectorinstitute.ai/internships.

← 63.  For North Carolina MEP, see https://ncmep.org/lean-helping-small-manufacturers-test-artificial-intelligence. Information on South Carolina MEP is available at www.scmep-online.org/courses/introduction-to-machine-learning-and-artificial-intelligence.

← 64.  See Scale AI’s partner training programmes at www.scaleai.ca/education/individuals.

← 65.  See Scale AI’s training courses at www.scaleai.ca/training/businesses-how-to-apply-for-funding.

← 66.  More information on the Virtual Training Centre is available at www.mxdusa.org/vtc.

← 67.  See the programme website at https://aisingapore.org/industryinnovation/aiap.

← 68.  Information on these programmes is available at https://learn.aisingapore.org.

← 69.  See Mobility Data Space at https://mobility-dataspace.eu.

← 70.  Author's communication with Mobility Data Space staff, 29 August 2022.

← 71.  See the programme website at https://migarage.digicatapult.org.uk.

← 72.  These resources are available at https://aisingapore.org/aiproducts/ai-ready-bricks.

← 73.  This repository can be accessed at https://nhsx.github.io/skunkworks.

← 74.  Unauthorised data transmission to a third party may happen in various ways, such as through file transfers or unlawful system access (hacking).

← 75.  The “Not Invented Here” syndrome refers to the tendency to avoid ideas, services, products or business solutions developed outside the organisation.

← 76.  Several diffusions institutions indicated that adopting AI generally does not lead to staff layoffs. Rather, AI systems often free up employees to engage in more productive work.

← 77.  For an overview of policies supporting the adoption of digital technologies, see www.youtube.com/watch?v=xwAtBd40pSQ.

← 78.  Some diffusion mechanisms are more resource-intensive than others. For example, institutions that deliver technology extension services or grants for applied public research tend to work with fewer firms per year compared to those hosting networking and collaborative platforms.