Latin America put AI productivity on the ministerial table before publishing the regional scoreboard

THE SIGNAL IN ONE SENTENCE

A ministerial meeting can put artificial intelligence on the table. A scoreboard tells everyone whether anything useful left the room. That is the gap after the OECD's 11th Latin America and the Caribbean Regional Programme Ministerial Summit, held in Montevideo, Uruguay, on September 29 and 30. The summit was titled Inclusive Innovation and Artificial Intelligence Revolution. Its agenda connected AI with firm productivity, small-business adoption, worker skills, education, public policy and the stubborn problem of making economic growth reach more people. The official framing begins with a hard number. The OECD's summit page and Key Issues Paper say labour productivity in Latin America and the Caribbean has stagnated or declined over recent decades and remains at about one third of the OECD average. They describe a regional productivity trap shaped by low-sophistication industries, informality, weak technology diffusion and limited growth among firms. AI was presented as one possible lever out of that trap. The summit's public sessions asked how companies could move from experiments to actual productivity gains, especially small and medium-sized enterprises. Other sessions covered technology transfer, industrial and trade policy, inclusive innovation, education, skills and the use of OECD policy tools. Those are sensible subjects. They are not yet a result. The official Key Issues Paper says a majority of countries in the region have not adopted a national AI strategy. It points governments toward the OECD AI Policy Navigator, which it describes as holding more than 2,300 initiatives across nearly 90 jurisdictions, plus the OECD/GPAI AI Policy Toolkit and an incidents monitor built from media reports. The paper then raises a revealing question: what indicators and features would be most valuable in a regional OECD AI dashboard for Latin American and Caribbean policymakers? In other words, the meeting discussed the scoreboard before the scoreboard existed. The official public documents page listed the Key Issues Paper, background research and statements from stakeholders. When verified on October 1, it did not publish a final ministerial declaration, a regional AI dashboard, funded country commitments or an implementation schedule. That does not mean private discussions produced nothing. It means the public record did not yet give citizens a way to measure what governments promised, who owns the follow-up or when progress should be visible. Business at OECD, the OECD's institutional business advisory body, held an official side event in Montevideo on September 28. Its September 30 account says government, private-sector and OECD representatives discussed business adoption, investment, infrastructure, small firms and worker skills. The group called for interoperable AI rules, stronger digital and computing infrastructure, support for smaller businesses and wider use of existing OECD tools. That is an advocacy position, not a ministerial agreement. Business at OECD represents employers and industry groups. Its priorities deserve to be read as one participant's desired policy package, not as proof that governments adopted it. Its priorities paper nevertheless exposes the measurement problem clearly. The document cites third-party surveys and the 2025 Latin American Artificial Intelligence Index to argue that AI use is rising but advanced implementation remains limited and uneven. It says Brazil holds more than 90 percent of the region's measured computing capacity, while more than half of the assessed countries lack high-performance computing infrastructure. Those figures come through a business advocacy paper and should be checked against their original studies before becoming official targets. The broader point is harder to dispute. A company opening a chatbot account is not the same as a factory improving output, a clinic serving patients faster or a small exporter reaching a new market. Adoption counts activity. Productivity measures whether the activity produces more value from the time, capital and materials used. That difference matters in a region where firms vary enormously. A large bank in São Paulo, a software company in Montevideo, a family manufacturer in Guadalajara and a tourism business on a Caribbean island do not face the same connection speed, electricity price, access to credit, training options or market size. A single regional adoption percentage can make unequal access look like shared progress. A useful scoreboard should therefore begin with firm size, sector, country and location. It should separate experimentation from production. It should distinguish using a foreign general-purpose service from developing local data, software, models or specialized applications. It should show whether small firms can afford the tools and whether rural businesses have the same basic connectivity as capital-city offices. The first panel should measure adoption honestly. How many firms tried an AI tool? How many use one every week? How many integrated it into a core process? How many abandoned the effort, and why? Reporting only the success stories turns policy into a brochure. The second panel should measure productivity. Did output per hour increase? Did delivery times fall? Did error rates improve? Did energy or material use decline? Were gains sustained after the pilot budget ended? A vendor estimate of hours saved is not enough. Governments need audited or reproducible measures that account for the cost of software, compute, integration, review and correction. The third panel should follow workers. Which jobs changed? Who received training during paid work time? Did wages rise with productivity? Were tasks made safer or merely more intense? Did automation remove repetitive work, reduce headcount or quietly move unpaid checking onto employees? A productivity policy that never measures job quality can improve the numerator while making ordinary life worse. The fourth panel should track skills without confusing attendance with capability. Counting course enrollments is easy. The better questions are whether people completed training, whether small firms could release workers to participate, whether instruction was available in relevant languages and whether graduates could use AI responsibly in real work. The fifth panel should map infrastructure. Reliable broadband, electricity, cloud services, data centers and computing access are not glamorous side notes. They determine who can build and who can only buy. A regional ledger should show price, availability, reliability, environmental burden and concentration. If one country holds nearly all measured compute, a regional average becomes a magic trick. The sixth panel should measure local value. How much AI spending stays with local workers, universities, startups and suppliers? How much leaves through foreign cloud bills, licenses and consulting contracts? Are public datasets and language resources being improved? Are locally developed tools reaching other countries in the region? Productivity can rise while most of the financial return travels elsewhere. The seventh panel should track harms and correction. The OECD's incidents monitor can help identify patterns, but media reports are not a complete incident registry. Governments need routes for workers, consumers and public agencies to report failures. The ledger should include discriminatory decisions, privacy breaches, cyber incidents, unsafe automation, misleading content and the time required to investigate and repair them. The eighth panel should cover government promises. Each country commitment needs an owner, budget, deadline, procurement status and public progress update. Regional cooperation needs the same discipline. If ministers agree to share evaluations, build training or adapt the policy toolkit, the public should be able to see who volunteered, what was delivered and what remains late. This is not a plea for one giant ranking. Rankings encourage countries to optimize whatever produces a better number. A useful dashboard should combine shared definitions with country notes and disaggregated data. It should show uncertainty, missing reports and changes in methodology. A blank cell is evidence too. It says the government does not know, cannot measure or has not published. The region also needs a way to challenge the scoreboard. Small-business associations, unions, universities, Indigenous communities, disability groups, civil society and young people should be able to question indicators and add evidence. The official summit did include business, labour, education and youth-related activities around the wider programme. Participation becomes meaningful when it can change what gets counted. Uruguay is a fitting place to start this argument. The summit took place at the Torre Ejecutiva in Montevideo and included officials from Uruguay's planning, innovation, digital-government and AI institutions. The country can help demonstrate what a public ledger looks like. It should not become a template that pretends every country begins with the same institutions or resources. There are practical steps that do not require another summit. Publish the proposed dashboard schema. Define the difference between an experiment and a production deployment. Ask countries to report a small common set of measures. Release the data in machine-readable form. Let researchers reproduce calculations. Add country notes for local context. Update quarterly or twice a year. Archive revisions so progress cannot be rewritten after the fact. For small businesses, pair measurement with service. A firm that reports a skills gap should be able to find accredited training, technical assistance or a financing route. A company without suitable connectivity should appear in infrastructure planning, not vanish from the AI adoption denominator. A pilot that fails should teach the next participant something. Governments should also publish the costs. AI programmes often announce grants, labs and training targets without showing ongoing cloud bills, consultant spending, electricity demand or staff time. The scoreboard should count the whole system. Otherwise a cheap demonstration can become an expensive obligation after the photographers leave. The summit's language about inclusion deserves the same test. An AI programme is not inclusive because a strategy document uses the word. Inclusion shows up in who gains access, whose language works, whose data is used, who gets paid, who bears risk and who has a remedy when the system fails. The OECD has enough policy tools to begin. Its own paper says the region needs evidence-informed governance, incident monitoring, peer learning and practical implementation. The missing step is turning those resources into a public cadence of commitments, measurements and corrections. That is where the ministerial conversation can become useful beyond Montevideo. Countries do not need identical strategies. They do need comparable evidence for the claims they make in common. One country may focus on agriculture, another on public services, another on manufacturing or tourism. The shared scoreboard can still ask whether projects reached production, improved outcomes, included smaller firms, raised worker capability and managed harm. The plain signal is that Latin America and the Caribbean have moved AI productivity into the regional policy conversation. The official material identifies the productivity trap, the infrastructure gaps, the missing national strategies and even the need for a regional dashboard. What it does not yet provide is the dashboard itself, a public action plan or a funded set of country commitments. Until that ledger appears, the region has a serious agenda and an unfinished score.