DIGITAL PROGRESS AND TRENDS REPORT 2025 Case Study 3 CO-WORKER, COACH, OR COMPETITOR? How AI Is Transforming the Future of Digitally Delivered Services DIGITAL PROGRESS AND TRENDS REPORT 2025 Case Study 3 Co-worker, Coach, or Competitor? How AI Is Transforming the Future of Digitally Delivered Services Johan Bjurman Bergman and He Wang © 2025 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW, Washington DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org This work is a product of the staff of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy, completeness, or currency of the data included in this work and does not assume responsibility for any errors, omissions, or discrepancies in the information, or liability with respect to the use of or failure to use the information, methods, processes, or conclusions set forth. The boundaries, colors, denominations, links/footnotes, and other information shown in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. The citation of works authored by others does not mean The World Bank endorses the views expressed by those authors or the content of their works. Nothing herein shall constitute or be construed or considered to be a limitation upon or waiver of the privileges and immunities of The World Bank, all of which are specifically reserved. Rights and Permissions The material in this work is subject to copyright. Because The World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2625; e-mail: pubrights@worldbank.org. Cover images: AI professions set: artificial intelligence workers, machine learning, software engineers, digital data analyst, robotics specialists, scientists. Flat vector illustration isolated on white background, by Good Studios, from Adobe Stock. Further permission required for reuse. Cover design: Veronica Elena Gadea / World Bank Group Interior design: William Pragluski, Critical Stages LLC Contents Main Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix Purpose and Scope of the Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Background and Literature Review. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 DDS drive development and job creation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 DDS and AI are increasingly intertwined. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 AI is disrupting DDS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 AI in DDS: A framework for considering opportunities and risks. . . . . . . . . . . . . . . . . . . . 16 Country Cases. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 The Philippines: Mature lower-skilled exporter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Uzbekistan: Emerging builder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 Policy Considerations and Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Policy-relevant barriers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Policy-relevant risks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 Policy considerations by country archetype. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 Toward a framework for addressing the impact of AI adoption on DDS. . . . . . . . . . . . . . . 59 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 iii iv  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Boxes CS3.1 AI and productivity of software engineers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 CS3.2 Stock market suggests AI will profoundly impact BPM outsourcing firms in the Philippines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 CS3.3 The double-edged sword of AI-powered accent neutralization tools in the Philippines. . 27 CS3.4 Spotlight on Viet Nam: A future hub for AI-driven DDS? . . . . . . . . . . . . . . . . . . . . . . 35 CS3.5 Building a spoken language corpus for Uzbek LLMs as a foundation for local AI tools. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 CS3.6 Private and public-private initiatives to address AI skills shortage in Uzbekistan. . . . . 51 CS3.7 Upskilling software developers for outsourcing in an AI-enabled world. . . . . . . . . . . . 52 CS3.8 Potential impact of increased economic uncertainty and protectionism on DDS trade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 Figures CS3.1 GenAI use in the United States, by industry, December 2024. . . . . . . . . . . . . . . . . . . . 10 CS3.2 AI impacts on productivity, employment, outsourcing opportunities, and wages in DDS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 BCS3.1.1 AI’s impact on software development productivity. . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 CS3.3 Four-archetype framework based on DDS exports, digital skill level, and intensity of GenAI use for select countries, 2023 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 CS3.4 ChatGPT visits per internet user in the Philippines and other regions, May 2025. . . . 21 CS3.5 AI exposure in EAP, by country . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 CS3.6 AI adoption impact on jobs as seen by IT and business process firms in the Philippines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 CS3.7 Job postings in the Philippines, by occupation, 2022–24 . . . . . . . . . . . . . . . . . . . . . . . 23 BCS3.2.1 Stock prices of large contact centers and BPM firms, 2020–25. . . . . . . . . . . . . . . . . . . 26 CS3.8 Employees and revenues of companies registered with IT Park Uzbekistan, 2017–24. . . 38 CS3.9 ChatGPT visits per internet user in Uzbekistan and other regions, May 2025. . . . . . . 40 CS3.10 Quarterly job postings by occupation in Uzbekistan, 2023–24. . . . . . . . . . . . . . . . . . . 40 CS3.11 High-level framework for shaping a response to the impact of AI on DDS. . . . . . . . . . 60 Map CS3.1 DDS exports as a percentage of gross domestic product, 2023. . . . . . . . . . . . . . . . . . . . 7 Table CS3.1 DDS and commonly outsourced tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Main Messages Most workers and firms surveyed in this study view narrow and generative artificial intelligence (GenAI) tools as co-workers and coaches, with few seeing them as competitors. Providers of digitally delivered services (DDS) in the Philippines and Uzbekistan generally see artificial intelligence (AI) as a means to enhance work efficiency, quality, and learning. Most discount the threat of AI as a competitor that would displace them and their sectors while also acknowledging that AI may take over many simple tasks. Many acknowledge the potential need for large-scale up-skilling and re-skilling of workers to adapt to jobs and tasks changed by AI. However, such efforts have yet to materialize. Workers and firms are adopting AI tools, but most use only a fraction of their capabilities. This is especially true for smaller, less-well-resourced firms. Contributing factors include the limited availability of specialized GenAI tools for specific tasks, the limited awareness of how AI could help solve specific problems, and gaps in in-house capability to evaluate and integrate solutions. Firms say they need to identify use cases and integrate the right tools themselves—or build custom systems—which demands product, data, and engineering skills that many lack. Taken together, these factors can slow AI adoption and make it more costly. AI is driving productivity and quality enhancements for tasks ranging from simple to complex. Tasks range from writing emails in a second language using GenAI chatbots, automating password resets using machine learning (ML) and natural language processing (NLP), and automating the translation of patient logs to medical codes using ML and NLP to writing code assisted by a GenAI copilot, training new staff using ML-powered adaptive simulations, or developing artwork for computer games using GenAI tools. Signs of AI-driven job displacement are appearing in DDS. However, firms say the displacement is still modest. Nevertheless, because it seems to affect primarily lower-skilled and junior workers, adoption of AI tools at scale in DDS could disproportionally affect youth and poorer groups, potentially exacerbating inequality. Indeed, postings for DDS jobs in some economies have not rebounded after the drop seen after ChatGPT was introduced in November 2022. AI seems to be redistributing value creation in DDS toward larger, multinational, and higher-skilled players. Large firms leverage advanced, custom AI systems to boost efficiency and pursue higher- value projects, widening the gap between them and smaller competitors relying on off-the-shelf solutions. Multinational firms, supported by proprietary global tools and mandatory training, integrate responsible AI practices to capture more value than local firms lacking similar resources. Higher-skilled sectors, such as software development, amplify worker productivity and innovation with AI, while lower-skilled sectors risk greater automation and fewer upskilling opportunities, limiting their potential to capture value. The nascent transformation from software as a service v vi  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S to services as software, in which AI agents deliver complete outcomes, is likely to accelerate this redistribution of value. Technology-driven firms seem to be adjusting more quickly than labor-intensive firms. Technology- driven firms—including information technology services and software development firms—may have more to gain from AI adoption given their capabilities to develop and sell AI solutions that complement their offering. Meanwhile, labor-intensive firms—including contact center and business process management firms—may have more to lose from AI disruption, at least in the short term, because they are less likely to have in-house capabilities to develop and sell AI solutions and because such AI solutions may compete with their existing offering. If the latter group is unable to proactively develop AI solutions and upskill staff to use these as co-workers, it may exacerbate labor market disruptions by making transitions sudden and unplanned. The inability of governments and regulators to adapt to the rapid pace of AI innovation may prevent them from mitigating AI-driven increases in economic inequality. Key barriers include access to talent that can build AI and work alongside it, policy uncertainty, and insufficient digital and data infrastructure. The perceived lack of coherent and swift government response is slowing down AI adoption and prompting private sector action. This issue implies that firms with resources to spend on training and digital transformation will capture the benefits of the AI transition, while smaller ones are left behind. In turn, progress on AI readiness of the digital economy may slow down, potentially exacerbating inequality. The perceived lack of a coherent government and regulatory response to rapid advances in AI may be slowing adoption and blunting both productivity gains and the management of disruptions. Because government action is perceived as incoherent and slow, AI adoption is becoming hesitant and piecemeal, as well as skewed in favor of large firms. In a context where firms must fend for themselves, progress depends on firm-level capacity: Those able to secure AI-capable talent, navigate policy uncertainty, and invest in adequate digital and data infrastructure move ahead, while smaller firms lag. The result is fragmented, firm-by-firm adoption that slows overall AI readiness and heightens the risk of widening inequality. Acknowledgments This case study was undertaken by Johan Bjurman Bergman (Digital Specialist, World Bank, and lead author) and He Wang (Economist). The team is grateful for the valuable guidance of Yan Liu (Senior Economist) and the teamwork and considerable contributions from World Bank Digital teams in the Philippines, including Naoto Kanehira (Senior Digital Specialist), Mitch Abdon (Senior Digital Consultant), and Jamie Guerrero Javier (team assistant); in Uzbekistan, including Sandjar Saidkhodjaev (Digital Development Specialist); and in Viet Nam, including Huong Tran (Senior Digital Specialist); as well as global team members Sharmista Appaya (Senior Digital Specialist) and Sandra Sargent (Senior Digital Specialist). The team also expresses its gratitude to the IT and Business Process Association of the Philippines and IT Park Uzbekistan for their invaluable collaboration in facilitating data collection, as well as to the firms, workers, and policy makers who dedicated their time to share their perspectives and insights. The team is grateful to Yegor Denisov-Blanch and Simon Obstbaum of the Software Engineering Productivity Research Group at Stanford University for generously sharing their insights, as well as to Liubov Guryeva for excellent editorial assistance. vii Abbreviations abbreviation definition AI artificial intelligence ASEAN Association of Southeast Asian Nations BPM business process management BPO business process outsourcing DDS digitally delivered services EMDE emerging and developing economy FDI foreign direct investment GDP gross domestic product GenAI generative AI HICs high-income countries HR human resources IBPAP IT and Business Process Association of the Philippines ICT information and communication technology IDEs integrated development environments IP intellectual property IT information technology LICs low-income countries LLM large language model LMICs lower-middle-income countries MICs middle-income countries ML machine learning NLP natural language processing Q quarter QA quality assurance R&D research and development RN registered nurse UMICs upper-middle-income countries VoIP voice-over-internet protocol WTO World Trade Organization ix Purpose and Scope of the Study The rapid advancement of artificial intelligence (AI) is reshaping the global landscape of outsourced digitally delivered services (DDS), raising questions about the evolution of the human workforce in this industry. Characterized by large pools of relatively low-cost, lower-skilled workers performing routine tasks, DDS industries—such as contact centers and business process management (BPM)— are now facing a potential technological inflection point. The future profile of the DDS workforce and the division of labor between the industry’s human workforce and AI will be actively shaped by today’s policies. This study contributes to the policy-oriented discussion on the impact of AI on DDS by exploring how AI is transforming outsourced DDS, with a focus on labor market implications in developing and emerging economies that rely heavily on this sector for job creation and export revenue. This study complements the Digital Progress and Trends Report (DPTR) 2025 by addressing a gap in the research on AI use at work in export-oriented DDS and developing economies more generally. It combines a literature review with qualitative insights from two case studies of firms producing and exporting DDS in the Philippines and Uzbekistan. It highlights how firms and workers use and develop AI tools, the impacts they are seeing, and the barriers they face in scaling up AI adoption. The case studies include semistructured interviews with representatives of 40 digital services export firms, 20 workers, and 15 government and other private sector stakeholders in the Philippines and Uzbekistan. In the Philippines, interviews were conducted in firms employing more than 30,000 contact center and BPM workers, 9,000 health care services workers, more than 7,000 workers in information technology (IT) services and software development firms, and about 100 workers in animation and game development in offices in metro Manila, Taguig, Ortigas, and Iloilo. In Uzbekistan, interviews were conducted in firms that employed more than 1,200 employees in contact centers and BPM, more than 800 employees in IT services and software development, and more than 300 employees in specialized logistics services in Tashkent, Urgench, Jizzakh, Samarkand, and Khiva. This report is structured as follows. The “Background and Literature Review” section briefly introduces DDS, reviews the literature on AI’s impact on this sector, and lays out an analytical framework for understanding the impacts of AI on productivity, employment, outsourcing opportunities, and wages. It also introduces four country archetypes to guide policy-oriented discussion on AI and DDS. The “Country Cases” section zooms in on the case study countries, offering a brief history of the evolution of their DDS and outlining their characteristics as digital services exporters and the use of AI across subsectors. It also includes insights from firm interviews. The “Policy Considerations and Summary” section tackles barriers to AI adoption and policy- relevant risks, pinpoints key areas for a policy response to AI’s impacts, and outlines a high-level policy action framework to support policy makers as they navigate this landscape in the coming years. 1 Background and Literature Review Introduction This section defines digitally delivered services (DDS), traces their rise via digitalization and outsourcing from high-income to developing markets, maps key segments and markets, and highlights how DDS and artificial intelligence (AI) are increasingly intertwined. It highlights the enabling foundations for AI-enabled DDS and the gaps that keep many low-income countries from participating at scale. Finally, it lays out a framework for thinking about how targeted investments can turn DDS growth into inclusive jobs in an AI era. DDS drive development and job creation DDS have been transforming how businesses operate for the past 70 years. Beginning in the 1950s, some of the earliest examples of digital services delivery involved companies in the United States installing large onsite computers to manage tasks such as payroll processing and manufacturing process automation. In the intervening decades, the introduction of smaller and faster computers and increased access to high-speed internet connectivity expanded the scope and scale of what services could be delivered via digital means. DDS still has no single accepted definition but includes a wide range of services delivered through digital means. The WTO et al. (2023) defined the term to include service sector activities delivered through computer networks, such as the internet, apps, emails, voice and video calls, and digital intermediation platforms. This encompasses digital services such as information technology (IT) and software development services, as well as traditional services delivered digitally, such as digital marketing, voice and nonvoice customer support, back-office operations, and sector-specific support services. DDS are increasingly outsourced from high-income countries (HICs) to developing countries. In the early 2000s (Amiti and Wei 2024), services began to be delivered remotely and digitally, both domestically and from overseas, giving rise to services outsourcing—the practice of contracting tasks or services to third-party providers. Outsourcing helps firms reduce costs, access specialized skills, flexibly scale operations, and provide around-the-clock support. 3 4  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S In recent years, digitalization and business process modernization efforts by firms in HICs have driven growth in outsourcing. In turn, digitalization has increased the complexity of outsourced tasks, and outsourcing providers are now expected to offer not only skilled staff but also domain expertise and technology platforms (Arora et al. 2021). Table CS3.1 provides an overview of key segments of DDS that are often outsourced. Global outsourcing sector revenue is estimated to exceed US$700 billion, and the sector creates millions of good jobs. Leading outsourcing providers include Accenture, India’s Tata Consultancy Services and Infosys, and Capgemini, which collectively generated approximately US$135 billion in revenue and employed 2 million people worldwide in 2024. Through sophisticated business networks and offshore-onshore delivery models, outsourcing firms provide a wide range of IT services, digital consulting, and business process outsourcing (BPO). This has made them essential nodes in the DDS value chain, channeling the expertise of key outsourcing hubs—such as India and the Philippines—to meet the demand of firms located mostly in advanced economies. Global demand for DDS is primarily driven by firms in HICs. Although both businesses and individuals contribute to the demand, firms account for the majority of DDS buyers, particularly in IT and business services. Individual users primarily consume digital services in media, entertainment, e-commerce, and personal finance, with demand in developing countries rising as internet access expands. Because of their substantial market size in services, HICs play a significant role in both the demand and the supply of digital services. Indeed, with around two-thirds of developing countries’ DDS exports destined for HICs, they are the primary importers of DDS from developing countries. Outsourcing has helped DDS become the fastest-growing segment of international trade. Global exports of DDS have increased fourfold since 2005, reaching US$4.25 trillion in 2023, up 9.0 percent year-on-year—outpacing goods (4.8 percent) and other services (4.6 percent) to account for more than 54 percent of total services exports (WTO 2024). In 2022, HICs accounted for 76 percent of DDS exports, whereas developing countries contributed 24 percent, with notable growth from 19 percent in 2010 (UNCTAD 2023). Economies such as India (6.0 percent of global DDS exports), China (4.9 percent), Poland (1.0 percent), the Philippines (0.7 percent), Romania (0.5 percent), and Malaysia (0.3 percent) contribute significantly to DDS exports (refer to map CS3.1). In addition, countries such as Ghana, Morocco, Pakistan, South Africa, and Uzbekistan have seen strong growth in recent years, highlighting the growing importance of the sector as a driver of well-paying jobs in developing countries (WTO et al. 2023). Low-income countries (LICs) are not yet able to participate meaningfully in DDS trade, mainly because of limited access to digital infrastructure and skills. The 44 least-developed countries continue to lag, cumulatively contributing just 0.2 percent of global DDS exports in 2023, whereas all African countries cumulatively contributed about 0.9 percent (WTO et al. 2023). The divergence between middle-income countries (MICs) and LICs is primarily driven by limited access to reliable power supplies, high-speed internet, sufficiently advanced computers, and skills, which increase the cost of delivering digital services and create barriers to scaling up to meet demand. In some cases, these challenges are exacerbated by fragility and conflict, making reliable delivery even more difficult. TABLE CS3.1  DDS and commonly outsourced tasks Commonly outsourced Primary outsourcing Key client Key segment Example task Key provider market subsegment driver market Australia, Software-as-a-service Designing, coding, and testing Cost, specialized European Union, development software platforms competence India, Japan, Software United States development: End-to-end development of a Designing, Custom application mobile app, website, or internal tool, Cost, flexibility coding, testing, development including UI and UX China, Eastern Europe and maintaining software Maintenance, bug fixes, and (Poland, Ukraine), United States Software maintenance applications and software testing and quality Cost India, Latin America (nearshoring to and testing systems assurance (Argentina, Brazil, Latin America) Specialized development in AI and Mexico), Viet Nam Emerging technology Specialized ML, AR and VR, cloud integration, or development competence IoT Handling user support tickets and Cost, focus on core Help desk support trouble-shooting competency IT services: Infrastructure and Monitoring servers, networks, and Specialized Provision, network management data centers remotely competency European Union Eastern Europe management, (nearshoring to Set up antimalware software, Specialized (Poland, Romania, and support of IT Eastern Europe), Cybersecurity services managed security monitoring, threat competence; around- Ukraine), India infrastructure and United States B ackground and L iterature R eview    systems intelligence, and incident response the-clock support Manage cloud architecture and Specialized Cloud services DevOps competence Building of dashboards, regular Business intelligence reports, and performance metrics Access to talent, cost Eastern Europe, India and reporting Data analytics: tracing Collecting, Large-scale data cleaning, validation, European Union, Data entry, processing, processing, and migration, annotation, and Access to talent, cost Kenya, Philippines United Kingdom, and storage analyzing data to warehousing United States extract business Developing predictive models (<50 percent) insights Advanced analytics and or performing complex analysis, Access to talent, cost India AI modeling delivering AI services (vision, language, decision support) (Continued) 5 TABLE CS3.1  DDS and commonly outsourced tasks (Continued) 6   Commonly outsourced Primary outsourcing Key client Key segment Example task Key provider market C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S subsegment driver market Content creation, search Developing blog posts, articles, and Eastern Europe, India, Cost, specialized engine optimization, website content optimized for search Philippines, freelance competence and marketing engines, manage search campaigns platforms Social media Posting, community management, Cost, specialized management, email, paid social advertising, email Eastern Europe, India Creative services: competence and web marketing campaigns, engagement analysis Online marketing, animation, and Character animation, logos, videos, European Union, India, Malaysia, game development Animation complex modeling, special effects, Cost, speed Japan, United Philippines short-form, visuals for digital ads States Programming games in specialized Eastern Europe, Latin European Union, engines; developing characters, Cost, specialized Game development America, Southeast Japan, United environments, and in-game objects; competence, speed Asia States conducting playtests Inbound customer service calls, Africa (Egypt, Arab Contact center and Call center services outbound telemarketing, and tech Cost Rep.; Morocco; South European Union business process (voice based) support calls Africa), Colombia, (to Eastern management: Costa Rica, India, Email handling, live web chat support, Europe), United Customer-facing Mexico, Philippines, Nonvoice social media customer care, chat- Cost Kingdom, support and sales Poland, Romania based technical support help desks United States services and non- customer-facing Accounting, finance and payroll, Eastern Europe; Egypt, (Philippines, administrative and Business process HR services, procurement, claims Arab Rep.; Ghana; India; nearshoring to Cost Latin America) operational support management processing, and document Kenya; Latin America; management Nigeria; Philippines Order and freight processing: Shipment tracking and customer Cost, around-the-clock China, India, United Logistics service, fleet management, logistics, support States finance, and accounting Specialized Medical coding and Translating patient charts into United Kingdom, services: Requiring Cost India, Philippines billing standardized codes and creating bills United States significant domain- specific knowledge Front-office tasks (for example, scheduling and reminders, telehealth Patient care Cost, specialized United Kingdom, follow-ups) and back-office tasks (for India, Philippines coordination competence United States example, managing patient data, referrals, and care plans) Source: Original table for this publication. Note: AI = artificial intelligence; AR = augmented reality; DDS = digitally delivered services; DevOps = software development and IT operations; HR = human resources; IoT = internet of things; IT = information technology; ML = machine learning; UI = user interface; UX = user experience; VR = virtual reality. B ackground and L iterature R eview   7 MAP CS3.1  DDS exports as a percentage of gross domestic product, 2023 Egypt: Business services, ICT services, Romania: ICT services and Uzbekistan: IT services and and nancial services other business services software development Philippines: Business process outsourcing Digitally delivered services exports, 2023 (% GDP) (50, 200] (20, 50] (10, 20] (5, 10] (2, 5] (1, 2] (0.5, 1] Kenya: Business and nancial services, India: Business process outsourcing, [0, 0.5] Brazil: Business services and telecommunication services, IT services, and software development No data software development data processing, and labeling Sources: Data from World Bank, DDS exports data from WTO, and GDP data from IMF. Note: DDS = digitally delivered services; GDP = gross domestic product; ICT = information and communication technology; IMF = International Monetary Fund; IT = information technology; WTO = World Trade Organization. Asian countries still capture most of the DDS outsourcing market, but Eastern Europe and Latin America are growing, with a rapid expansion forecast in Africa in the coming years. India is the global leader, with a strong focus on IT services and software development, employing more than 5 million people, whereas the Philippines is the global leader in contact centers and BPO, with more than 1.6 million employees. Eastern Europe and Latin America are emerging as competitive destinations for a range of services. Argentina, Brazil, Mexico, Poland, Romania, and Ukraine are all grabbing market shares in IT services and software development, driven by competitive costs, skilled labor pools, and favorable time-zone alignments with the European and US markets. Africa has a nascent digital services outsourcing sector, employing about 1.2 million people (Yieke 2024). However, recent forecasts indicate that countries such as the Arab Republic of Egypt, Ghana, Kenya, Morocco, Nigeria, and South Africa are poised for significant growth in the next 5 years, driven by increasing investments in digital infrastructure and talent development that could see the sector add up to 1.5 million new jobs. The growth in DDS exports has been enabled by investments in infrastructure and supportive domestic and international laws and policies. Investments in reliable power, affordable high-speed internet and digital devices, improved competition policy, strengthened regulation of intellectual property (IP), data privacy and security, and cybercrime, as well as investment in worker training, have laid the foundations for growth. In addition, generous fiscal incentives, the relaxation of foreign ownership laws, and the creation of technology parks and free zones that provide reliable access to inputs, coupled with export promotion initiatives and engagement of diaspora workers, have been key to attracting talent and foreign direct investment (FDI). Finally, strengthened policy frameworks to facilitate cross-border digital transactions and cross-border data flows, which are the vehicles for exporting DDS, as well as efforts to increase the availability of cloud services, have been fundamental to enabling the rapid increase in digital services trade (World Bank 2021). 8  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Development benefits to the economies that export DDS are significant, particularly for women and youth. Digital services can help drive countries’ integration into global value chains, expand their technological capabilities, and support economic diversification. With women making up about 54 percent of the Philippine BPO sector workforce, which is about 1.5 million strong, and about 34 percent of the Indian IT workforce, they are among the direct beneficiaries of DDS growth (World Bank and WTO 2023). The following impacts of DDS exports on exporters’ economies have been well documented (WTO 2023). • Job creation and increase in export revenue. India’s IT and business process industry employs more than 5 million people, and the Philippine BPO sector employs more than 1.6 million people just in contact centers. Strong demand is also driving firms to seek talent outside of large cities, supporting peri-urban job creation and shared prosperity. With the BPO market valued at about US$280 billion (Grand View Research 2025) and the IT outsourcing market approaching US$600 billion—and with growth rates of about 8 percent per year in both sectors—outsourcing economies have—and still can—significantly increase their export revenues by tapping into these segments (Statista 2025). • Technology diffusion. Firms providing FDI—such as foreign-owned DDS firms—play a crucial role in the diffusion of digital technologies. For example, in the Philippines, foreign-owned service firms that have access to high-speed broadband use 3 times as much data and software per worker as domestic-owned firms with high-speed broadband (World Bank 2024a). • Workforce upskilling. First, exposure to advanced technologies and best practices from international clients help local firms refine processes and adopt innovative solutions and enable workers to learn by doing (World Bank 2024a). Second, meeting global standards incentivizes continuous workforce development and upskilling. Third, ongoing collaboration in cross-border projects provides real-time feedback, spurring iterative improvements and product innovation. For example, the Philippine BPO sector has gained expertise and technical know-how by working closely with multinational partners. This report focuses on four types of DDS, each requiring different types and levels of skills and competencies: • Software development and IT services, including custom application development, software maintenance, testing emerging tech development, and IT services, including help desk support, infrastructure and network management, cybersecurity services, and cloud services, demand advanced technical skills such as programming and problem-solving. • Specialized services, including health care and logistics, include segments that require specialized domain knowledge and interpersonal skills, as well as more repetitive tasks. • Creative services rely on creativity and specialized design skills. • Contact centers and BPM typically require fewer specialized skills and involve many repetitive tasks but often also require interpersonal skills. DDS and AI are increasingly intertwined The AI boom has driven a deeper integration between AI and DDS. Although AI systems have been powering commercial DDS for at least 40 years (Newquist 1994), the emergence of deep learning techniques in the early 2000s and large language models in the late 2010s made AI tools far more efficient and effective. By 2016, commercial investment in AI had risen to more than US$8 billion a B ackground and L iterature R eview   9 year (Lohr 2016), and big technology firms such as Google, Facebook, Apple, Amazon, Microsoft, and Baidu made their existing digital services (maps, search, translation, and so forth) AI-enabled (Lewis-Kraus 2016). Since the launch of ChatGPT in November 2022, workers and firms across the globe providing DDS have been able to use off-the-shelf generative AI (GenAI) tools and build solutions on top of available GenAI models. Today, popular GenAI tools have more than a billion users, and usage timing indicates they are being applied to work tasks by people across HICs and MICs (Liu and Wang 2024). The AI value chain begins upstream with scarce capital‑intensive inputs, and much of the value creation is concentrated in HICs. Specialized chips, vast computer clusters, proprietary databases, and specialized talent are increasingly controlled by a handful of suppliers of graphics processing units, such as NVIDIA, and big tech platforms that generate and monetize data. Korinek and Vipra (2024) have warned that this concentration of computing power, data, and talent could lock many developing‑country firms into low‑value niches unless policy makers expand affordable access to these inputs and foster open standards. Midstream, a small group of foundation‑model labs (OpenAI, Google DeepMind, Anthropic, and so forth) converts those inputs into large models; the high fixed cost of pretraining and the low marginal cost of deployment reinforce oligopoly power and drive vertical integration with cloud and chip vendors. Downstream, system integrators and service exporters in developing economies fine‑tune these models or build task‑specific applications—chatbots, code migration tools, AI copilots—for global clients; here, value accrues to firms that can pair local market knowledge with rapid deployment, translating models into productivity gains for end users. A more detailed discussion of the AI value chain is available in the Digital Progress and Trends Report 2025 (World Bank 2025). Firms providing DDS across the globe are using, integrating, and developing AI tools and solutions. Firms can use AI-enabled tools in their internal operations, such as recruitment, human resources management, payroll, and training. They can integrate AI into the workflows, tools, and software they use to deliver services and solutions to clients, for example, by integrating a GenAI copilot into their customer relationship systems to write better emails in a second language, natural language processing to summarize and analyze the success of customer calls, and machine learning-powered accent neutralization tools to improve clarity of client interaction or by integrating GenAI into support programming. They can also develop AI-enabled solutions and tools for clients, for example, by creating specialized customer service chatbots, developing predictive analytics systems to forecast firm revenues, or creating computer vision AI tools to improve quality assurance (QA) for products such as avocados and mechanical parts. Increased development and use of AI also engender new data privacy and security risks and raises  cybersecurity and ethical issues. These are discussed in greater detail in DPTR 2025 (World Bank 2025). DDS industries are leading AI adoption, although significant variation exists among segments and occupations. Two representative surveys of individual GenAI use in the United States conducted in August and December 2024 indicate that about one-third of employed respondents had used AI at work at some point since the introduction of such tools in November 2022. However, the studies estimated that GenAI assisted in only about 1–5 percent of all work hours (Bick, Blandin, and Deming 2025; Hartley et al. 2024). 10  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Jobs in computer and mathematical occupations show active AI usage rates of 50–60 percent, with the information industry exceeding this level (Hartley et al. 2024). At the other end of the spectrum are office and administration occupations, which include data entry and information processing workers and customer service representatives with usage rates of around 35 percent (US BLS 2024) (refer to figure CS3.1). These survey results demonstrate a high correlation with predictions of AI exposure, such as those done by the International Monetary Fund as well as researchers at OpenAI, suggesting that such assessments could be a valuable tool for thinking about future effects in developing economies (Cazzaniga et al. 2024; Eloundou et al. 2024). AI is disrupting DDS Although AI is already affecting productivity, employment, outsourcing patterns, and wages of DDS, the pace, scale, and timing of impacts in the coming years remain uncertain. Early indications of impacts combined with the rapid pace of both AI adoption and improvement in AI capabilities make it clear that changes are on the horizon (Klein 2025). However, historically, the full adoption of general-purpose technologies—such as electricity and the internet—has been slowed down by the resistance of powerful incumbents, because such technologies require fundamental reconfigurations of organizational structures, processes, and relationships. It is plausible that although AI’s capabilities will develop rapidly, firms’ capacity for social, organizational, and institutional change will not keep up, delaying the impact on DDS. Moreover, although HICs are likely to be the first to feel the impacts of AI in DDS due to stronger incentives to automate labor and lower barriers to technological adoption, developing countries relying heavily on outsourcing for job creation may not be far behind, as HIC-based customers impose new requirements on cost and speed, which may require the adoption of AI solutions at scale. Figure CS3.2 summarizes the impacts of AI to date on productivity, employment, outsourcing opportunities, and wages in DDS, and the following section discusses a theoretical framework grounded in the analysis of the available literature that can help policy makers navigate this landscape in the years ahead. This framework will require frequent updating to capture emerging patterns, directions, and magnitudes as the understanding of the use of AI at work and its impact develops. FIGURE CS3.1  GenAI use in the United States, by industry, December 2024 Administration and support services Banking, nance, and insurance Professional and business services Arts, entertainment, and recreation Management of companies Information services 0 5 10 15 20 25 30 35 40 45 50 55 60 65 Percent Source: Original figure for this publication based on data from Hartley et al. 2024. Note: GenAI = generative AI. B ackground and L iterature R eview   11 FIGURE CS3.2  AI Impacts on productivity, employment, outsourcing opportunities, and wages in DDS Potential e ects on productivity, employment, DDS types Relevant AI capabilities Impact on tasks and skills outsourcing opportunities, and wages Productivity Software E ect: Likely increasing for most DDS. development • Task substitution Distribution: Tend to improve productivity more for entry-level positions and and (Al handles part or low performers on both simple and more advanced tasks and for high IT services all of a task) performers on tasks requiring signi cant domain knowledge or contextual • Task awareness. recomposition (task are split or Employment • Automation of merged; Al with E ect: Uncertain; likely to involve both job displacement and creation. Strong routine human-in-the-loop demand, client risk-averseness, and business process inertia may mitigate processes Specialized control) labor disruption e ects of AI in the short term. • Augmentation services • Volume increase Distribution: Decreased hiring of entry-level roles, as AI improves or copilot for (potential to do productivity of existing workers, and lower-skilled jobs, as AI automates complex tasks more tasks when simple tasks, disproportionally a ecting youth and more economically • Quality and Al makes it less vulnerable groups. speed gains expensive or faster • Personalization • Value chain shift at scale Outsourcing opportunities (introduction of Al • 24/7 availability E ect: Uncertain; potential for shifting the type and location of outsourcing provokes a change Creative • Translation/ opportunities. Contact centers and BPM. in the mix of tasks services multilingual Distribution: Potentially decreasing for routine digital tasks that AI can done on-, near-, or support automate. Potentially increasing for tasks where workers can be o -shore) • Code complemented by AI to move up the value chain. Firms that can quickly adopt • Skill mix change generation AI to o er lower cost with maintained quality are set to bene t. (AI may shift balance of entry vs. senior roles and Wages Contact create new demand E ect: Depending on AI exposure and complementarity of jobs. centers and for coordination or Distribution: Potentially increasing for high-complementarity jobs where AI business training skill sets) frees up time for humans to do more value-added work. Potentially process management decreasing for low-complementarity jobs where AI takes over tasks and reduces the skill demand and value added of the job. Source: Original figure for this publication. Note: Capabilities are model agnostic. Outcomes will vary by context, maturity, and pace of adoption. AI = artificial intelligence; BPM = business process management; IT = information technology; DDS = digitally delivered services. Productivity AI can substantially boost productivity across segments thanks to automation and process optimization. Field experiments show that AI assistance enables workers to accomplish more in less time. For example, a study of a Fortune 500 customer service center found that a GenAI tool , with large gains for less- helped agents resolve 14 percent more service chats per hour on average​ experienced and lower-skilled workers (Brynjolfsson, Li, and Raymond 2023). AI coding assistants can also dramatically speed up programming tasks. In an experiment with professional programmers, those given access to a GenAI pair programmer (GitHub Copilot) completed a coding task in less than half the time taken by the control group, with no loss in quality​and slightly higher gains for lower-skilled programmers (Peng et al. 2023). Indeed, some software development firms say they will only hire developers who work alongside AI because productivity gains are so significant. In knowledge work, GenAI likewise enhances output: Consultants using ChatGPT in a set of business case tasks were more productive and produced higher-quality analyses, with the largest 12  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S improvements observed among those who initially performed worse on an assessment (Dell’Acqua et al. 2023; refer to box CS3.1). In writing and content creation, access to GenAI has been shown to save time and improve work quality, again disproportionately helping those with weaker baseline skills. These studies suggest that AI’s capability to democratize expert knowledge potentially reduces the skill premium for similar tasks. BOX CS3.1  AI and productivity of software engineers According to a Stanford study of 100,000 software engineers in 500 companies across the globe, artificial intelligence (AI) is driving net productivity gains of 15–20 percent in software development. Using AI tools increases new code generation by 30–40 percent while also raising rework by 20–30 percent, leading to overall productivity improvements of about 15 percent (refer to figure BCS3.1.1). Although AI significantly increases productivity by accelerating coding, increased error rates demand robust human oversight to benefit from these productivity gains. FIGURE BCS3.1.1  AI’s impact on software development productivity 30%–40% However, a lot of that new code has Al makes it easy to bugs and must be reworked generate new code 15%–25% 15%–20% The net overall software engineering productivity gains from Al are ~15% New code Rework Net productivity gains Task complexity +0%–10% +10%–15% High complexity in legacy systems is High Even in new projects, complex tasks require deeper constrained by outdated code and intricate human insight, limiting Al's assistance dependencies, so gains are modest +35%–40% +15%–20% These tasks are often repetitive and Legacy projects still bene t on simpler tasks, Low well de ned, allowing Al to automate routine although integration challenges can work very e ectively reduce the impact Green eld Brown eld Project maturity Source: Denisov-Blanch and Obstbaum 2025. Note: AI = artificial intelligence. (Continued) B ackground and L iterature R eview   13 BOX CS3.1  AI and productivity of software engineers (Continued) Productivity increases from AI use depend on task complexity and project maturity. AI delivers the highest gains of up to 35–40 percent efficiency improvements in low-complexity tasks in new (greenfield) software development projects without many legacy constraints (refer to figure BCS3.1.1). However, as task complexity rises, AI’s impact diminishes, requiring deeper human oversight, particularly in brownfield projects, where legacy systems and integration challenges constrain gains to 0–20 percent. Coding language support also affects outcomes, with AI providing 10–25 percent productivity boosts in popular languages, such as Java and Python, but showing minimal or even negative effects in niche ones. Over the next 5 years, commercial AI capabilities in software engineering may evolve from basic assistance to fully autonomous development. A strategic focus will be on systems that proactively codevelop software with minimal human intervention. Current AI tools are progressing toward higher complexity levels, expanding from basic code assistance and optimization to recent advancements in reasoning support. The next stage may focus on more advanced and automated capabilities, ranging from automated workflows, which can convert requirements into code and run tests automatically, to autonomous development, where AI systems could potentially build software with minimal human oversight. As these higher-complexity applications evolve, AI is expected to become more deeply integrated into software engineering workflows, significantly enhancing productivity and driving software development to new levels of efficiency. However, for tasks that require significant creativity and deep domain knowledge, AI seems to disproportionally benefit high performers. One experiment with an AI advice tool for entrepreneurs saw high-performing users improve profits, whereas lower-performing users experienced a slight decline, possibly because of differing usage patterns. Although many cases show AI raising the floor of productivity by automating simple or time-consuming tasks and optimizing workflows, some recent studies that focus on more advanced and ambiguous work suggest that AI can reinforce skill bias by disproportionately increasing the productivity of higher performers with deeper domain knowledge. Indeed, AI may significantly reduce the need for basic expertise while simultaneously benefiting top experts the most (Mollick 2025). Which productivity impact will be more pronounced will, in the near term, likely depend on how AI is deployed in firms rather than on the development of new AI capabilities. Employment The effect of AI on digital services jobs is uncertain and will likely involve both job displacement and job creation. All jobs are a collection of tasks requiring different levels and kinds of skills. Assuming that most DDS jobs are highly exposed to AI because of their digital nature, its net effect on employment in a given job will be influenced by the share of tasks in that job that AI complements or substitutes for, as well as changes in demand (Pizzinelli et al. 2023). Consistent with this theory, studies have suggested that, in the near term, AI can increase the risk of displacement— that is, essentially compete with humans—for digital services jobs that predominantly consist of lower-skilled routine tasks by automating these tasks and leaving the human worker with less work, potentially widening the gap between high- and low-skilled workers, as well as between senior and junior roles (Drydakis 2025). Employment in occupations highly exposed to AI, as well as in firms that use it intensively, has already begun to edge downward since the arrival of ChatGPT, driven by automation 14  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S (Eisfeldt et al. 2023). However, unlike previous automation technologies, AI’s reasoning capabilities also enable it to complete tasks that require more advanced skills, such as coding, translation, some research, and tutoring (De Simone et al. 2024). Indeed, freelance and gig jobs have been affected by AI substitution, because they often consist of unbundled tasks. Once AI automates such a task- based job, demand for human labor decreases significantly, because AI can complete the task more quickly and cheaply (Demirci, Hannane, and Zhu 2023). Nevertheless, to the extent that tasks are combined in unpredictable ways to form jobs that also require advanced analytical skills, creativity, decision-making, adaptability, and deep domain knowledge, AI will be more likely to complement humans, acting as a helpful co-worker that allows the human to spend more time using advanced skills. In this way, higher complementarity and higher-skilled jobs could entail a lower risk of labor displacement, supporting the idea of AI being a driver of skill-biased technological change (Violante 2008). However, even in cases in which AI complements labor, if it makes workers twice as productive while the demand for the digital service less than doubles, the ultimate demand for human working hours will decline. The outcomes may depend on how sensitive the demand for the service is to the changes in price driven by technological changes (Bessen 2018). Firms are adopting AI at scale, with some effects on hiring of junior and lower-skilled staff already visible. The IBM Global AI Adoption Index 2023 found that approximately 42 percent of large enterprises (those with more than 1,000 employees) are actively deploying AI solutions. Although the effects are not yet visible in official labor market statistics, corporate leaders say they will slow the hiring of junior staff (Martin 2025), including developers, and such trends can already be seen in data on job postings in some locations, such as the Philippines (refer to the “Country Cases” section). In digital services, AI-enabled digital transformation may create a few jobs in areas such as data analytics and AI application development, benefiting individuals with AI-relevant skills. Consistent with this idea, the use of GenAI tools for coding has been shown to increase the number of developers in a country (Quispe and Grijalba 2023). For high-skilled jobs that require domain expertise, AI is likely to complement humans, allowing them to focus more on analytical and creative task that directly leverage their domain expertise. However, some low- and mid-skilled jobs could be displaced. Previous technological changes suggest that job displacement through AI adoption will be transitional rather than permanent, with medium- and longer-term impacts on workers depending on policy responses and the speed of adaptation. Increased AI use could create jobs by catalyzing new firm creation in developing countries. This was the case during previous waves of technological change, such as the mobile and internet revolutions (Houngbonon, Mensah, and Traore 2022). However, because most firms leveraging AI will rely on similar foundational AI models and infrastructure, they will need to identify a different defensible competitive differentiator. Key differentiators are likely to include ownership of proprietary data, ownership of client relations and workflows that enable distribution of AI services, the ability to design products that fit into those workflows, and access to the relationships and infrastructure necessary to deploy the solutions in cost-effective ways. To enable innovators and entrepreneurs to capture this value and create jobs instead of creating companies that can be easily copied, investors and policy makers must provide targeted mentorship, training, and incentives. AI is transforming employment dynamics for freelance workers and gig professionals, allowing some to advance up the value chain, while also reducing demand for simple and text-heavy tasks. For example, AI-powered translation, transcription, and design tools allow individuals to perform tasks that previously required specialized skills, expanding the scope and speed of DDS. AI adoption also seems to affect earnings per contract for freelancers, with Upwork, a freelancing platform, B ackground and L iterature R eview   15 reporting an average 11.5 percent rise (Liu, Deng, and Monahan 2024). Nevertheless, for simple and text-heavy tasks, such as copy editing and proofreading, which GenAI can do well, demand and wages seem to be decreasing. Early studies suggest that image based GenAI could have similar effects on simple creative work (Hui and Reshef 2025). Matching platforms increasingly use AI to connect clients with workers in real time, optimizing for availability, performance, and fit—thereby increasing utilization rates and compressing turnaround times. In parallel, automated QA systems and AI-driven customer support tools are allowing solo providers and small teams to operate at standards previously reserved for larger firms, shifting the boundary of who can participate competitively in export-oriented service delivery. Outsourcing opportunities AI has the potential to alter global outsourcing patterns by changing client firms’ considerations of what to outsource and where. AI could disrupt outsourcing in two main ways. First, by automating routine digital tasks, AI could reduce the need to offshore certain jobs. If an AI system can handle basic customer inquiries or generate basic code, it may be less cost-effective to outsource these functions, leading firms to carry them out in-house. However, to enable this development, firms would need to make the upfront investment in an AI system and retain the skills needed to manage it themselves. Faced with this hurdle, a firm may choose to keep the task outsourced while requiring the service to be priced significantly lower, knowing that it is performed by AI. This could be an opportunity for outsourcing firms that have both customer service and software development capabilities. Indeed, as firms digitalize, software is likely to grow faster than services (Arora et al. 2021). Second, AI could enable human workers in offshore locations to move up the value chain, supervised or empowered by AI, shifting the nature of outsourced tasks. As researchers have shown, AI disproportionately improves the performance of lower-skilled workers on tasks such as coding and customer service, allowing them to handle more client work in less time and making offshore teams even more competitive as productivity goes up and human labor costs remain low. The impacts will depend on the pace of AI adoption, as well as geo-economic considerations. Countries with high specialization in DDS, such as IT, finance, and BPM, show a high uptake of AI tools. AI-augmented labor could help these countries stay competitive as firms choose between domestic and foreign labor and consider AI automation as a factor in their outsourcing decisions (Liu and Wang 2024). However, nearshoring, driven by both geopolitical considerations and AI impacts, may drive companies to keep AI-sensitive processes closer to home, including for data security or regulatory reasons, which could hurt higher-skill outsourcing, such as data analytics and specialized software development, delivered far from client markets. With AI adoption, routine digital work could see less labor arbitrage as automation replaces some roles, and higher-skilled work could become more globally distributed as AI empowers workers to move up the value chain. Wages The impact on wages will depend on the distribution of productivity gains from AI adoption. The evidence discussed earlier in the “Productivity” subsection suggests that AI could complement higher-skilled workers, increase wage inequality, enable lower-skilled workers to perform higher- skilled tasks, potentially diminishing pay differentiation, and commoditize lower-skilled workers’ competencies, exacerbating wage inequality. The actual impact on wages will depend on whether firms reward performance improvements equally for low-skilled or junior workers and for highly skilled workers. If this is not the case, AI could exacerbate labor market polarization. 16  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S What is already clear is that firms are putting a premium on AI expertise, with AI software developers and product managers commanding salaries of close to US$1 million per year (Nair 2023) or more, and that demand is increasing for workers able to deploy AI for data analytics, software development, and marketing (Drydakis 2025). It is possible that a similar pattern could extend to other digital services as well, with firms paying higher wages to workers with experience in using AI tools. This would increase incentives for workers to upskill themselves to work with AI—as AI co-workers—and for policy makers to ensure education systems provide such upskilling at scale. The ways in which AI adoption affects DDS will have important implications for gender, education, and income inequality. First, AI may increase inequality between low- and high-skilled workers, because, as discussed earlier, AI adoption may push the wages of the high-skilled workforce upward and the wages of low-skilled workers downward. Second, AI may increase gender inequality. In a representative survey of US workers, 38.0 percent of men and 27.8 percent of women reported using AI at work (Hartley et al. 2024). Higher-income individuals are also more likely to use AI tools, creating the risk of further income divides. Third, those with a college degree or higher have greater potential to benefit from the widespread adoption of AI than those with a high-school education or lower, who are least likely to benefit from AI adoption. Finally, when looking at income levels, potential adverse impacts seem to be evenly spread across the income distribution, whereas benefits are concentrated at the top (Hartley et al. 2024). AI in DDS: A framework for considering opportunities and risks A framework of four archetypes of DDS economies can help policy makers think strategically about the effect of AI on their digital services sector. We propose categorizing economies into four broad archetypes (refer to figure CS3.3) on the basis of their reliance on digital services exports and the skill level of their digital economy, using software developer activity as a proxy indicator:1 • Digital export leaders (top right quadrant) have highly skilled workers in the digital economy and high DDS exports and are well positioned to harness and contribute to AI advancements. These countries often outsource lower-skilled tasks to countries with cheaper labor while they focus on high-skilled, high-value tasks. • High-skilled, high-potential economies (top left quadrant) currently generate modest DDS exports despite their highly skilled workforce, presenting significant potential for capturing value by increasing DDS exports either AI based or driven by AI developments. • Mature lower-skilled exporters (bottom right quadrant) earn considerable income from DDS exports, the majority of which require less-advanced skills. These countries risk both missing out on the benefits of AI adoption and disproportionately suffering the negative effects of AI-driven labor displacement unless they diversify, upskill their workforce, and invest in complementary infrastructure and technology capabilities. • Emerging builders (bottom left quadrant) are at an early stage of DDS exports and often also at an early stage of digital economy development. This gives them a chance to identify global best 1 The use of developer activity as a proxy for higher-skilled workers is primarily meant to highlight the fact that software development requires logic and problem-solving skills as well as the ability to learn complex concepts. All else being equal, the more widespread such skills are in an economy, the easier it is to upskill the workforce in that economy. The choice of this indicator does not mean that software development is resilient to the impact of AI. Refer to boxes CS3.1 and CS3.7 for more insight on this point. B ackground and L iterature R eview   17 practices and the opportunity to develop an AI-native DDS sector with targeted and strategic investments in skilling and infrastructure. Countries at the same level of digital services exports may specialize in different sectors. For example, countries such as India that have a significant share of digital exports in noncomputerized services such as research and consulting are more likely to be complemented than substituted for by AI. Those with higher skill levels are better positioned to move into higher value-added segments, such as advanced software development and professional service outsourcing, which could offer greater long-term growth and sustainability in an AI-driven age. We also factor in the intensity of AI use, proxied by ChatGPT usage intensity (monthly traffic per internet user) and represented by the bubble size in figure CS3.3, to show the country’s familiarity with AI tools, recognizing that experience with GenAI is important for reaping the benefits of AI in DDS. Each of the four archetypes faces different opportunities for and risks to their DDS exports. The case studies of the Philippines and Uzbekistan that follow represent the mature lower-skilled exporter and emerging builder archetypes, respectively, and highlight the different challenges and opportunities the two groups of countries face in the course of accelerating AI development and adoption. The case of Viet Nam, a high-skilled, high-potential economy, is also featured later in the section on country cases. FIGURE CS3.3  Four-archetype framework based on DDS exports, digital skill level, and intensity of GenAI use for select countries, 2023 Github pushes per 1,000 population ages 18–35 1,000 EST SGP KOR FRA USA High-skilled, high-potential economies 100 BRA VNM RUS Digital export leaders KEN ROU PER EGY PHL 10 UZB MYS ETH SUR Emerging builders TGO Mature lower-skilled exporters 1.0 0.1 1.0 10 100 DDS exports (% GDP) Source: Original figure for this publication using calculations from Github, IMF, Semrush, and WTO. Note: The horizontal and vertical lines represent the median values of the x- and y-axes. Circle size indicates ChatGPT usage intensity, measured as total website visits per internet user in each country for March 2024. Countries such as Ireland that have a significant population of foreign firms that inflate their digital services exports are not included in this diagram. For a list of country codes, refer to https://www.iso.org/obp/ui/#search. DDS = digitally delivered services; GDP = gross domestic product; IMF = International Monetary Fund; WTO = World Trade Organization. Country Cases Introduction This section presents findings from the case studies of the impact of artificial intelligence (AI) and digitally delivered services (DDS) providers in the Philippines and Uzbekistan. These studies examine the challenges and opportunities of the mature lower-skilled exporter and emerging builder archetypes. After a brief introduction to the structure and characteristics of DDS in each country, this section discusses the adoption and impact of AI on DDS and the observed impacts of AI on the supply of DDS jobs. It then proceeds to qualitative insights into how AI solutions are being adopted by firms in various sub-segments of DDS and examines the barriers to adopting and developing AI solutions and AI’s impacts on firms and workers. The section also includes a spotlight on Viet Nam, a country of the high-skilled, high-potential archetype, exploring opportunities and hurdles to increased DDS export in the context of AI adoption (refer to box CS3.4). The Philippines: Mature lower-skilled exporter The Philippines is a leading global hub for DDS delivery. DDS exports rose from US$4.2 billion in 2005 to US$29.4 billion in 2023, supported by foreign direct investment (FDI)-friendly policies that also heighten reliance on foreign ownership and US demand. The sector employs about 1.82 million workers—around 3.7 percent of all jobs in the Philippines—with 1.6 million in contact centers, representing roughly 40 percent of the global customer experience management workforce. Higher-skill segments are smaller but meaningful, including about 250,000 in back-office functions in technology, banking, and financial services; about 190,000 in health care services; and nearly 180,000 in information technology (IT) services and software development. GenAI usage is advanced, with ChatGPT intensity at 1.67 monthly visits per internet user in March 2024—about 4 times the world average. This suggests a solid foundation for AI adoption, with the right support. The case introduces sector structure, job creation trends, revenue trends, and AI exposure. It then presents sector-specific findings for contact centers and business process management (BPM), specialized health care services, creative services, and software and IT services, focusing on AI adoption pathways, barriers, implications for workers and firms, and near-term challenges. 19 20  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Background DDS exporting firms employ about 1.8 million workers—about 3.7 percent of all jobs in the Philippines—with about 1.6 million in contact centers, representing roughly 40 percent of the global customer experience management workforce. The sector is the main source of foreign exchange and has helped diversify the otherwise agriculture- and remittances-driven economy. As of October 2024, the sector employs about 1.82 million workers, which equals about 3.7 percent of all jobs. About 1.6 million of these jobs are in contact centers, representing about 40 percent of the global headcount of workers in the customer experience management market.2 The majority of contact center firms are foreign owned, and successful home-grown firms tend to be acquired by foreign ones. Policies such as the Philippines Special Economic Zone Act of 1995, which provides tax incentives, and the Philippines Foreign Investments Act of 1991, which has permitted 100 percent foreign ownership since the 1990s, have encouraged foreign presence in the Philippines. Although this has driven FDI inflows and may have supported skills development and global competitiveness, it has also reduced the potential for local reinvestment because firms repatriate profits, limited domestic control because business decisions are made overseas, and made the sector more sensitive to policy and geo-economic changes because of the strong reliance on the US market. A smaller share of jobs is found in higher-skilled, higher-value-added segments such as specialized services, IT services, and creative digital services. Since the 1980s, international companies have opened their own captive centers in the Philippines. About 250,000 jobs are in business process functions such as back-office support, primarily in banking, financial services and insurance, and tech sectors. The health care services sector hosts about 190,000 jobs, with the majority focused on claims management, billing, medical coding, and patient management. IT services and software development provide about 180,000 jobs, primarily focused on application development management, infrastructure services, front- end and interface design, and enterprise software development. The Philippines has also seen some success in attracting highly advanced jobs. For example, Dyson, a global appliance brand, has a center for software and AI innovations, and Samsung has a research and development (R&D) center, both in Manila. Adoption and impact of AI in DDS GenAI adoption in the Philippines surpasses that of its regional peers, and the country is relatively exposed to AI-driven job displacement, especially in services. As of March 2024, ChatGPT usage intensity in the Philippines is significantly higher than in most of its regional peers, reaching 1.67 monthly visits per internet user, nearly 4 times the world average (0.43) (refer to figure CS3.4). This figure places the Philippines closer to global GenAI leaders such as Singapore (2.20), highlighting its strong engagement with AI tools. Moreover, more than 40 percent of employment in the Philippines consists of routine cognitive and manual tasks vulnerable to automation. Among East Asian peers, the country sits in the middle to upper range for such routine-intensive work, with the second largest share of routine workers in the service sector after Malaysia (refer to figure CS3.5, panel a). 2 Industry Overview Q4 2024, provided to the authors by the IT and Business Process Association of the Philippines on December 15, 2024. C ountry C ases   21 FIGURE CS3.4  ChatGPT visits per internet user in the Philippines and other regions, May 2025 Visits per internet user 7.00 6.25 6.00 5.00 4.00 3.00 2.00 1.41 1.51 1.05 0.84 0.91 1.00 0.54 0.61 0 PHL UMICs IDN THA World VNM MYS SGP Geographic area Source: Original figure for this publication using data from Semrush. Note: IDN = Indonesia; MYS = Malaysia; PHL = Philippines; SGP = Singapore; THA = Thailand; UMICs = upper-middle-income countries; VNM = Viet Nam. FIGURE CS3.5  AI exposure in EAP, by country a. Jobs, by nature of tasks b. AI-displacement exposure Percent Complementarity-adjusted AI exposure, latest available year Less likely to 100 be displaced by 4.6 AI or robotics 80 MYS Agriculture 60 Nonroutine manual 4.4 PHL Nonroutine cognitive PNG VNM Routine cognitive KHM IDN 40 THA Routine manual 4.2 TLS FJI MNG MMR More likely to 20 LAO be displaced by AI or robotics 0 4.0 7 8 9 10 11 12 m Ca ina Ind ia Ph esia Th s Ma d Mo sia a e oli an od Na pin lay Ch ng GDP per capita, log ail on mb t ilip Vie EAP Other EMDE AE Sources: World Bank 2024b, with calculations based on Felten, Raj, and Seamans 2021, Pizzinelli et al. 2023, and World Development Indicators (https://databank.worldbank.org/source/world-development-indicators). Note: Panel b shows the complementarity-adjusted AI exposure measure at the country level. A higher exposure value indicates a greater risk of job displacement because of AI. AE = advanced economies; AI = artificial intelligence; EAP = East Asia and Pacific; EMDE = emerging and developing economy; GDP = gross domestic product. A U-shaped relationship exists between gross domestic product (GDP) per capita and AI displacement risk (refer to figure CS3.5, panel b), putting middle-income economies such as the Philippines near the lower part of that curve. However, the Philippines lies slightly above this U-shaped trend line, implying a higher replacement risk than would be expected at its current income level (World Bank 2024b). 22  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Two-thirds of digital services exporting firms in the Philippines report actively implementing AI, but just one-quarter say they have observed a shift in jobs structure. According to a survey carried out by the IT and Business Process Association of the Philippines (refer to figure CS3.6), firms report improvements in employee productivity, operational efficiency, and service quality. 3 Just 8 percent of firms reported job losses because of AI adoption, whereas 13 percent reported job gains. About one-quarter of firms reported shifting demands for job roles and a significant need for reskilling and upskilling as a result of AI adoption, and one-third said they have not seen any shifts in the workforce structure thus far. Nevertheless, some firms are concerned that the sector is in denial about the potential impacts of AI, causing workers, firms, and policy makers to remain complacent instead of proactively upskilling and reskilling workers to prepare them for shifting labor market realities. Some firms report a fear that AI tools may get too powerful if they use them and give them too much information. They try to avoid this by not using these tools. Labor market data suggest that IT and BPO jobs declined sharply after the launch of ChatGPT and do not seem to have recovered 2 years later. After ChatGPT’s launch, demand for IT jobs initially surged but declined sharply after the first-quarter (Q1) 2023, reflecting shifts in market dynamics and the impact of GenAI-driven automation trends (refer to figure CS3.6). BPO jobs followed a similar pattern, albeit with less volatility. By 2024, the demand for both IT and BPO jobs had dropped to less than 50 percent of their levels in fourth-quarter (Q4) 2022. In contrast, demand for local jobs, represented by shop sales assistants, showed resilience, experiencing steady growth until 2023 and subsequently stabilizing at a level higher than in Q4 2022. These trends highlight the varied impacts of digital transformation across occupations, emphasizing the importance of targeted policy initiatives to address declining demand in digital services exporting sectors. FIGURE CS3.6  AI adoption impact on jobs as seen by IT and business process firms in the Philippines Reported job losses 8 Reported job gains 13 Shifted demand for job roles 24 Require signi cant reskilling and upskilling 26 No shift in workforce structure 29 0 5 10 15 20 25 30 35 Percent Source: Original figure for this publication using data from IBPAP Mid-Year Performance and Outlook Survey for 2024–25. Note: IBPAP = IT and Business Process Association of the Philippines. 3 Industry Overview Q4 2025, provided to the authors by the IT and Business Process Association of the Philippines on December 15, 2024. C ountry C ases   23 FIGURE CS3.7  Job postings in the Philippines, by occupation, 2022–24 Job postings (thousands) 175 150 125 100 75 50 25 0 22 2 2 22 23 3 3 23 24 4 4 24 02 02 02 02 02 02 20 20 20 20 20 20 ,2 ,2 ,2 ,2 ,2 ,2 1, 1, 1, 1, 1, 1, l1 y1 l1 y1 l1 y1 ry er ry er ry er ri ri ri Jul Jul Jul tob tob tob ua ua ua Ap Ap Ap Jan Jan Jan Oc Oc Oc BPO jobs IT jobs Local jobs Source: Original figure for this publication using data from LightCast. Note: All series are normalized to a value of 1 in fourth-quarter 2022—the quarter when ChatGPT launched. Occupations are classified at the 4-digit level according to the International Standard Classification of Occupations. BPO jobs include the two most in-demand occupations, enquiry clerks and contact center information clerks. IT jobs include software developers and ICT service managers. Local jobs are represented by shop sales assistants. BPO = business process outsourcing; ICT = information and communication technology; IT = information technology. Dashed line indicates the quarter when ChatGPT first launched. Sector-specific findings This section presents detailed findings from each focus sector. It highlights how contact center and BPM firms are deploying AI for training, coaching, and call summarization and notes that legacy systems slow down AI adoption. Health care services and software and IT use AI to automate coding and the patient care journey and shift toward AI apps and cloud, while creative firms test tools cautiously amid intellectual property (IP) concerns. Across segments, firms try new business models and report seeing talent shortages grow while entry-level roles reduce in number. To accelerate responsible AI adoption that preserves jobs and keeps more value creation in the Philippines, firms say they need rapid reskilling support and clearer data and AI governance. Contact centers and BPM The Philippines is the world leader in the contact center and BPM industry. Nearly three-quarters of clients are in the United States, with the remainder split evenly between Europe and Asia and the Pacific. The segment employs about 1.6 million people with annual revenues of about US$32 billion as of 2024. The most mature services are voice- and non-voice-based contact center functions, which include customer support hotlines, technical and IT support help desks, and telemarketing and sales calls. Emerging services include HR management and finance and accounting support. Key outsourcing drivers include the Philippines’ large English-speaking labor pool, cultural affinity with the United States, and significantly lower labor costs. Some leading firms in the Philippines are TTEC, Concentrix, Cognizant, Teleperformance, Alorica, Accenture, and Wipro. 24  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Use Cases • Internal. GenAI employee chatbots to access information, ML-powered training of new call center agents, and ML-powered personalized agent coaching to improve performance. • Client facing. ML and NLP automation of simple customer service tasks, such as password resets, profile updates, and account balance inquiries. Tools • Client facing. Variant (speech analysis), Google Contact Center as a Service (contact center AI platform; call summarization), RealSkill (AI-powered training), and Amplify (AI-powered coaching). Adoption Contact center firms are adopting AI to enhance employee performance, improve efficiency, and strengthen client engagement. AI-powered training tools are enabling agents to reach proficiency 20 percent faster by hosting mock calls and identifying employees’ strengths and weaknesses without manual intervention from team leads. Firms also see less attrition among new hires because the AI-based curriculum helps them learn and feel empowered (TTEC 2025). AI-driven coaching tools provide targeted feedback, helping agents improve their performance by highlighting areas for development. Speech analytics tools enhance efficiency by offering insights into call sentiment, demand times, and call types, allowing for proactive improvements in both call handling and team management. Call summarization tools automate the summarization process done at the end of each call, reducing the time spent on each call by 30–60 seconds and improving efficiency by 10 percent. Taken together, these tools enable workers to reach performance targets faster and more effectively while also enhancing customer satisfaction and supporting client retention. Firms that have both contact center and software development capabilities leverage these to accelerate the sector’s AI-driven transformation. In some firms, the software service capabilities grew from a need to develop internal solutions and serve occasional external clients, initially constituting a small share of revenues. However, these firms now have an edge because they can offer a specialized value proposition whereby the contact center teams up with the software service division to identify which specific aspects of each client’s process would yield the highest return if automated using AI and to build in-house solutions that can be licensed to other firms. To drive process improvements, firms offer incentives for contact center agents to spot opportunities for automation. Barriers Client-side barriers are the most powerful decelerators of AI adoption for contact centers. Legacy client processes can hinder the seamless integration of new AI-driven solutions, and poor data quality further complicates efforts by limiting the ability to train accurate and reliable AI models. Moreover, many clients remain reluctant to make the initial investment required for AI automation because of uncertain returns. To overcome this issue, firms sometimes use proof-of-concept to demonstrate their results to clients. C ountry C ases   25 Impact on firms Contact center firms are exploring alternative business models to diversify AI disruption risks. Some firms have started offering employer-of-record services, handling all legal and administrative employment tasks on behalf of foreign companies, allowing them to hire employees in the Philippines without a local entity. Clients include companies that previously used freelance workers but that are scaling up and need a more secure and structured model. Roles include project managers, engineers, and architects. This model helps companies mitigate AI-related revenue risk by diversifying the range of skilled roles in their talent pool. However, AI seems already to have impacted the future outlook of BPO firms, judging by stock price trends (refer to box CS3.2). Firms are also seeking to take on the role of AI consultants by guiding clients through the necessary groundwork for deploying AI—such as process documentation and data preparation— and helping them experiment with different AI solutions. By ensuring clients are prepared and supported throughout the integration, these firms aim to become essential partners in successful AI adoption. BOX CS3.2  Stock market suggests AI will profoundly impact BPM outsourcing firms in the Philippines Stock price trends for major contact center and business process management (BPM) firms indicate caution after their market capitalization collapsed around the time ChatGPT was released (late 2022) (refer to figure BCS3.2.1). However, the subsequent rebound seems to be correlated with firms’ share of revenue from contact centers. For example, Accenture derives a smaller but still meaningful share of its revenue from the BPM sector, whereas Wipro’s contact center arm represents a smaller fraction of its overall revenues. Concentrix and Teleperformance both rely heavily on customer experience outsourcing, emphasizing large-scale contact centers, suggesting that the market believes that artificial intelligence (AI) could significantly shrink returns in these sectors going forward unless the cost base is changed. Genpact, the only firm in the sample that seems to have remained somewhat steady, obtains a larger portion of revenues from digital transformation consulting and is focused on domain consulting, analytics, and AI-driven services. Indeed, the firm has used strategic acquisitions in digital consulting and analytics to move up the value chain. Firms that have large operations in multiple middle-income countries, such as Accenture and Wipro, also will be better prepared to distribute their workload between countries and adapt to more software-intensive work than smaller companies with more concentrated operations. Given the size of the workforce of these companies—Concentrix employs 100,000 people and Accenture 85,000 people in the Philippines—such shifts could create significant labor market disruptions for disfavored locations. (Continued) 26  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S BOX CS3.2  Stock market suggests AI will profoundly impact BPM outsourcing firms in the Philippines (Continued) FIGURE BCS3.2.1  Stock prices of large contact centers and BPM firms, 2020–25 Percent 200 100 0 –100 ry 3 ry 4 ry 0 ry 1 ry 2 Jul 2021 Jul 2022 Jul 2023 Jul 2024 Jul 2020 3 4 0 1 2 ril 0 ril 1 ril 2 ril 3 ril 4 25 ua 02 ua 02 ua 02 ua 02 ua 02 tob 02 tob 02 tob 02 tob 02 tob 02 Ap , 202 Ap , 202 Ap , 202 Ap , 202 Ap , 202 20 Jan r 1, 2 Jan r 1, 2 Jan r 1, 2 Jan r 1, 2 Jan r 1, 2 Oc y 1, 2 Oc 1, 2 Oc 1, 2 Oc 1, 2 Oc 1, 2 1, 1, 1, 1, 1, 1, 1 1 1 1 1 y y y y ry e e e e e ua Jan Date Accenture Concentrix Genpact Teleperformance Wipro Sources: Original figure for this publication using Bloomberg and Nasdaq data; Ando and Davies 2024. Note: Normalized as of January 1, 2020. Data before late 2020 were unavailable for Concentrix because the stock was not publicly traded until then. The availability of sufficiently qualified talent, already a constraint on growth in the BPO sector, is expected to become more pronounced as simple tasks are automated with AI. Some firms expect up to half of simple tasks and calls—such as password resets and user profile updates—to be taken over by AI in the coming 3 years. This will require agents to take on more complex calls involving problem-solving, the use of several systems, financial transactions, or complaints and will increase the skills requirements for the hundreds of thousands of new workers recruited every year. Higher skill requirements may decrease the pool of eligible new talent and decrease today’s 15 percent share of successful applicants. In turn, this issue will make it harder for firms to meet demand, which is expected to remain strong in the near term and may contribute to slowing jobs growth in the sector. Increased reliance on software over human labor improves cost competitiveness but risks moving businesses away from the Philippines to more established software development hubs. The ability of some contact center firms to offer in-house technology solutions makes it easier for firms to reach the annual cost reduction guarantees that clients often require, enhancing their competitiveness. With cost reduction being a key driver of contact center outsourcing and AI solutions offering a way to further lower costs, the adoption of AI solutions in this segment could accelerate quickly. If, however, AI substitutes for human labor and global firms develop and host their AI IP abroad, this would transfer a larger share of value abroad posing a risk not just to the C ountry C ases   27 Philippine contact center sector but to the economy as a whole. As a result, even if AI enhances the competitiveness of services delivered from the Philippines, a smaller share of outsourcing revenues would stay in the Philippines with more accruing to software hubs in other countries (refer to box CS3.3 on the possible impact of the development of accent neutralization software on Philippine BPO firms). BOX CS3.3  The double-edged sword of AI-powered accent neutralization tools in the Philippines Artificial intelligence (AI)-powered accent neutralization tools help address communication barriers and misunderstandings in the contact center outsourcing sector by ensuring that agents are more universally understood, enhancing the customer experience, and reducing misinterpretation because of regional speech variations. This work leads to improved efficiency and customer satisfaction and potentially lower error rates. As these tools become available, the impacts on the business process outsourcing (BPO) market in the Philippines—which prides itself on its neutral English accent and cultural affinity with the United States—are uncertain. On the one hand, these tools could significantly expand the Philippine BPO workforce by lowering language barriers and enabling companies to recruit beyond Metro Manila, particularly in tier 2 and 3 cities. By offering real-time translation and accent training, these solutions can reduce rejection rates of job applicants, shorten onboarding time, and unlock a wider pool of applicants, thereby addressing the high demand for skilled agents while capitalizing on the country’s cultural alignment with US clients. However, these same tools also pose risks by potentially reducing a key competitive edge of the Philippines’—neutral English accents—in voice-based services. If countries with lower labor costs or less natural cultural alignment can use AI to match an American accent, it may intensify global competition and drive voice service wages downward. BPO leaders and policy makers thus face a dilemma: Although accent neutralization can help fill staffing gaps and enhance operational efficiency, it could also erode the unique linguistic and cultural advantages that have long distinguished the Philippine BPO market. However, the Philippines’ existing BPO ecosystem benefits from a large pool of English- and Tagalog- language data, which could help strengthen the accuracy of AI-based translation and accent solutions. This advantage, coupled with significant US foreign direct investment in the BPO market, will support the country to maintain a dominant position relative to emerging players. By establishing a multilingual AI translation workflow—starting with English–Tagalog—the Philippines’ BPO sector could extend to additional languages such as Chinese and Spanish. Doing so still requires language-specific models, data, and quality assurance (especially for speech), but the same infrastructure and processes carry over, enabling quicker access to broader markets. 28  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Impact on workers AI adoption is likely to decrease the number of simple contact center jobs, leaving many workers unable to secure employment in the sector. Many contact centers organize their workforce into three levels by task complexity, with employees advancing through each level of expertise and responsibility: • Level 1 handles basic customer service functions, such as retail support, logistics, and case management. • Level 2 includes specialized roles, such as subject matter experts for different client services—for example, hotels or rental cars. • Level 3 represents higher-trust, higher-value functions such as fraud prevention, executive relations, and product specialists, where human-to-human contact can be more valuable. Because AI is already automating many level 1 tasks such as password resets and profile updates and increasing the efficiency of all calls through automated summaries, companies say they are already planning reductions in level 1 jobs of up to 10 percent in 2025. This change may increase the pressure on workers to upskill for level 2 and 3 jobs. Given the high turnover in the sector, firms have limited incentives to upskill workers. Moreover, employees who were previously narrowly able to qualify for a level 1 job may not have the basic education needed for upskilling. These two factors indicate a need for public sector intervention to support upskilling and job transition. Contact centers have little incentive to invest in substantial AI upskilling—once agents improve their capabilities, competitors readily recruit them. Therefore, government-financed training programs become critical to help the sector address AI risks and capture opportunities. (director of a domestic conglomerate) AI-driven sectoral changes may increase disparities between low- and high-skilled workers. Early studies suggest that AI tools disproportionately benefit lower-performing BPO workers (Brynjolfsson, Li, and Raymond 2023). These studies, however, do not consider potential changes in the employment structure of contact centers after AI adoption. First, if AI tools automate much of level 1 work, these tasks will no longer provide work for a majority of the human agents currently performing them. Although some workers may be able to upskill to level 2 and 3 tasks, the lowest-skilled agents may not. Second, for firms to invest in upskilling of level 1 workers, there must be enough level 2 and 3 tasks to take on. Although firms could likely absorb near-term increases in the number of level 2 and 3 workers, absorbing all level 1 workers who can be upskilled to perform level 2 and 3 tasks would require significant further increases in client demand. Thus, although AI tools might help bridge performance gaps among higher-skilled but lower-performing agents, they could widen inequalities between lower- and higher-skilled contact center and BPM workers, because some may not be able to upskill, and future demand may not meet the supply of retrained workers. Agents want more challenging work, so letting AI handle simple calls can boost their job satisfaction. (manager of a global BPM firm in the Philippines) C ountry C ases   29 Specialized services: Health care The health care services sector in the Philippines primarily serves hospitals, clinics, and insurance companies in the United States. The segment employs about 190,000 people and generates about US$4.2 billion in revenues. Mature services include claims management, coding, billing and collections, and clinical services, including patient care coordination, building on the Philippines’ strong nursing workforce. Key outsourcing drivers include reducing administrative burdens on US-based medical staff, cutting costs, and leveraging experts for complex, high-volume tasks. Firms include Cardinal Health, Carelon, Tenet, and WorldSource. Use Cases • Internal. Computer vision, NLP and ML to translate patient logs into medical codes. • Client facing. ML to personalize and optimize treatment pathways. GenAI chatbots to respond to patient queries. Tools • Coding. MediCodio (NLP coding tool), Fathom AI, and CodaMetrix. • Care coordination. AIDoc. Adoption Firms are training AI to automate medical coding, giving incumbents with large amounts of data an advantage. Medical coding involves translating medical records written by doctors into standardized codes denoting different medical procedures for insurance claims, which are turned into bills and sent to patients and insurance companies. Because of its repetitive and standardized nature, medical coding is exposed to AI disruption. Medical coding firms are developing custom AI solutions to automate the translation of patient records into medical codes. In the early stages, AI tools suggest codes that are then confirmed by human coders while more mature operations become increasingly automated. Firms with extensive historical data, such as those from health care partnerships or long industry experience, can more rapidly develop accurate AI systems, potentially raising barriers to entry for the segment. Firms are adopting AI to make the patient experience more seamless and less costly. Patient care coordination includes managing appointments, coordinating referrals between specialists, tracking medication adherence, communicating with patients regarding treatment plans, and ensuring timely follow-up care. In the health care outsourcing sector, this job is often done by local nurses who have obtained nursing licenses in the client country. Registered nurses (RNs) trained in the Philippines constitute 33 percent of all foreign-born US RNs. Firms are using NLP and ML to streamline communication, optimize scheduling, and integrate data from multiple sources to ensure seamless care transitions. ML-based predictive analytics also help identify at-risk patients, support proactive intervention by clinicians, and provide decision support based on patient history and medical guidelines, improving efficiency and personalized treatment. 30  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Barriers Data quality and accuracy of AI outputs remain key bottlenecks to AI adoption for medical coding. Although AI can handle most emergency department records, which are generally simpler, involving just one doctor providing care on one date, medical coding still requires significant quality assurance (QA) and often fails when data are inconsistent or care is provided by multiple doctors or over multiple days, as with inpatient and outpatient care. Data availability and quality, which largely rely on the sophistication of the electronic medical record system at the client site, are also a key bottleneck to training and adopting AI. Moreover, AI seems to be making fewer but more financially costly errors than human coders. For example, in one company AI missed coding certain procedures as critical care, reducing the reimbursement from insurance—a mistake that would likely not be made by a human coder. These issues undermine trust in AI solutions, making firms less willing to try them out and adopt them at scale. Impact on firms AI is improving quality and efficiency across health care services but may gradually shift the value creation abroad. Firms providing patient care coordination are leveraging AI-driven value stream mapping to enable more accurate predictions of cost savings from outsourcing and improved quality of service as measured by patient case closure rates, allowing firms to set clear expectations with clients for operational outcomes. For medical coding firms, AI is reducing costs and improving both efficiency and accuracy for simpler tasks, improving the competitiveness of their offering. Although firms highlight knowledge of the European and US health care systems as a key differentiator for the Philippines, which creates a sustainable barrier to entry, AI has the potential to drive disruption in Philippine operations of foreign-owned health care services firms. This issue occurs because their IT departments are often located in a different country, such as India. Hence, as AI takes over tasks, it could not only disrupt jobs in the Philippines but also shift the value creation abroad with potential negative effects on sector revenue. Moreover, the drive to automate and identify cost efficiencies in administrative and revenue cycle functions of the health care system is strong. Such functions account for up to 25 percent of US health care spending, making even marginal efficiency improvements result in large improvements in firms’ bottom line (Turner, Miller, and Lowry 2023). Hence, as AI tools become more accurate, firms have strong incentives to adopt them. Impact on workers In medical coding, AI is expected to lead to staff reductions and potential disruptions to industry recruitment practices. Although AI automation of medical coding initially requires significant QA, these needs decline progressively, reaching 5 percent of tasks after 6 months, according to interviewed firms. As a result, organizations anticipate workforce reductions in both coding and quality control roles, and some experienced coders will transition to specialized QA or subject- matter-expert positions. Some firms say they will need to reduce their headcount by 15 percent in 2025 because of AI automation unless they see an increase in new accounts that would require high levels of QA. Experienced manual coders are least exposed to disruption because they can more easily be reskilled C ountry C ases   31 to become quality control specialists for AI-driven processes, leveraging their workflow knowledge and foundational skills for verification tasks. In addition, they can be deployed to take over if AI systems encounter issues, ensuring continuity and accuracy in the coding process. The decrease in hiring and training of fresh graduates by some medical coding firms as a result of AI adoption may lead to a reduced pipeline of trained staff, who are typically absorbed by larger, more well-resourced firms. This shift could disrupt the sector’s structure, creating a gap in skilled workforce availability and potentially increasing competition for talent among larger firms, benefiting experienced coders. AI automation of patient care coordination could reduce the demand for human workers for simple tasks while creating opportunities for upskilling to more specialized roles. AI adoption could reduce demand for human coordinators by automating scheduling, referrals, and patient communication while also reducing the need for human workers to plan care, because AI enables predictive analytics. This shift may reduce demand for workers performing routine tasks while creating additional opportunities for clinically skilled staff who can be upskilled to work alongside AI assistants. Creative services: Animation and game development The animation services sector in the Philippines primarily serves Asian and US studios. The industry has been around since the 1980s and employs about 20,000 people, generating about US$180 million in revenue per year. Mature services include production and post-production services, primarily animation, that is, making predefined characters move and act in predefined environments, which is a labor-intensive endeavor. However, studios are expanding to character and background design, creating their own original content and IP. Most studios remain locally owned, such as PlayLab and TopDraw. Use Cases • Internal. GenAI chatbots for research. GenAI image tools to create first drafts of artwork for games. Tools • ChatGPT; Dall-E; and Adobe Animate, Autodesk Maya, and Blender, which are commonly used in these studios, incorporate computer vision and ML to facilitate tasks such as lip- syncing (matching sound to image), computer-generated 2D animation, and integration of multimedia elements into projects. Adoption Rising demand for more complex products is a driver for AI adoption, but animation and game development firms have yet to incorporate AI into production flows. Firms say they are open to AI and understand the potential benefits it can bring in terms of increased productivity and cost savings; however, thus far, they use GenAI tools only for research and not in their core business activities, such as character design, background design, and animation. Nevertheless, many commonly used animation software-as-a-service offerings, such as Adobe Animate, have begun to incorporate narrow AI and ML to, for example, match sound to images. 32  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S However, the online streaming industry’s insatiable appetite for new content is driving a shift in the animation industry, requiring firms to deliver completed productions. This is difficult for small firms because of large upfront costs and the broad skillset required. AI could play a significant role in decreasing the cost of creating these products and equipping workers with additional capabilities, potentially allowing smaller firms a better chance to compete. As a sign of what is to come, animation programs and annual industry competitions are opening up to the use of AI tools, as long as animators explain what and how much of their work was AI assisted. Barriers Key barriers to AI adoption include commitment to human creativity, workers’ fear of commoditization of their work, and uncertainty about AI-related copyright. Staff in animation studios recognize that AI can be useful in supporting activities such as storyboarding. However, despite managers encouraging adoption, they have not taken up AI tools for creative tasks driven by a strong commitment to unassisted human creativity. Some game studios have started using GenAI and prompting to create basic artwork, which is then edited by a human artist to correct issues and cultural misalignments. Uncertainty about AI and copyright has also been slowing down adoption. However, the US Copyright Office (2025) recently affirmed that combined AI and human work can be copyrighted if a human is adding, changing, or selecting elements—potentially setting the direction for other countries to follow, which should help address such concerns. Impact on firms Creative firms are seeking to move up the value chain toward IP development as AI lowers entry barriers and reshapes copyright norms. Although AI tools are still not able to independently produce culturally aligned original artwork of sufficient quality, they may lower barriers to entry for new firms or individuals without formal training in the creative industries to do less complex high-volume work such as basic-level generation for games, simple text-to-animation work, and simulating crowds for animated movies. This could increase competition in these lower-value- added parts of the sector, pushing firms toward more complex and higher-value-added work in preproduction, such as character design and storyboarding, to create their own IP. Indeed, managers fear that the use of AI could affect business models, because clients may not be willing to pay as much for AI-generated animation. Impact on workers Labor disruption is not yet on the horizon, but AI is driving the need for mindset shifts and skills adjustments. Creative services firms do not yet seem to foresee AI-driven reductions in their workforce because they believe increased demand driven by streaming services and gaming studios in the European Union and the United States may outweigh the productivity impacts of the technology. The skills requirements in the Philippine sector do not appear to be changing yet, but some studios are recruiting talent who are more open to AI use and adaptable to start using it. The Philippine Skills Framework for Digital Animation and Game Development, a promising collaborative initiative between the Department of Trade and Industry and various training and industry stakeholders, provides a shared understanding of industry workforce needs and supports continuous learning. It mentions AI as a tool for planning, executing, and optimizing game development tasks for both C ountry C ases   33 junior and senior developers but does not make it a central part of the skills requirement for employment in the sector. Software development and IT services The software development and IT services sector in the Philippines primarily serves EU and US clients. The segment employs about 180,000 people and generated about US$6.1 billion in 2024. Mature services include application development, testing, and maintenance, and emerging services include data center and cloud services, as well as network and security services. Cyber security and AI services are emerging, driven by shifting client demand. Key firms include IBM, DXC Technology, and Accenture. Use Cases • Internal. HR and IT help desk GenAI chatbots; ML-powered analytics to forecast computer maintenance and optimize computer upgrading timelines. • Client-facing operations. GenAI copilots supporting the writing, testing, and building of software in integrated development environments (IDEs). • Client products. Building GenAI chatbots for clients; implementing ML-powered data analytics; using GenAI to automate storage, organization, and access to information to find and use firm knowledge more efficiently (knowledge management); deploying GenAI accelerators, a service to help businesses integrate GenAI into their processes; using GenAI to rewrite legacy code bases to newer languages. Tools • Internal. ChatGPT. • Client-facing operations. Microsoft Copilot as part of IDEs. • Client products. Building client solutions on Amazon Bedrock and Microsoft AI platforms. Adoption IT services and software development firms in the Philippines are integrating AI solutions in three ways. First, they use AI in their internal operations, such as HR and IT, to respond to employee queries and identify when laptops need to be exchanged. This has helped save on laptop and staff costs. Second, they use AI in their client-facing operations by integrating AI copilots into development environments. This has helped enhance productivity, which is important for managed IT services contracts, where companies compete on costs and clients seek price reductions of 5 percent per year. Third, they deploy AI solutions to their clients to enhance the efficiency and quality of their operations and services. Although clients are showing interest in GenAI in areas such as knowledge management, marketing, and synthetic data, most are still in the early proof-of- concept phase. One interesting use case is translating clients’ code bases written in legacy languages such as COBOL into modern languages such as Java and generating detailed documentation that allows developers to understand the underlying business logic and rewrite it in their preferred language. This significantly speeds up this gargantuan and business-critical task, also improving outcomes for clients. 34  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Barriers Client demand for AI solutions is growing, but firms say data quality in client systems and risk aversion at global headquarters limit the pace of AI adoption. On the client side, poor data management and governance practices, as well as low data quality, are key constraints on immediate, large-scale AI implementation. Internally, some international firms have a global AI council that vets all AI client projects through red teaming, model verifications, and deployment modalities before a project is delivered. This centralized approach is perceived by some to create a bottleneck to delivery, potentially hampering efforts to capture market share and undermining job growth in the Philippines. Access to talent remains the key barrier for firms to capture demand for emerging technology services, and the government is perceived as too slow to respond. The upskilling needed to train recent graduates for client-facing AI-related work is significant, indicating a significant skills gap between industry demands and education system qualifications that sometimes forces firms to decline contracts because of the lack of staff. Previously, software firms partnered with the Technical Education and Skills Development Authority for training but were discouraged by perceived excessive red tape. Today, several firms have memoranda of understanding with universities and work with them to update curricula, incorporating emerging areas such as data analytics and AI. These collaborations involve subject matter experts, tailored electives, and internship opportunities to better align academic offerings with industry needs. This signals a shift by firms to go directly to the source of talent instead of involving the government, which they perceive as being too slow in driving the changes they need to stay competitive. Firms say they are missing a cohesive, large-scale strategy from the government to develop AI skills that can support on-the-job training as well as the establishment of advanced degree offerings in software and AI-related topics within the country. The recently published Philippine Skills Framework for Analytics and AI developed by the Department of Information and Communications Technology and the Analytics and AI Association of the Philippines provides the basis for such programs by laying out career progressions and skills requirements. Impact on firms AI adoption is driving increased revenues in AI application development and cloud services and may also decrease operational costs. AI may help improve margins for firms by automating processes and decreasing costs in internal operations, such as HR and IT support. Demand for AI-based software development projects such as GenAI chatbots, AI marketing automation, and specialized AI tools for other sectors, including contact centers and business process management, is increasing, benefiting software development firms. As AI automates simpler outsourcing tasks, revenues shift toward firms that can build AI applications. Firms also report that revenue is shifting from data center management toward application development and cloud-based services, driven by AI deployment and use. Some larger firms are establishing an asset-based consulting model, whereby a software solution is made marketable to be deployed to many different clients to hedge against AI’s impact on lower-skilled software development and establish recurring revenue streams. C ountry C ases   35 AI-driven displacement in the [business process operations] sector poses risks for overall employment in the country but also creates an opportunity for IT firms to upskill workers with finance, accounting, and HR expertise to meet rising demand for IT specialists with sector-expertise. (global IT firm manager in the Philippines) Impact on workers AI is driving increased demand for specialized software and IT roles, but talent shortages are constraining growth. Software companies are seeing a growing need for full-stack developers and software development and IT operations professionals who can integrate AI functionalities into their work. These changes are driving demand for upskilling and reskilling. For example, firms are reskilling engineers in infrastructure management, where demand is declining, and in cloud-based solutions, where clients are investing to access specialized AI computing. Data engineers are also being reskilled to support large language models, including data preparation, storage, and data science, and firms are supporting staff to earn certifications in Microsoft AI, Amazon Bedrock, and other platforms. As GenAI copilots for software development improve developer productivity by taking over basic tasks, lower-end jobs in the sector may be commoditized, making it harder for junior or low-skilled developers to secure jobs. This may benefit skilled software developers who can work alongside GenAI tools, which will make them more productive and allow them to dedicate time and energy to system design and code optimization. Given talent shortages, firms may have strong incentives to invest in upskilling developers to take on more advanced tasks and find innovative ways for developers to be AI co-workers to enhance their output. Indeed, firms emphasize that on-the-job training is vital for IT sector jobs, enabling workers to continuously upskill and remain competitive. However, although several universities produce capable graduates, the lack of postgraduate-level programs often pushes students to seek specialization abroad after completing their bachelor’s degree and gaining some work experience in the Philippines. Firms believe that by introducing these advanced programs domestically, the country could retain talent, foster deeper expertise, and help reduce brain drain. BOX CS3.4  Spotlight on Viet Nam: A future hub for AI-driven DDS? Country Archetype: High Skill, High Potential Viet Nam has demonstrated its capacity for rapid industrial transformation, most notably through its remarkable growth in high-tech manufacturing exports, which grew from less than 10 percent to more than 46 percent of total manufacturing exports between 2008 and 2024. This achievement was driven primarily by three key factors: strategic attraction of foreign direct investment (FDI), effective government support, and sustained investment in workforce skills. Today, Viet Nam has an opportunity to emulate this move up the value chain toward artificial intelligence (AI)-driven services in digitally delivered services (DDS). Viet Nam could position itself as a leading hub for AI-driven DDS by capitalizing on three key strengths—which are similar to those that supported its move into advanced manufacturing—and overcoming three hurdles. (Continued) 36  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S BOX CS3.4  Spotlight on Viet Nam: A future hub for AI-driven DDS? (Continued) Three Key Strengths Three key strengths to capitalize on are as follows: 1. Strong foundations in digital services exports. Viet Nam’s DDS exports grew dramatically—more than eightfold between 2012 and 2023—surpassing US$4.8 billion, with high-skilled computer-related services accounting for more than 30 percent (WTO 2025). This rapid growth outpaces that of regional peers, including Malaysia, the Philippines, and Thailand, highlighting Viet Nam’s momentum and opportunity to capitalize on a high-skilled workforce to shift toward more sophisticated AI-enhanced services exports. 2. Robust human capital with high educational standards. Viet Nam possesses a substantial and capable talent pool, including more than 400,000 information technology professionals and an annual addition of approximately 50,000 graduates in information and communication technology–related fields. This makes Viet Nam one of the largest suppliers of software engineers in the world (Báo Thanh niên 2024). Nearly 30 higher education institutions conduct some form of AI research or training, with several offering master’s programs in AI (EduRank 2025). The country’s educational system consistently achieves high rankings in the OECD’s Programme for International Student Assessment assessments, significantly above regional peers, a sign that students are being equipped with core competencies in math, reading, and science, as well as with proactive learning skills essential for adopting advanced technologies such as AI (OECD 2022). 3. Emerging local AI innovations, high-profile FDI in AI, and government commitment. Viet Nam is steadily building a vibrant AI innovation ecosystem, illustrated by successful initiatives such as the following: • Chong Lua Dao, an AI-powered platform that integrates with popular web browsers to protect users from online scams by identifying fraudulent websites. • FPT Corporation’s AI center, a US$173 million project in Binh Dinh province, focusing on research, software production, digital transformation (DT) assistance, and cybersecurity solutions. • VinAI’s phoGPT, a pioneering Vietnamese-language large language model, enhancing language- specific digital services and customer interactions. Notable FDI in AI include NVIDIA’s 2024 acquisition of VinBrain, a health care AI startup, and its partnership with Viet Nam’s FPT Corporation to build a US$200 million AI factory in Hanoi dedicated to research, software production, and DT. Moreover, NVIDIA’s December 2024 announcement of a state-of-the-art AI research and development (R&D) center in Viet Nam further underscores the potential international firms see in the country’s human capital for AI. However, to date, Indonesia, Malaysia, and Thailand have been preferred locations for AI-related FDI, attracting more than US$30 billion in planned investments and indicating significant upside potential for Viet Nam. Strong government commitment supports this innovation and FDI, as demonstrated by the National AI Strategy (2021–30), which aims to position Viet Nam as a leading AI innovation hub in the Association of Southeast Asian Nations by 2030 (Hoa 2025). The June 2024 decision outlining nine principles for responsible AI R&D also provides a foundation for responsible AI deployment in the country. In 2025, Viet Nam is expected to implement comprehensive AI regulations focused on flexible governance, infrastructure improvements, workforce training, and stronger intellectual property protection. Three Hurdles Three hurdles to overcome are the following: 1. Limited availability and retention of advanced AI talent. Although Viet Nam has abundant junior tech professionals, it currently experiences a shortage of experienced AI specialists. Retention of advanced (Continued) C ountry C ases   37 BOX CS3.4  Spotlight on Viet Nam: A future hub for AI-driven DDS? (Continued) AI talent remains challenging because of attractive opportunities abroad, driven by higher salaries and more favorable personal income tax regimes elsewhere. A draft law on the digital technology industry introduces a 5-year tax exemption for experts working on eligible projects, which, if implemented well, could help (Hoa 2025). 2. Restricted data access and tightening regulation. Data availability and sharing remain constrained, with critical data sets often fragmented or locked within government ministries and corporations. Moreover, the government introduced a decree on the protection of personal data in July 2023 that tightened rules on cross-border data flows. Companies intending to complete cross-border transfers of data now have to seek approval from the Ministry of Public Security on an ad hoc or case-by-case basis (OECD  2024). A  balanced regulatory framework that ensures data security without limiting innovation could help alleviate this challenge. 3. Underdeveloped computing infrastructure. Viet Nam currently faces infrastructure constraints, with just 45  megawatts of domestic data center capacity primarily supplied by local telecom firms such as VNPT, Viettel IDC, FPT Telecom, and CMC Telecom (“Vietnam’s Data Centre” 2023) and no major hyperscale cloud providers operating within the country, save for Amazon Web Services edge locations in Hanoi and Ho Chi Minh City (Sufianti 2024). Although some international firms have announced investments in data centers that could double the current capacity, these have yet to come online. A key challenge is limited domestic demand, a primary driver of investments in data centers. For Viet Nam, expanding computing infrastructure to facilitate advanced AI workloads is critical to attracting further investment and scaling up digital services capabilities, particularly given prevailing data localization laws. The telecommunications law that took effect in July 2024 opens the digital industry to 100 percent foreign investment, which could support this goal (Dezan Shira and Associates 2025). Nevertheless, to support such expansion, the government will need to expand the provision of reliable and clean energy (Sufianti 2024). Addressing these hurdles strategically could help Viet Nam emulate its successful transition to high-tech manufacturing by moving to high-value AI-driven DDS. Uzbekistan: Emerging builder Uzbekistan is emerging as a hub for DDS delivery. Today, DDS firms employ nearly 40,000, up from about 8,000 in 2021, and the information and communication sector counts nearly 100,000 self-employed workers. DDS export revenues reached about US$900 million in 2023, up from just US$22 million in 2005. While most jobs—about 25,000—are in IT services and software, contact centers have grown rapidly from a small base and today employ about 8,000. Creative services and specialized health care contribute more modestly to both job creation and export revenues but are expanding. Overall, the exporter base is young but growing, supported by FDI incentives that are starting to attract multinational firms to Uzbekistan. AI adoption in the Uzbek DDS sector is still in the early stages, in part due to a delayed release of popular GenAI tools in the country and in part due to policy and talent barriers. The case introduces sector structure, job creation trends, revenue trends, and AI exposure. It then presents sector-specific findings for contact centers and BPM, specialized services in the logistics sector, creative services, and software and IT services, focusing on AI adoption pathways, barriers, implications for workers and firms, and near-term challenges. 38  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Background Uzbekistan’s DDS sector has been growing rapidly, in both jobs and revenues, over the past 3 years. From Q1 2021 to third-quarter (Q3) 2024, total formal employment in DDS, based on data from IT Park Uzbekistan, increased nearly fivefold, from 8,000 to nearly 40,000 employees, and the number of self-employed information and communication technology workers registered with IT Park Uzbekistan increased 10-fold from 10,300 in 2020 to 100,000 in 2024 (refer to figure CS3.8, panel a). FIGURE CS3.8  Employees and revenues of companies registered with IT Park Uzbekistan, 2017–24 a. Jobs in ICT sector in Uzbekistan b. IT Park companies’ employees by sector Number of jobs Number of employees 200,000 100 30,000 180,000 25,000 160,000 53.3 20,000 140,000 39.9 120,000 15,000 100,000 18.7 10,000 80,000 26.1 38.0 2.5 3.0 4.6 10.3 18.0 60,000 61.8 9.9 5,000 59.7 6.0 60.7 59.3 61.7 57.6 52.7 53.0 40,000 0 2017 2018 2019 2020 2021 2022 2023 2024 ril 21 Jan er 1, 21 Jan er 1, 22 Jan er 1, 23 4 ry 1 ril 22 ry 2 ril 23 ry 3 ril 24 Oc 1, 1 Oc 1, 2 Oc 1, 3 y1 4 02 ua 202 ua 202 ua 202 Jul , 202 Jul , 202 Jul , 202 Jul , 202 Ap , 20 tob 20 Ap , 20 tob 20 tob 20 Ap , 20 Ap , 20 ,2 1 1 1 1 1 1 1 1 Self-employed in ICT ry y y y ua Jan IT Park residents Other enterprises of the ICT sector BPO Game development and creative IT education IT service c. IT Park companies’ revenue by sector Revenue (UZS, bilions) 5,000 4,000 3,000 2,000 1,000 0 ril 21 Jan er 1, 21 ril 22 Jan er 1, 22 ry 2 ril 23 Jan er 1, 23 ril 24 4 ry 1 ry 3 Oc ly 1, 21 Oc 1, 2 Oc 1, 3 y1 4 ua 202 02 ua 202 ua 202 Jul , 202 Jul , 202 Jul , 202 Ap , 20 tob 20 Ap , 20 tob 20 Ap , 20 tob 20 Ap , 20 Ju , 20 ,2 1 1 1 1 1 1 1 1 ry y y ua Jan BPO Game development and creative IT education IT service Source: IT Park Uzbekistan statistics. Note: In panel a, “other enterprises of the ICT sector” includes telecom and manufacturing jobs. BPO = business process outsourcing; ICT = information and communication technology; IT = information technology. C ountry C ases   39 Total revenue grew even faster, expanding 10-fold from UZS 500 billion in Q1 2021 to nearly UBZ 5,000 billion (US$380 million) in Q3 2024 (refer to figure CS3.8, panel c). DDS exports, which include a broad range of DDS such as financial services, have also grown rapidly from US$155 million in 2010 to over US$1 billion in 2024, according to UNCTAD 2023 data. With 25,000 employees in Q3 2024, the IT services and software development sector leads in both the number of employees and revenues, exceeding combined values of the contact center and BPM, IT education, and creative sectors (refer to figure CS3.8, panel b). The contact center and BPM sector, now the second-largest employer in DDS with 8,000 employees, has grown the fastest—expanding from fewer than 500 employees just 3 years ago. Despite rapid growth, Uzbekistan’s DDS exporters remain in the early stages. Most DDS exporters are small, domestically owned companies primarily serving smaller overseas clients or  acting as subcontractors for larger international firms. For example, at the time of field interviews, a company located in Urgench, a city in western Uzbekistan, was developing a chatbot for a large European electronics company but provided the service as a subcontractor to another firm holding a direct contract with the multinational firm. This arrangement reflects the sector’s nascent stage and limited direct engagement with major global clients. However, international firms are beginning to enter the market. The government is offering incentives for foreign DDS firms to establish themselves in Uzbekistan. According to IT Park Uzbekistan, the number of foreign-owned firms among DDS exporters grew from 7 in Q1 2021 to 316 in Q3 2024, accounting for 33 percent of the exporters in the IT Park Uzbekistan data. Notably, EPAM, a global software company, and several contact center and BPM firms, including Nearsol, Tafaseel, and UpStream, have either established or announced plans to set up operations in Uzbekistan, signaling growing interest in the country’s DDS potential. Adoption and impact of AI in DDS exports Although data on AI adoption by Uzbek firms is scarce, anecdotal evidence suggests that it is at an early stage. Interviews with managers and employees of software firms working on AI implementation in Uzbekistan suggest that the majority of firms are currently focused on digitalization, data collection, and preparation for AI adoption by deploying enterprise resource planning systems. A small share of firms—estimated to be a few percent by the manager of an AI and data science firm—are already at a stage when their data are good enough to adopt business intelligence and data analytics to generate data-driven insights, and an even smaller number have established the capabilities needed to progress to AI implementation. This suggests that most firms will need to invest in their data foundations and practices in the short term to get ready for AI adoption and compensate for GenAI’s delayed introduction in the country. Because of a staggered launch schedule, GenAI tools such as ChatGPT did not become available in Uzbekistan until November 2023, about a year after its global launch. In March 2024, there were just 0.25 visits to ChatGPT per internet user in Uzbekistan, significantly fewer than among regional comparators (0.99) and countries with the same income level (0.45), reaching only about one-quarter of the world average and 10 percent of the US level (refer to figure CS3.9). Although savvy digital workers purchased sim cards from neighboring Kazakhstan, where ChatGPT became available in late 2022, the delay in the broad adoption of off-the-shelf AI tools put Uzbek firms at a disadvantage. 40  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Labor market data reveal a shifting landscape of DDS jobs in Uzbekistan after the introduction of GenAI tools. Although local jobs experienced a steady decline, IT jobs saw an even sharper drop in new positions after Q4 2023, when ChatGPT officially became available in Uzbekistan (refer to figure CS3.10). Meanwhile, BPO jobs surged briefly in early 2024 before declining, yet their overall number remained small, with just 40 postings in Q4 2023 compared with 1,200 for IT jobs. FIGURE CS3.9  ChatGPT visits per internet user in Uzbekistan and other regions, May 2025 Visits per internet user 3.00 2.47 2.50 2.00 1.50 0.99 0.91 1.00 0.65 0.45 0.50 0.25 0.25 0 UZB KGZ KAZ LMICs ECA World USA Geographic area Source: Liu et al. (2025), based on website traffic from Semrush and internet users from ITU. Note: ECA includes developing countries. ECA = Europe and Central Asia; KAZ = Kazakhstan; KGZ = Kyrgyzstan; LMICs = lower-middle- income countries; USA = United States; UZB = Uzbekistan. FIGURE CS3.10  Quarterly job postings by occupation in Uzbekistan, 2023–24 Number of job postings (normalized to Q4 2023 = 100) 350 300 250 200 150 100 50 0 October 1, 2023 January 1, 2024 April 1, 2024 July 1, 2024 October 1, 2024 BPO jobs IT jobs Local jobs Source: Authors’ calculations using data from LightCast. Note: Occupations are classified at the 4-digit level according to the International Standard Classification of Occupations. BPO jobs include the two most in-demand occupations: enquiry clerks and contact center information clerks. IT jobs include software developers and ICT service managers. Local jobs are represented by shop sales assistants. ChatGPT became publicly available in Uzbekistan starting from Q4 2023. BPO = business process outsourcing; ICT = information and communication technology; IT = information technology; Q4 = fourth quarter. C ountry C ases   41 Sector-specific findings This section presents detailed findings from each focus sector. It highlights how contact centers using AI transcription, quality assurance, and reputation management tools face Uzbek-language gaps and client trust hurdles; how logistics firms apply AI to dispatch and client communication but run into English language barriers, sector-specific tooling hurdles, and privacy constraints; how creative studios accelerate concept development with image tools yet wrestle with cultural fit and high compute costs; and how small IT firms lean on free copilots while larger firms respond to client privacy expectations with policies, training, and internal knowledge tools. Across segments, companies are beginning to build proprietary tools, local language resources, and training pipelines, but scaling will depend on stronger skills, better availability of data, cloud, and data governance enablers. Contact centers and BPM Uzbekistan’s contact center and BPM sector primarily serve Russian-speaking countries. Mature services include customer support hotlines, technical support help desks, telemarketing and sales calls, and IT help desks. The sector employs about 8,000 people. Use Cases • Call center quality control. Several business process outsourcing companies use AI-powered speech-to-text tools to transcribe every call, with ChatGPT reviewing them for potential issues, such as complaints or service gaps. • Proactive reputation management. ChatGPT with a voice-over-internet-protocol tool can analyze call transcripts in real time, identifying dissatisfied customers before they leave negative reviews. Managers can then reach out proactively to resolve issues, preventing bad ratings and maintaining strong customer relationships. Tools • ChatGPT and custom-built natural-language processing tools. Adoption Uzbek contact centers implement both off-the-shelf and tailored AI solutions to improve service quality. AI-based solutions that automatically transcribe all customer service calls and flag potential issues (for example, trigger words, technical mistakes, or negative sentiment) for QA review are being adopted, improving both service quality and employee satisfaction. Because of their efficiency, they are already reducing the demand for QA managers. However, the use case in contact centers for off-the-shelf chatbots is limited because customers expect concrete, highly specialized, and quick answers that such tools cannot yet provide. Some firms recognize that global partnerships and collaboration to gather quality data for training will be important to develop the best AI tools. One firm has set up its own AI development team and partnered with peers in Europe, North America, and South America to share solutions and insights with a view to developing an AI platform that can be licensed to sector peers. 42  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Barriers Language limitations and client trust are key barriers to AI adoption for Uzbek contact centers and BPM firms. AI automation of simple queries, such as customer identification, works well in Russian and English but not yet in Uzbek because of the shortage of AI models trained in the language. However, efforts are under way to address this, as explained in box CS3.5. In addition, firm representatives have noted that, although AI tools help operators answer questions by listening to the customer and suggesting replies, the savings that these tools currently produce are not significant enough to justify their cost. Some interviewees suggested that lack of client trust in AI tools remains a key barrier for outsourcing firms to provide AI-enabled services. They drew parallels with the development of the Russian outsourcing market, which grew significantly over the past 15 years as clients gained trust in the quality of services that outsourcing firms could provide at lower costs. Some entrepreneurs believe that their clients are currently testing AI solutions for customer support internally to understand whether they can trust them and, if the tests go well, may be willing to outsource the management of AI solutions in the future, creating an opportunity for contact centers and BPM firms who are ready to take on such work. Impact on firms AI integration is increasing demands on both training and recruitment for firms and may drive shifts in the contact center business model. Some contact center firms see a need to extend agent training duration because the nature of customer inquiries has become more complex, partly because of AI-driven service changes. They are hiring more trainers, quality controllers, and information technology staff to build and manage comprehensive knowledge bases that help agents handle sophisticated questions. Previously, firms needed to specialize in providing quick answers to simple questions. With AI, they now need to provide quick answers to more sophisticated questions. AI is also helping manage volatile demand, for example, handling surges in refund requests to retail firms during the holiday season, at limited additional costs to firms. One firm estimated that AI would allow them to meet a demand equivalent of hiring 10,000 staff with only 4,000 high-quality recruits, ensuring faster growth without compromising service standards through AI-based automation and augmentation of certain tasks. If such projections are realized, the sector’s job creation potential—a key value added from a policy maker’s standpoint—will be reduced. This scenario points out a potential trade-off between firm growth and job creation for future policy. Business models for contact center firms may also change to build resilience against AI disruption, as indicated by firms developing their own AI tools for licensing. Today, AI supports human agents, improving their productivity. By 2030, human agents will support AI, improving its ability to deal with customers. (chief executive officer of contact center firm) C ountry C ases   43 Impact on workers AI may increase demand and wages for more high-skilled agents and improve their job satisfaction while reducing opportunities for junior and low-skilled staff. For contact centers serving export markets, the introduction of AI-based solutions to automate simple queries may drive increased hiring of more experienced agents at the expense of junior ones, because agents are now required to deal with more sophisticated questions. Firm representatives shared that previous waves of automation reduced the number of agents and calls per agent while at the same time increasing demands on agent skills and increasing the revenue per call, which enabled firms to pay agents higher salaries. Nevertheless, with a growing demand for contact center services, some interviewees suggest that AI will augment rather than replace workers by improving response speed and accuracy, leaving humans to provide empathy and handle complex issues. Some managers also expect AI deployment to improve employee satisfaction by reducing repetitive tasks. Although AI is currently an aid to humans, some firms believe this may change by 2030 so that humans become aids to AI, verifying its outputs and identifying issues. BOX CS3.5  Building a spoken language corpus for Uzbek LLMs as a foundation for local AI tools Background and Motivation To support the development of Uzbek language models, the Uzbekvoice initiative spent 2 years collecting and curating spoken data from diverse sources. Recognizing that existing digital materials (for example, online libraries and news sites) offered limited coverage of everyday and regional speech, the project aimed to capture a broad range of linguistic forms—from formal literary Uzbek to informal language and dialects seldom seen in written media. The government of Uzbekistan provided an initial grant for the Uzbekvoice initiative. However, recognizing the benefits of the corpus, the private sector provided both in-kind and financial support for the initiative. Approach and Methodology • Text sourcing. The team gathered open-source texts from websites, literature, and news articles. They then segmented these texts into small chunks (around 5 words each) suitable for crowd-sourced audio recordings. • Crowdsourcing audio. Volunteers were recruited through a Telegram bot (later replaced by a dedicated mobile app) to read and record these text snippets. This allowed the collection of both standard and region-specific speech. • Dialect inclusion. To capture natural dialects that do not typically appear in print, field recordings were conducted with participants wearing microphones during everyday conversations (for example, in markets and at social gatherings). This helped gather real- world, unscripted speech representing all 12 regional accents. • Quality control. More than 100,000 volunteers participated, incentivized through competitions and a government-sponsored retreat. To counter fraudulent submissions—such as silent or (Continued) 44  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S BOX CS3.5  Building a spoken language corpus for Uzbek LLMs as a foundation for local AI tools (Continued) irrelevant audio—Uzbekvoice implemented blind human reviews and later used artificial intelligence (AI) models for automated verification. Challenges • Dialect variation. The four main dialect clusters in Uzbekistan are so distinct that speakers may struggle to understand each other. The lack of standardized written forms meant that conventional read-aloud methods were not sufficient; natural recordings were essential. • Data quality and cheating. With large-scale crowdsourcing came the risk of fake or low- quality audio. Manual review was time consuming, but combining human validation with AI screening improved efficiency and accuracy. • Resource coordination. Collaborations with the national university, along with government support, were critical for recruiting and training volunteers, as well as securing funding for large-scale data collection events. Outcomes and Impact • Open-source dataset. The resulting corpus is now publicly available, driving the development of multiple Uzbek language models as well as text-to-speech and speech-to-text solutions being deployed to support digital government services. • Wider participation. Government sponsorship through the youth affairs agency fostered a sense of community ownership, enabling more extensive and inclusive data gathering. • Enhanced linguistic resources. By documenting dialects and natural speech, Uzbekvoice created a robust foundation for future Uzbek-language technologies and research. Key Lessons for Other Countries • Diverse data sources. Combine formal and informal texts, ensuring coverage of literature, news, and everyday speech to capture the full linguistic spectrum. • Balanced representation. Actively include regional dialects, even those not typically found in written form, to build more inclusive language models. • Crowd-sourcing and incentives. Engage volunteers through apps and community events, offering rewards or recognition to maintain motivation. • Quality assurance. Use both human reviewers and AI-based checks to validate audio submissions and maintain data set integrity. • Government and institutional support. Leverage public agencies and academic institutions for funding, logistical backing, and broader reach. • Open access. Open-sourcing the final corpus fosters innovation and encourages collaboration among researchers and developers, accelerating the progress of local-language large language models. Conclusion This case demonstrates that a well-planned, inclusive, and community-driven approach—supported by technology and institutional partnerships—can produce a high-quality spoken language data set that benefits not only model developers but also the broader linguistic ecosystem. C ountry C ases   45 Specialized services: Logistics The logistics outsourcing services sector in Uzbekistan serves Uzbek fleet owners in the United States as well as domestic and regional logistics companies. Although employment numbers are not available, they are likely in the region of several thousand. Mature services include fleet management, freight brokerage, and driver recruitment, and emerging services encompass advanced route optimization, real-time vehicle tracking, and digital logistics platforms. Ancillary services such as regulatory compliance and predictive maintenance are growing in response to evolving client demand. Key firms include UzLogistics, TransUz, and Global Freight Experts. Use Cases • Facilitate client conversations. Dispatchers used ChatGPT with global positioning system data to automate real-time location updates, freeing them to secure more contracts and boost reliability. • Driver recruitment assistance. ChatGPT provided alternative outreach tools (OpenPhone, Facebook, and LinkedIn) when a calling app failed, helping business process outsourcing discover new ways to engage and recruit drivers. • Lowering language barriers. Non-English-speaking trucking firms used ChatGPT to draft clearer insurance claims and police reports, speeding responses from US authorities and insurers. Tools • ChatGPT and RingCentral (AI enabled). Adoption AI adoption is in its early stages but is growing rapidly for a wide range of uses. Not all surveyed logistics firms have implemented AI tools; however, the majority said they were planning to do so during 2025. For firms already using AI, it is helping streamline driver recruitment, customer engagement, and reputation management. For example, when a local firm struggled to reach new US-based drivers because their voice-over-internet protocol calling application was not working, they used ChatGPT to discover alternative communication methods and received step-by-step guidance on how to implement each platform effectively. Another logistics firm is using an AI speech-to-text tool to analyze customer calls and identify potential client dissatisfaction, which allows managers to proactively call affected clients and promptly address any issues. Finally, firms see opportunities to use GenAI-powered training to shorten onboarding times for new staff, but such solutions are not yet being implemented. Barriers English language proficiency, lack of specialized AI tools for the logistics sector, and data privacy concerns are key barriers to AI adoption. Most employees in Uzbekistan’s logistics industry primarily speak Uzbek, making it difficult for them to write effective prompts in English and fully leverage 46  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S generative AI tools such as ChatGPT. Furthermore, a shortage of specialized solutions designed for logistics processes forces companies to rely on generic AI platforms that do not meet their specific needs, such as bidding for and coordinating pickups of freight or creating insurance claims in the case of truck accidents. Finally, strict data privacy concerns—especially regarding police reports, which need to be filed in the case of accidents, and confidential databases containing financial and personal information—further limit the use of AI, because firms remain uncertain about how securely these tools handle their data. Impact on firms AI may improve firm productivity and is unlikely to decrease barriers to entry because of the importance of interpersonal ties with client firms. Firms expect AI adoption to spur their growth, because it will enable them to capture more of the increasing demand for their services at lower cost. Close ties with fleet operators in the United States—many of whom are Uzbek—will continue to play an important role in gaining clients and establishing trust, making this sector less prone to displacement from lowering barriers to entry. Impact on workers AI may lower demand for human workers to perform routine tasks and allow managers to focus on customer relationship management. Some roles, such as back-office jobs dealing with more routine tasks, may be displaced. Meanwhile, client-facing roles in logistics firms, which require workers to be fluent in English, have domain-specific knowledge, and complete a range of tasks in unpredictable ways, may be more resilient to AI impacts, at least as long as sector- specific AI tools are not available. The turnover in the sector is high, and employees sometimes change companies for a US$100 salary increase, so firms may not have incentives to upskill workers. Intelligent call analysis systems summarize key points from recorded conversations, freeing managers from hours of manual review. With the time saved, managers can focus on resolving issues that might affect ratings on Google or industry review sites, ensuring they address negative feedback promptly and maintain a strong reputation. Managers also hope that AI can help reduce the need for them to work night shifts, which is common among logistics firms serving the United States. Creative services: Game development and animation The game development and animation sector in Uzbekistan serves primarily domestic and regional media companies, with some starting to serve the international market. While employment numbers are not available, several small studios and freelancer teams provide mature services, including 2D or 3D animation production, game art and asset creation, and post-production and clean-up. Emerging services include AI-assisted concepting and storyboards, rapid character and environment rendering, short video-to-video generation, and early motion-capture alternatives. C ountry C ases   47 Use Cases • Concepting. Instant storyboards and image prompts for characters, props, and settings. • Asset build. Rapid character and location renders, 30‑second video‑to‑video clips, and palette boosts. • Post-production. Drop‑in 3D characters, auto‑upscale and clean footage, and future AI motion capture. • Business operations. Auto‑generated pitch visuals and Telegram‑bot office administration. Tools • Midjourney—a primary tool for image generation. • DALL·E—Supplementary image generation. • Media.io—Image and video generation and enhancement. • GoEnhance—Video-to-video and image enhancer (30-second clips). • Wonder Studio (WonderDynamics)—Library of ready-made 3D characters for insertion into live-action scenes. Adoption Uzbek game development and animation firms are adopting AI tools to drive productivity. Interviewed representatives of animation firms have integrated AI into their workflows, mainly for image generation. Although AI-generated visuals meet about 50 percent of these firms’ needs, video generation remains insufficient, requiring local animators to fill the gaps. Despite these limitations, they anticipate continued advancements in AI technology and expect its role in creative processes to grow over time. Firms are keen for these tools to be trained with more Uzbek and other diverse content—or to train or fine-tune their own large language models to local needs. Barriers Cultural gaps in AI output and the high cost of computing constrain adoption efforts. Although AI speeds up scene creation, its output still lacks cultural and religious sensitivity, sometimes misplacing elements such as mosaics in mosques or missing distinctions in head coverings. Although AI has the potential to replace costly motion capture studios, cloud solutions remain too expensive for local firms, which opt for in-house computing. Although in-house systems entail high up-front costs, firms take this route to avoid the unpredictable running costs that may arise when using cloud services. Impact on firms AI tools are helping improve productivity, market reach, and visual quality of Uzbek creative studios. Surveyed firms report that character concepts that once took 2–3 days to create can now be done in a few hours, and location renders—whether modest homes or complex cityscapes—are produced roughly 3 times faster. Some firms say this shift has already led to an increase in profits because of reduced staffing costs, smaller office space requirements, and improved efficiency, with 48  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S character design time decreasing from 5 or 6 days to just 1. Faster turnaround lets firms pitch concepts earlier, secure more contracts, and even prepare impressive client presentations without mobilizing full teams. Yet the same models often miss subtle religious and cultural distinctions or Uzbekistan‑specific motifs, so artists must still retrofit culturally accurate details. Impact on workers AI enables faster project delivery and may reduce the need for junior workers. AI adoption in animation is reducing the need for beginner and mid-level workers, allowing companies to operate with smaller teams of highly skilled professionals while supplementing the project workforce with freelancers. Some game development companies are hiring individuals who lack traditional drawing skills but can use AI tools to create art, hoping to produce more artwork at a lower cost. However, delivering a large volume of consistently styled high-quality art remains difficult, even though AI is helpful for idea generation and concept development. Overall, the industry faces a long- term challenge in developing senior talent because fewer junior workers are being trained within companies. Software development and IT services Uzbekistan’s software development and IT services sector serves both domestic and international clients, increasingly targeting markets in Asia and Europe. The segment employed roughly 25,000 professionals and generated an estimated US$330 million in revenue in 2023. Mature services include web and mobile application development, testing, and maintenance, along with emerging offerings in cloud network security, with cybersecurity and AI services gaining prominence. Key players include EPAM Systems, Uzinfocom, MAAB, and Mohirdev. Use Cases • Distribute work tasks. A software development company integrated ChatGPT with its customer relationship management system via application programming interfaces, allowing it to read task lists and automatically assign them using predefined team roles. This significantly reduced the managers’ workload, improving overall team efficiency. • Deploy tools and services. A software development and information technology operations engineer turned to ChatGPT for guidance on securely deploying various software tools, reducing the risk of data loss and security breaches. This led to smoother implementations and more reliable outcomes for clients. • Learn new code libraries. A programmer needed to learn the Quartz library in C# to handle background jobs. Instead of taking a class, he used ChatGPT to understand its best practices and quickly implemented it in his project, accelerating the development process. • AI in internal operations. Software companies use AI for HR, IT, and employee management tasks (for example, responding to queries and managing equipment). • Copilots in development environments. AI is integrated into development environments to enhance productivity in client-facing operations. Tools • ChatGPT and GitHub Copilot. C ountry C ases   49 Adoption The majority of Uzbek software and IT services firms are small and tend to adopt ad hoc off-the- shelf AI solutions to improve productivity and elevate quality. Firms report using—primarily free— ChatGPT or GenAI copilot tools to write basic code and fix coding errors, which results in significant time savings. They also leverage ChatGPT for technical writing and design tasks. The fact that most surveyed companies are satisfied with free AI tools suggests they are not tapping into their full potential. However, developers acknowledge that AI-generated code still requires verification and iteration to be optimized, with humans working alongside AI. Indeed, they believe the best AI use case is auto- complete, whereby the programmer writes about 20 percent of the code and then uses AI to complete it in their style. To gain greater advantages from AI use, some firms are planning to build their own generative pretrained transformer systems using open-source AI models, underscoring the importance of open-source technologies that enable firms in emerging economies to harness the benefits of AI. The few international firms in Uzbekistan benefit from centrally developed AI tools, knowledge bases, training, and access to sophisticated clients that enable them to develop their expertise. Natural language processing and AI-powered résumé analysis helps firms identify candidates with specialized technical competencies, such as SAP OpenText, making a significant impact in a region where advanced technical skills are scarce. Some firms also have introduced mandatory GenAI training for all employees, covering both foundational and job-specific skills to ensure consistent AI literacy and promote safer data-handling practices. In addition, AI-based performance management tools are streamlining performance reviews by aggregating employee achievements into concise summaries, thereby facilitating data-driven feedback. To improve team composition, an AI-driven matching model is optimizing staffing on the basis of availability, skills, performance, and seniority, which ultimately leads to improved project outcomes and a more efficient HR allocation. Finally, AI-powered smart search tools and classifiers quickly access internal documents and generate draft responses to letters, streamlining time-consuming tasks. Barriers AI fatigue and the need for verification of AI-written code may be dampening staff enthusiasm for AI solutions. Firm representatives say staff feel overwhelmed by the AI hype that permeates all facets of life, suggesting that they may be better served by rolling out more focused efforts in areas in which they know AI will make a material difference. Managers say they need guidance on using AI in a way that balances efficiency with authentic communication, and employees still prefer direct human interaction for certain help desk and HR tasks. In addition, resistance exists to relying on AI for complex or sensitive inquiries despite the potential time savings for routine administrative functions. These cultural and operational hurdles—along with the need for training and strategic rollouts—can slow the adoption of AI tools, even when they offer clear benefits for efficiency and workload management. Data localization requirements bar us from leveraging the many production‑ready AI solutions already running on AWS, Azure, and other global clouds, slowing down adoption. (founder of a business intelligence company) 50  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Impact on firms Large firms are aware of AI-related data privacy, security, and ethical issues because of their clients’ expectations, whereas smaller firms are not exposed to such concerns. Only one of the firms interviewed, the largest, reported having internal guidelines for AI use. Managers of larger firms say their clients sometimes require AI-generated content to be watermarked or forgo the use of AI tools altogether, including copilots for development, because of data privacy and security concerns. By being able to toggle AI tools on and off, they reassure clients while still allowing the organization to harness the speed and creativity of AI when possible. Small firms did not report such issues, possibly because of a lack of awareness and client demand. Firms across the size spectrum noted that, although most clients are keen to implement AI solutions to improve efficiency and quality, they require substantial guidance to understand how such solutions can be implemented. Firms with sufficient resources use technical presale teams to work with clients to identify specific AI opportunities and then develop them into projects. AI may benefit small firms working on simpler projects in the short term, but it could prevent them from developing the capabilities needed to take on more advanced work. GenAI-driven code generation may offer smaller firms the ability to produce code rapidly, but it also creates a competitive imbalance. Firms relying on these tools risk undercutting the development of workers’ technical skills and, hence, their ability to tackle complex projects and advance up the value chain. With the increasing volume of code generated by AI, companies will face heightened pressure to employ highly skilled professionals who can refine and optimize these outputs, ensuring technical precision and creative consistency. Firms that do not secure such talent may struggle to compete with larger organizations that already have robust capabilities in this area. The software development industry needs talent for developing AI—with skills such as applied mathematics, computer science, and ML—as well as workers trained to use AI at work, such as developers able to use GenAI copilots. Some software companies have also launched their own training initiatives. One software company directly collaborates with local schools and government officials in 9 regions to run free training academies, offering 6- to 10-month programs in software development for high school students who score high on English and critical thinking tests. The initiative began with 52 schools in the Khorazm region and has since expanded to 9 regions, screening thousands of applicants but accepting only the top 0.1 percent (refer to box CS3.6 for examples). Impact on workers AI is reducing stress and making workers more productive, but it could hamper the development of deep expertise, with negative consequences for career development. Overall, workers say that AI tools have reduced work-related stress by helping them meet tight deadlines more efficiently. However, they find it difficult and time consuming to identify the best AI solution for various use cases, in part because capabilities are expanding so quickly. However, for individual developers, the rise of AI tools can be a double-edged sword. Although these tools accelerate code production, they may also encourage a reliance on superficial expertise, where workers produce functional code without a deep understanding of core concepts. This gap in expertise could hinder their ability to diagnose errors, optimize performance, and solve complex problems. Consequently, developers may need to invest in deeper skill development to remain competitive in an environment in which advanced technical knowledge will be crucial (refer to box CS3.7). C ountry C ases   51 BOX CS3.6  Private and public-private initiatives to address AI skills shortage in Uzbekistan Several private and public-private training programs in software engineering, machine learning, deep learning, and data analytics have been launched in recent years in response to growing market demand. IT Park University IT Park University is a partnership between EPAM, a global software company, and IT Park Uzbekistan under the Ministry of Digital Technologies of Uzbekistan. It offers bachelor’s and master’s programs in software engineering, emphasizing practical IT education through a hybrid learning format. Students engage in four practical internships, including international opportunities, to gain real-world experience. The curriculum is designed to align with global IT industry standards, ensuring graduates are well prepared for careers in data science, artificial intelligence (AI), and related fields. The university operates in a hybrid format, in large part because of parental skepticism about fully online education, with the IT Project Management and AI Master’s Program being the most popular. It enrolls 20 students annually and is taught by EPAM foreign experts. IT Park branches in each region serve as physical locations for the university and sites where aspiring students can take admission exams. MAAB Academy Established in 2019 and accredited as a Microsoft Solutions Partner in Data and AI, Digital and App Innovation, Infrastructure and Security in 2023, MAAB Academy offers comprehensive 10-month programs in data analytics and data engineering, focusing on practical skills in advanced Python and SQL and on industry-standard tools such as Azure Data Factory and Power BI. Students engage in hands-on learning with real-world projects, guided by experienced mentors, to master data extraction, transformation, analysis, and visualization. The academy’s campus provides a collaborative environment, including 24/7 co-working spaces, to support immersive education in data and AI disciplines. Mohirdev Mohirdev is an online platform that offers comprehensive data science and AI courses based on curricula from Google, IBM, and Kaggle, enabling students to become proficient in these fields within 6 months. The programs emphasize practical experience, guiding learners through real- world projects and providing hands-on training with modern technologies. According to firm leadership, Mohirdev has about 2,000 students studying AI. By collecting and analyzing job market data from job boards in Uzbekistan in both Uzbek and Russian, they keep their courses aligned with market demand. 52  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S BOX CS3.7  Upskilling software developers for outsourcing in an AI-enabled world As artificial intelligence (AI) automates certain software development tasks, human oversight still remains essential. Predictions that AI will replace developers within 1 or 2 years are not supported by currently available data. AI has started automating tasks at the lower end of the skills spectrum, which are more commonly undertaken in developing countries, but human oversight remains essential in shaping, refining, and understanding AI-generated work. New skills required from software developers by AI automation will be focused on managing AI systems. Because large-scale AI adoption is still in the early stages, upskilling developers to handle AI workflows is crucial in the short to medium term. This includes practical AI integration, including setting up AI-based workflows and agentic systems, testing and refining AI outputs, and handling model upgrades from vendors as well as high-level product management skills to clarify project goals and requirements, oversee AI-driven prototypes and implementations, and balance technical feasibility with user needs. Outsourcing projects of the future may be increasingly focused on implementing prototypes. Future outsourced projects might feature a product team in a high-income country creating three prototypes of a product, creating the specifications using AI, and delegating the implementation of those three prototypes to an AI-enabled human team in a developing country. To enable such projects, it will be essential to train developers in practical AI integration skills, such as setting up agentic workflows, testing and troubleshooting AI systems, and managing frequent model updates. To acquire such skills, developers will need an intermediate background in software engineering plus a good understanding of AI and product management concepts. Software developers from developing countries will also need to be upskilled to think more strategically about product goals and how best to guide AI systems toward those objectives. Source: Denisov-Blanch and Obstbaum 2025. Firms need support for talent development to take advantage of AI opportunities. The software development industry needs talent for developing AI—with skills such as applied mathematics, computer science, and ML—as well as workers trained to use AI at work, such as developers able to use GenAI copilots. Some software companies have also launched their own training initiatives. One software company directly collaborates with local schools and government officials in 9 regions to run free training academies, offering 6- to 10-month programs in software development for high school students who score high on English and critical thinking tests. The initiative began with 52 schools in the Khorazm region and has since expanded to 9 regions, screening thousands of applicants but accepting only the top 0.1 percent. Policy Considerations and Summary Introduction This section distills policy-relevant barriers and risks for artificial intelligence (AI) adoption in digitally delivered services (DDS) in developing countries. It summarizes key challenges facing firms, as discussed in the country cases, in adopting AI solutions and outlines risks and opportunities for the years ahead. It first highlights five barriers slowing adoption: policy uncertainty, cross-border data limits, developer-skills gaps, insufficient worker upskilling, and uneven digital infrastructure. It then examines two risk channels—shifts in value creation and tightening skills demand—that could disadvantage firms—particularly smaller ones—and constrain job growth. Finally, the section outlines priorities for each country DDS archetype and suggests a framework for thinking about policy development at the intersection of AI and DDS. Policy-relevant barriers The country case studies in the previous section point to five policy-relevant barriers that are slowing AI adoption in DDS and undermining the development of local AI ecosystems. • Policy uncertainty deters investment and experimentation. • Restrictions on cross-border data flows raise costs and limit access to training data and markets. • Fragmented responses to the AI-developer talent shortage leave firms competing for a very limited pool of skilled workers. • Insufficient worker upskilling limits frontline adoption. • Digital-infrastructure gaps across regions constrain deployment. Together, these frictions delay firm-level uptake, stall the formation of a local ecosystem capable of developing AI solutions, and slow efforts to prepare workers for an AI-enabled workplace. Policy uncertainty Policy uncertainty—from the absence of clear guidelines to vaguely formulated policies—has forced firms to adopt a risk-hedging approach, making them reluctant to invest in technologies and approaches that could help advance AI development and adoption but may be deemed noncompliant or unlawful and, therefore, punishable, either now or in the near future. Because of such hesitancy, firms miss out on opportunities that could significantly enhance their AI capabilities. This factor puts firms operating in an uncertain policy environment at a competitive disadvantage compared with rivals operating in jurisdictions with clear and stable regulatory frameworks. 53 54  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Limitations on cross-border data flows Data localization laws, which mandate that certain types of data remain within national borders, are a key barrier to AI development and deployment in DDS. Such laws not only prevent service providers from using the many AI tools and solutions already implemented in international cloud providers’ environments, such as AWS and Microsoft, but also hinder their access to high-powered computing. Although some firms use synthetic data to get around this issue, this approach has its limitations. These restrictions drive up the cost of developing AI solutions, putting them out of reach for all but the most resource-rich firms. Under restrictive data localization laws, companies must invest in local infrastructure that often falls short of global standards because export restrictions limit imports of hardware. Lack of a cohesive approach to the AI developer skills shortage Specialized talent remains a significant barrier to AI development and deployment, as firms struggle with a shortage of specialists with the foundational skills in data science, machine learning, applied mathematics, and Python required to develop AI models and applications. Ad hoc initiatives of private sector firms that partner with universities to create training programs cannot replace a cohesive government-supported approach and large-scale investment in AI talent development. AI development and integration and, by extension, the provision of competitive AI-enabled services to international clients cannot successfully develop in the absence of government-supported education and training programs. Limited efforts to upskill workers to use AI Where basic education is weak or lacking, firms are already struggling to find the talent they need to fill even lower-skilled jobs, such as in contact centers. With AI automating routine tasks, the demands on human employees’ skill sets will become more complex, further squeezing the talent pool. Private firms, particularly in low-skilled, high-turnover sectors, lack the incentive to provide upskilling at scale, even when foundational competencies of the workforce allow this. Indeed, just 23 percent of workers participating in a global survey received training from their employers on how to use AI at work (Adecco Group 2024). Firms look to governments to fill in the skills gap by instituting large-scale proactive upskilling and reskilling programs that would help workers adapt to the demands of using AI at work and broaden the talent pool for DDS exporters, so that they can continue growing their footprint and graduate to a higher level of value added. Digital infrastructure limitations in regions Limited access to reliable connectivity and power outside of large cities significantly hinders firms from recruiting local talent. As new AI tools, such as, for instance, accent neutralization software, create job opportunities for new pools of candidates, firms are often unable to connect with them or fully utilize the talent available in regions lacking essential connectivity and power, which constrains DDS growth and job creation. Policy-relevant risks AI in DDS poses two policy-relevant risks: • Shifts in value creation and • Changing skills demand. P olicy C onsiderations and S ummary   55 First, value may migrate toward larger, more technology-intensive firms and offshore locations, disadvantaging smaller or less advanced domestic firms and threatening job creation. Second, AI may raise skill requirements, tightening already scarce talent and weakening DDS firms’ ability to expand employment. These risks are more pronounced for lower-skill segments—such as contact centers and business process management—that generate large numbers of relatively well-paid jobs in many DDS-exporting economies, including those with modest education levels. As AI automates portions of this work, the job-creation engine in these sectors may slow. Therefore, the policy challenge is twofold: Invest in rapid upskilling or reskilling to move workers into higher-value roles, and cultivate alternative, lower-skill sectors that are less exposed to AI—ideally without significant wage losses for workers who cannot be upskilled. Potential shifts in value creation AI adoption may disproportionately benefit larger firms, multinationals, and high-skilled sectors and countries, shifting value creation away from small firms, local companies, and lower-skilled sectors and countries. From smaller to larger firms Large digital services export firms get a competitive edge by leveraging proprietary and custom- developed AI systems to tailor technology to specific use cases and significantly boost operational efficiency. This strategic advantage not only delivers substantial cost savings but also allows these firms to pursue more sophisticated, high-value projects, further widening the gap between them and smaller competitors. Meanwhile, smaller firms rely primarily on off-the-shelf AI tools in an ad hoc manner that does not fully exploit their potential because they typically serve less-sophisticated clients, which limits their exposure to the complex challenges of AI-related ethical issues, AI bias, data privacy, security, and cyber risks. Not being well versed in these AI-related issues, small digital services export firms may find it increasingly hard to beat a path to working on high-value projects and compete with larger firms for more advanced and lucrative initiatives. From local to multinational firms Multinational firms are advancing rapidly in AI integration by leveraging proprietary global tools, which include mandatory, role-specific AI training courses for employees and firm-paid access to high-capability AI tools. They further enhance operational efficiency by providing employees with AI-powered access to the firm’s entire knowledge base and by embedding responsible AI practices through centrally developed policies and frameworks. As a result, these firms are better positioned to capture higher value in an AI-enabled world than most local firms in developing countries that have limited access to AI tools, AI use frameworks, and global knowledge bases. From lower-skilled, labor-intensive to higher-skilled, technology-intensive sectors and countries Firms in higher-skilled sectors, such as software development, are integrating advanced AI tools such as coding copilots that enhance worker productivity to accelerate routine tasks while preserving human insight for complex problem-solving. This approach not only enables workers to focus on higher-value activities but also fosters an environment of innovation and skill development that can drive substantial value creation. 56  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S In contrast, firms in lower-skilled sectors such as contact centers tend to adopt AI solutions that fully automate basic tasks, which, although increasing efficiency, limit opportunities for employee upskilling and ultimately reduce the potential to capture higher value in an AI-enabled world. This shift may disproportionately benefit countries that specialize in higher-skilled, software-intensive sectors, such as IT services, software development, and data analytics. Differential impact on high-skilled and low-skilled jobs The development and adoption of AI in DDS jobs is likely to affect low-skilled and high-skilled jobs differently. The lowering of barriers to entry will be mostly felt at the low-skilled end, and many DDS jobs may require a higher skill level than today, fueling increased competition for AI-compatible talent. Reducing barriers to entry for low-skilled and insecure jobs AI is reducing barriers to entry in some DDS jobs by enabling workers with limited domain-specific skills to perform low-skilled tasks, such as creating basic animation and games, handling simple coding jobs, and executing straightforward analytical tasks. Although large digital services firms may increasingly automate these tasks, smaller upstarts will likely emerge to serve clients who cannot engage with big providers or who require one-off support. However, as these tasks become commoditized, with AI handling most of the work, the economic returns will be low and dependent on economies of scale, making it challenging for small firms to compete. Ultimately, this development could benefit freelancers in lower-income economies, provided that affordable and accessible AI tools remain available. However, these new jobs will likely remain low paying and precarious. Such low-skilled, low-paying AI-assisted jobs would primarily be taken on by workers who are unable to access, or do not have the foundational skills required for, upskilling. Increasing skill requirements for employment in stable jobs At the same time as low-skilled tasks are commoditized, the focus of AI-assisted DDS jobs may shift toward more complex work that skilled workers—assisted by AI—are better placed to perform. As AI’s analytical capabilities are enhanced, outsourcing destinations in low-income countries may begin handling advanced tasks traditionally performed in high-income countries, potentially narrowing the global skills gap; however, the real beneficiaries in these regions will be those workers who can upskill and adapt to higher-value roles that complement AI. In contrast, those in lower- skilled or junior positions will face a greater risk of displacement or wage stagnation because of the automation of routine tasks, highlighting the critical need for ongoing training and upskilling to maintain or increase wages and job security in the evolving digital services market. Disproportionally benefiting deep experts GenAI tools can already do much of the time-consuming work involved in generating ideas, concepts, or compounds. Such work—when performed by humans—requires not only basic domain expertise but also contextual understanding. A creative director needs to understand their client and audience, a car designer needs to grasp the requirements for the new vehicle, and the researcher needs to understand how materials bind together to form potentially useful new ones. GenAI tools can replace basic domain expertise but not contextual understanding, and, therefore, those who P olicy C onsiderations and S ummary   57 stand to gain the most from the adoption of these tools are contextual, or deep, experts who can select and refine GenAI output using their superior knowledge of the subject matter and context. This development may sharpen the divide between basic and deep expertise, commoditizing the former while increasing the premium for the latter. Intensifying competition for talent that can be upskilled As AI capabilities continue to advance, competition for talent among both firms and countries will likely intensify, not just for people with technical competencies for AI development but also for those who possess the foundational education and soft skills, such as empathy and critical thinking, needed to be upskilled to work with AI tools in DDS jobs. Larger firms, with the advantage of dedicated HR teams and the ability to offer higher salaries, are better positioned to attract and retain such talent, putting smaller firms at a significant disadvantage and potentially leading to an even greater concentration of higher-skilled workers. At the national level, countries with robust and well-developed education systems that emphasize both broad-based foundational learning and rapid upskilling through digital tools and AI-enhanced education are poised to excel, thereby giving them a competitive edge in the global race for high-value talent in the age of AI. Disadvantaging women, older adults, and those with lower education and income Current AI use patterns show that younger, more highly educated, higher-income individuals and men are the primary users of AI, gaining early productivity advantages that could further cement their position in the labor force. As experience with AI tools helps identify the best ways to use them, those who already have access and training are likely to pull further ahead, leaving older, less-educated, lower-income individuals and women at a disadvantage when it comes to DDS jobs (De Simone et al. 2024). Without targeted training and intervention to broaden AI access, these disparities will intensify, exacerbating existing social and economic inequalities. Box CS3.8 addresses a recently exacerbated risk: overarching considerations of economic uncertainty and protectionism. BOX CS3.8  Potential impact of increased economic uncertainty and protectionism on DDS trade Mounting tariff escalation and the stop‑start nature of trade talks have injected a “wait‑and‑see” premium into trans‑Atlantic business planning, already depressing discretionary technology budgets and stalling new outsourcing deals. Potential Near-Term Effects Higher input costs and the fear of still‑higher duties are already pushing EU and US clients to pause or shorten service contracts, as seen in project delays at India’s top information technology exporters and cuts to digital marketing and cloud‑spend that ripple through business process outsourcing hubs from Manila to Bangalore (Briones 2025). Firms also face rising compliance overhead as they hedge against multiple tariff scenarios and digest fast‑shifting rules, a drag that Deloitte warns is weighing on sentiment, investment, and artificial intelligence modernization road maps in 2025 (Barua and Wolf 2025). (Continued) 58  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S BOX CS3.8  Potential impact of increased economic uncertainty and protectionism on DDS trade (Continued) Potential Long-Term Effects If tariff threats harden into a permanently higher trade‑risk environment, three structural shifts loom: 1. Regulatory fragmentation, with EU‑US battles over digitally delivered services (DDS) taxes and data localization spilling into services trade and raising the cost of cross‑border data flows (Broadbent 2025). 2. Accelerated “friend‑shoring” and regionalization of delivery centers toward politically neutral or preferential‑trade markets, eroding the scale advantages that some specialists in developing countries—such as India and the Philippines—now enjoy. 3. Faster adoption of generative AI tools by EU and US clients to substitute for outsourced labor— trends that could chip away at developing countries’ price differential over time. Today’s turbulence clouds near‑term revenue pipelines for DDS while casting a longer shadow over the sector’s cost competitiveness and market access to the European Union and the United States. Policy considerations by country archetype Although each country will shape its own approach to harnessing the opportunities and addressing the risks of AI in the DDS sector on the basis of its specific needs and context, the case studies and literature review suggest policy considerations that are most relevant to each of the DDS country archetypes introduced in the “Background and Literature Review” section. Digital export leaders (higher skilled, higher digital exports), such as Malaysia, should • Continuously invest in innovation to strengthen comparative advantage. Prioritize AI-focused research and development, partnerships, and advanced skills training to sustain leadership in high-value DDS exports. • AI-enable the workforce. Scale up retraining and upskilling programs for workers to adapt to evolving AI technologies and maintain competitiveness in high-end DDS. • Strengthen supportive regulatory frameworks. Review regulatory frameworks to ensure they guide AI innovation toward the public good, strengthen trust, and help shape AI to reinforce comparative advantages in DDS. High-skilled, high-potential (higher skilled, lower digital exports) economies, such as Viet Nam, should • Facilitate and incentivize AI-enabled DDS exports. Invest in the digital, business, and financial infrastructure that allows firms to leverage existing high-skilled talent to effectively produce, deliver, and scale AI-enabled DDS internationally. Provide funding, incubators, and tax incentives for tech start-ups and scale-ups specializing in AI-driven services and encourage them to sell abroad. • Get workers AI ready. Implement targeted AI upskilling programs across key sectors— prioritizing high-skilled workers—to strengthen competitiveness by enhancing productivity and maintaining cost advantages while moving into AI-enabled DDS. P olicy C onsiderations and S ummary   59 • Address policy barriers to digital trade. Review regulatory frameworks to clarify intellectual property laws, align data privacy regulations with international best practices to enable cross- border data flows, and establish digital trade agreements to facilitate cross-border service delivery. Mature, lower-skilled exporters (lower skilled, higher digital exports), such as the Philippines, should • Identify and invest in developing a comparative advantage in higher-skilled services. Use a network of global firms in the country to identify potential higher-skilled subsectors of DDS, incentivize firms to develop capabilities in these subsectors, and commit to upskilling workers to meet firm demand. • Build local AI capability and enhance worker flexibility. Invest in AI upskilling for workers, AI developer training, and AI-related research to develop domestic capabilities to underpin DDS exports. In parallel, in the short term, strengthen high school curricula to develop foundational logic and the capability to work with AI tools to lay the groundwork for effective upskilling to higher-skilled and AI-enabled work. • Strengthen digital infrastructure. Improve digital connectivity and business environments to attract investment in higher-skilled DDS and reduce dependency on low-value exports. Emerging builders (lower skilled, lower digital exports), such as Uzbekistan, should • Leapfrog to enable AI-native DDS through investment. Prioritize investments in connectivity, computing, and data, as well as programs to train workers to develop alongside AI, and work with AI tools to produce DDS that use AI to enhance comparative advantage. • Identify and adopt emerging good practices in AI. Continuously capture emerging AI policy and programmatic lessons from established digital economies and adapt to local conditions to support the development of comparative advantage in AI-enabled DDS. • Build foundational digital and AI skills. Establish broad-based digital skills development programs that integrate AI upskilling and strengthen programs in science, technology, engineering, and math to prepare the workforce of the future for jobs in high-value, AI-native DDS. Toward a framework for addressing the impact of AI adoption on DDS To address the impact of AI adoption on DDS, policy makers should collaborate closely with the private sector and shape a joint approach. We propose a high-level framework to support policy makers in shaping a response in their countries as they diagnose the impact, determine a response, and deploy policies (refer to figure CS3.11). First, policy makers must diagnose the extent to which DDS jobs in their country are exposed to and can be complemented by AI, how prepared workers are for AI, and what the demand for these services is. Second, they must determine how AI interacts with market failures and how policies need to be adjusted to address emerging risks. Third, they need to design and deploy policies to address the impacts and harness the benefits of AI. Policy makers must engage the private sector throughout this process to understand trends, impacts, and firms’ plans, because these will necessarily inform policy development and implementation. 60  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S FIGURE CS3.11  High-level framework for shaping a response to the impact of AI on DDS Diagnose Determine Deploy Understand how and to what Assess how Al is a ecting existing Adjust and add targeted policies, extent existing DDS jobs are market failures and potentially creating programs, and investments in exposed to and can be new ones, review existing policies to infrastructure, human capital, labor Public complemented by Al, assess worker determine whether they need to be market policies, business and sector preparedness, and map the adjusted or new ones added to help investment climate, and country's comparative advantage in harness the bene ts of AI and mitigate international market integration. digitally delivered services. the impacts of AI-driven displacement on workers. Through ongoing dialogue, policy makers should seek to do the following: Understand at what pace and for Understand whether and how rms are Design interventions in close what use rms are planning to planning to upskill and reskill workers on partnership with the private sector Private adopt Al solutions in their Al and whether Al has introduced new and proactively engage rms as sector operations and how Al will a ect barriers to their operations in the partners in programs and investments their business plans in the country. country. to accelerate impact. Source: Original figure for this publication. Note: AI = artificial intelligence; DDS = digitally delivered services. References Adecco Group. 2024. Working through Change: Adapting to an AI-Drive World of Work. Annual Global Workforce of the Future Report. Zurich: Adecco Group. https://www.adeccogroup.com​ /global-workforce-of-the-future-research-2024. Amiti, M., and S.-J. Wei. 2024. Demystifying Outsourcing: The Numbers Do Not Support the Hype Over Job Losses. Washington, DC: International Monetary Fund. https://www.imf.org/external​ /pubs/ft/fandd/2004/12/pdf/amiti.pdf. Ando, R., and M. Davies. 2024. “The World’s Call Center Capital Is Gripped by AI Fever and Fear.” Bloomberg, August 27, 2004. https://www.bloomberg.com/news/features/2024-08-27/philippines​ -call-centers-navigate-ai-impact-on-jobs. Arora, A., H. Chheda, H. Heimes, A. Kadyan, J. Silver, and S. Singh. 2021. “Global Business Services: Sourcing Comes of Age.” McKinsey & Company. September 1, 2021. https://www.mckinsey.com​ /capabilities/operations/our-insights/global-business-services-sourcing-comes-of-age. Báo Thanh niên. 2024. “Billionaire Jensen Huang Explains Why NVIDIA Chose Vietnam as Its ‘Second  Home.’” Vietnam.vn, December 9, 2024. https://www.vietnam.vn/ti-phu-jensen-huang-noi​ -ly-do-nvidia-chon-viet-nam-la-ngoi-nha-thu-hai. Barua, A., and M. Wolf. 2025. “Tariffs Will Impact the Economy … and So Will Uncertainty.” Deloitte Global Economics Research Center. April 11, 2025. https://www2.deloitte.com/us/en/insights/economy​ /spotlight/united-states-tariffs-impact-economy.html. Bessen, J. 2018. “AI and Jobs: The Role of Demand.” Working Paper 24235, National Bureau of Economic Research, Cambridge, MA. http://www.nber.org/papers/w24235. Bick, A., A. Blandin, and D. J. Deming. 2025. “The Rapid Adoption of Generative AI.” Working Paper 32966, National Bureau of Economic Research. https://www.nber.org/papers/w32966. Briones, J. A. 2025. “U.S. Tariff Escalation Hits Global Services, Outsourcing Industry Braces.” Outsource Accelerator News, April 21, 2025. https://news.outsourceaccelerator.com/u-s-tariff-escalation-hits​ -global-services/. Broadbent, M. 2025. “Reading Tea Leaves on Transatlantic Digital Trade with Europe.” Italian Institute for International Political Studies. April 15, 2025. https://www.ispionline.it/en/publication/reading​ -tea-leaves-on-transatlantic-digital-trade-with-europe-205861. Brynjolfsson, E., D. Li, and L. R. Raymond. 2023. “Generative AI at Work.” Working Paper 31161, National Bureau of Economic Research, Cambridge, MA. https://www.nber.org/papers/w31161. Cazzaniga, M., F. Jaumotte, L. Li, G. Melina, A. J. Panton, C. Pizzinelli, E. J. Rockall, and M. M. Tavares. 2024. Gen-AI: Artificial Intelligence and the Future of Work. Washington, DC: International Monetary Fund. https://doi.org/10.5089/9798400262548.006. Dell’Acqua, F., E. McFowland III, E. Mollick, H. Lifshitz-Assaf, K. C. Kellogg, S. Rajendran, L. Krayer, F.  Candelon, and K. R. Lakhani. 2023. “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality.” 61 62  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Working Paper 24-013, Harvard Business School, Boston, MA. https://www.hbs.edu/ris/Publication %20Files/24-013_d9b45b68-9e74-42d6-a1c6-c72fb70c7282.pdf. Demirci, O., J. Hannane, and X. Zhu. 2023. “Who Is AI Replacing? The Impact of Generative AI on Online Freelancing Platforms.” SSRN. Unpublished manuscript, posted November 3, 2023, last revised February 21, 2024. https://ssrn.com/abstract=4602944. Denisov-Blanch, Y., and S. Obstbaum. 2025. “Impact of AI on Software Engineering Productivity.” Unpublished manuscript. De Simone, M. E., M. Barron, W. Mosuro, E. Dikoru, and F. Maniolo. 2024. “From Chalkboards to Chatbots in Nigeria: 7 Lessons to Pioneer Generative AI for Education.” World Bank (blog), September 18, 2024. https://blogs.worldbank.org/en/education/From-chalkboards-to-chatbots-in-Nigeria. Dezan Shira and Associates. 2025. “Decree 163 Implementation: Shaping Vietnam’s Telecom Industry in 2025.” Vietnam Briefing, January 8, 2025. https://www.vietnam-briefing.com/news/vietnam​ -introduces-decree-163-guidelines-for-implementing-the-2024-telecommunications-law.html/. Drydakis, N. 2025. “Artificial Intelligence and Labor Market Outcomes.” IZA World of Labor 2025: 514. https://doi.org/10.15185/izawol.514. EduRank. 2025. “28 Best Universities for Artificial Intelligence (AI) in Vietnam.” Updated March 2, 2025. https://edurank.org/cs/ai/vn/#:~:text=Best%20Artificial%20Intelligence%20,For. Eisfeldt, A. L., G. Schubert, M. B. Zhang, and B. Taska. 2023. “Generative AI and Firm Values.” SSRN. Unpublished manuscript, posted May 9, 2023. https://ssrn.com/abstract=4436627. Eloundou, T., S. Manning, P. Mishkin, and D. Rock. 2023. “GPTs Are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models.” Preprint, submitted August 22, 2023. https:// arxiv.org/pdf/2303.10130. Felten, E., M. Raj, and R. Seamans. 2021. “Occupational, Industry, and Geographic Exposure to Artificial Intelligence.” Strategic Management Journal 42 (12): 2195–217. Grand View Research. 2025. Business Process Outsourcing Market 2025–2030: Size, Share & Trends Analysis Report by Service Type (Customer Services, Finance & Accounting), by Outsourcing Type, by Deployment, by End Use, by Region, and Segment Forecasts, 2025–2030. San Francisco: Grand View Research. https:// www .grandviewresearch.com/industry-analysis/business-process​-outsourcing-bpo-market. Hartley, J., F. Jolevski, V. Melo, and B. Moore. 2024. “The Labor Market Effects of Generative Artificial Intelligence.” SSRN. Unpublished manuscript, posted April 10, 2025. https://ssrn.com​/abstract=5136877. Hoa, N. Q. 2025. “Vietnam’s AI Sector in 2025: Regulatory Frameworks and Opportunities for Investors.” Vietnam Briefing, January 31, 2025. https://www.vietnam-briefing.com/news/vietnams-ai-sector​ -in-2025-regulatory-frameworks-and-opportunities-for-investors.html/#:~:text=December%20 07%2C%202022-,Ministry%20of%20Science%20and%20Technology%20(MoST),March%20 15%2C%202024. Houngbonon, G. V., J. T. Mensah, and N. Traore. 2022. “The Impact of Internet Access on Innovation and Entrepreneurship in Africa.” Policy Research Working Paper 9945, World Bank, Washington, DC. https://openknowledge.worldbank.org/server/api/core/bitstreams/5cb7fdd3-ea2e-5392-b6d5​ -77baec4dbb9a/content. Hui, X., and O. Reshef. 2025. “Is Generative AI a Job Killer? Evidence from the Freelance Market.” Brookings Institution, Washington, DC. https://www.brookings.edu/articles/is-generative-ai-a-job​ -killer-evidence-from-the-freelance-market/. Klein, E. 2025. “The Government Knows A.G.I. Is Coming.” New York Times, March 4, 2025. https://www.nytimes.com/2025/03/04/opinion/ezra-klein-podcast-ben-buchanan.html. Korinek, A., and J. Vipra. 2024. “Concentrating Intelligence: Scaling and Market Structure in Artificial Intelligence.” Working Paper 33139, National Bureau of Economic Research, Cambridge, MA. R eferences   63 Lewis-Kraus, G. 2016. “The Great A.I. Awakening: How Google Used Artificial Intelligence to Transform Google Translate, One of Its More Popular Services—And How Machine Learning Is Poised to Reinvent Computing Itself.” New York Times Magazine, December 14, 2016. https://www.nytimes​ .com/2016/12/14/magazine/the-great-ai-awakening.html. Liu, T., C. Deng, and K. Monahan. 2024. “How Generative AI Adds Value to the Future of Work.” Upwork. February 13, 2024. https://www.upwork.com/research/generative-ai-work-value?utm_source​ =chatgpt.com. Liu, Y., J. Huang, and H. Wang. 2025. “Who on Earth Is Using Generative AI? Global Trends and Shifts in 2025.” SSRN. https://ssrn.com/abstract=5495020. Liu, Y., and H. Wang. 2024. “Who on Earth Is Using Generative AI?” Policy Research Working Paper 10870, World Bank, Washington, DC. https://documents.worldbank.org/en/publication/documents​ -reports/documentdetail/099720008192430535. Lohr, S. 2016. “IBM Is Counting on Its Bet on Watson, and Paying Big Money for It.” New York Times, October 17, 2016. https://www.nytimes.com/2016/10/17/technology/ibm-is-counting-on-its-bet​ -on-watson-and-paying-big-money-for-it.html?emc=edit_th_20161017&nl=todaysheadlines&nl id=62816440. Martin, H. 2025. “No More Junior Salesforce Developers? How AI Will Impact the Job Market.” Salesforce Ben. February 26, 2025. https://www.salesforceben.com/no-more-junior-salesforce​ -developers-how-ai-will-impact-the-job-market/#:~:text=Marc%20Benioff%20has%20said%20 that,a%20bid%20to%20push%20Agentforce. Mollick, E. 2025. “Speaking Things Into Existence.” March 11, 2025. https://www.oneusefulthing.org/p​ /speaking-things-into-existence?r=i5f7&utm_campaign=post&utm_medium=web. Nair, V. 2023. “$1M Salary Package: AI Companies Pour Money for GenAI Roles.” Analytics India Magazine, August 17, 2023. https://analyticsindiamag.com/ai-features/1m-salary-package-ai-companies​ -pour-money-for-genai-roles/. Newquist, H. P. 1994. The Brain Makers: Genius, Ego, and Greed in the Quest for Machines That Think. New York: Macmillan/SAMS. OECD (Organisation for Economic Co-operation and Development). 2022. “Viet Nam: Student Performance (PISA 2022).” https://gpseducation.oecd.org/CountryProfile?primaryCountry=VNM& treshold=10&topic=PI. OECD (Organisation for Economic Co-operation and Development). 2024. OECD Services Trade Restrictiveness Index: Policy Trends up to 2024. Paris: OECD Publishing. https://doi.org/10.1787​ /b9e5c870-en. Peng, S., E. Kalliamvakou, P. Cihon, and M. Demirer. 2023. “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot.” Preprint, submitted February 13, 2023. https://arxiv​ .org/pdf/2302.06590. Philippines Foreign Investments Act of 1991, RA 7042. https://asean.org/wp-content/uploads/2016/08​ /P79_Foreign-Investments-Act-of-1991.pdf. Philippines Special Economic Zone Act of 1995, RA No. 7916. https://www.peza.gov.ph/special​ -economic-zone-act. Pizzinelli, C., A. J. Panton, M. M. Tavares, M. Cazzaniga, and L. Li. 2023. “Labor Market Exposure to AI:  Cross-Country Differences and Distributional Implications.” IMF Working Paper 23/215, International Monetary Fund, Washington, DC. https://www.imf.org/en/Publications/WP/Issues​ /2023/10/04/Labor-Market-Exposure-to-AI-Cross-country-Differences-and-Distributional​ -Implications-539656. 64  C O - W O R K E R , C O A C H , O R C O M P E T I T O R : H O W A I I S T R A N S F O R M I N G T H E F U T U R E O F D I G I TA L LY D E L I V E R E D S E R V I C E S Quispe, A., and R. Grijalba. 2023. “Impact of the Availability of Chat-GPT on Software Development Activity.” Preprint, submitted June 16, 2024. https://arxiv.org/html/2406.11046v1. Shakked, N., and W. Zhang. 2023. “Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence.” Science 381 (6654): 187–92. https://doi.org/10.1126/science.adh2586. Statista. 2025. “IT Outsourcing—Worldwide.” https://www.statista.com/outlook/tmo/it-services/it​ -outsourcing/worldwide. Sufianti, S. 2024. “AI, Digital Investments May Be Next Leg of SE Asia’s FDI Growth.” Bloomberg Intelligence, December 9, 2024. https://www.bloomberg.com/professional/insights/regional-analysis​ /ai-digital-investments-may-be-next-leg-of-se-asias-fdi-growth/. Turner, A., G. Miller, and E. Lowry. 2023. High U.S. Health Care Spending: Where Is It All Going? New York: Commonwealth Fund. https://doi.org/10.26099/r6j5-6e66. TTEC. 2025. “RealSkill Shrinks Contact Center Attrition.” https://www.ttec.com/client-stories/realskill​ -bot-shrinks-contact-center-attrition. UNCTAD (United Nations Conference on Trade and Development). 2023. “Digitally Deliverable Services Boom Risks Leaving Least Developed Countries Behind.” September 28, 2023. https://unctad.org​ /news/digitally-deliverable-services-boom-risks-leaving-least-developed-countries-behind. US BLS (Bureau of Labor Statistics). 2024. “Occupational Employment and Wage Statistics Tables.” https://www.bls.gov/oes/current/oes_stru.htm. US Copyright Office. 2025. Copyright and Artificial Intelligence, Part 2: Copyrightability. Washington, DC: US Copyright Office. https://copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2​ -Copyrightability-Report.pdf. “Vietnam’s Data Centre Industrial Is One of the Fastest Growing in the World.” 2023. Savills (blog), December 10, 2023. https://www.savills.com.vn/blog/article/214096/vietnam-eng/viet-nam-s-data​ -centre-industrial-is-one-of-the-fastest-growing-in-the-world.aspx. Violante, G. L. 2008. “Skill-Biased Technical Change.” In The New Palgrave Dictionary of Economics, edited by S. N. Durlauf and L. E. Blume, 2nd ed. https://link.springer.com/referenceworkentry​ /10.1057/978-1-349-95121-5_2388-1. World Bank. 2021. “Crossing Borders.” World Bank, Washington, DC. https://wdr2021.worldbank.org​ /stories/crossing-borders/#:~:text=Cross%2Dborder%20data%20flows%20are. World Bank. 2024a. Services Unbound: Digital Technologies and Policy Reform in East Asia and Pacific. Washington, DC: World Bank. https://openknowledge.worldbank.org/server/api/core​ /bitstreams/64563d60-3afc-4cb6-b006-a79eb818ff58/content. World Bank. 2024b. World Bank East Asia and the Pacific Economic Update, October 2024: Jobs and Technology. Washington, DC: World Bank. http://hdl.handle.net/10986/41998. World Bank. 2025. Digital Progress and Trends Report 2025: Strengthening Artificial Intelligence Foundations. Washington, DC: World Bank. World Bank and WTO (World Trade Organization). 2023. Trade in Services for Development—Fostering Sustainable Growth and Economic Diversification. Washington, DC: World Bank Group. http:// documents.worldbank.org/curated/en/099258110092319807. WTO (World Trade Organization). 2023. World Trade Report: Digital Trade Developments 2023. https://www.wto.org/english/res_e/booksp_e/dtd2023_e.pdf. WTO (World Trade Organization). 2024. “Thirty Years of Trade Growth and Poverty Reduction.” Data Blog by the WTO Secretariat, April 24, 2024. https://www.wto.org/english/blogs_e/data_blog_e​ /blog_dta_24apr24_e.htm. R eferences   65 WTO (World Trade Organization). 2025. “Digitally Delivered Services Trade Dataset.” https://www.wto​ .org/english/res_e/statis_e/gstdh_digital_services_e.htm. WTO (World Trade Organization), Organisation for Economic Co-operation and Development, International Monetary Fund, and United Nations. 2023. Handbook on Measuring Digital Trade. 2nd ed. Washington, DC: International Monetary Fund. https://doi.org/10.5089/9789287073600.071. Yieke, L. 2024. “Africa’s Outsourcing Boom: Young Talent Fuels Industry Growth.” African Business, July 15, 2024. https://african.business/2024/07/trade-investment/africas-outsourcing-boom-young​ -talent-fuels-industry-growth. This case study, along with any associated content or subsequent updates, can be accessed at https://hdl.handle.net/10986/43822.