SDG monitoring marks one of the most significant advances by the global statistical community. Findings from the Task Team on Lessons Learned of the Inter-Agency and Expert Group on SDG Indicators showed how an ambitious measurement framework has evolved into a shared data infrastructure for sustainable development, with a common language, an effective process to translate shared aspirations into statistical measurement, a set of globally agreed methodologies and an expanding evidence base that supports policymaking, accountability and international cooperation.

There has been considerable progress in the availability of internationally comparable data, as the number of indicators in the SDG database has expanded from 115 in 2016 to 233 in 2026, with data records rising from around 330,000 to 3.2 million. Significant strides have also been made in methodological development. When the SDG framework was first established, nearly 40 per cent of indicators lacked internationally agreed methodologies. Today, every indicator has an agreed statistical methodology.

Perhaps the most enduring legacy of the SDGs is not the indicator framework itself, but the strengthening of national statistical systems. Many countries have established SDG coordination mechanisms, developed national reporting platforms, improved data-sharing arrangements and expanded the use of administrative records, geospatial information, citizen data and other innovative sources. SDG monitoring has also driven a global capacity-development agenda, helping countries adopt new methodologies and data standards. As a result, the share of indicators with good country coverage has more than doubled, increasing from roughly one third to almost 70 per cent.

At the same time, the experience demonstrates that ambitious global monitoring requires more than political commitment. Monitoring 17 goals, 169 targets and more than 230 indicators has placed unprecedented demands on national and international statistical systems. Many indicators required years of methodological development. Others remain difficult to measure consistently, either because the concepts are complex or the data sources remain limited. The experience highlights the importance of balancing political ambition with statistical feasibility and ensuring that implementation costs and national capacities are considered from the outset.

Persistent data gaps provide another important lesson. Although data availability has improved substantially, particularly across health, water, energy and partnerships – with some areas having trend data exceeding 80 per cent – data gaps remain a persistent challenge. Progress on measuring gender equality, sustainable cities, climate action, and peace and justice has been much slower, as fewer than one third of the SDG indicators have sufficient trend data. Many countries also struggle to disaggregate statistics by income, education, disability, migration status or other characteristics, leaving gaps in understanding who is being left behind.

Proportion of countries or areas with trend data (at least two data points since 2015), comparing 2019 and 2026 databases, by Goal (percentage)

The central lesson is clear: sustainable development monitoring depends not only on ambitious global indicator frameworks but also on sustained investment in methodologies, institutions, partnerships and national statistical systems. The data infrastructure built through the SDGs is a global public good whose value extends well beyond 2030. Preserving, strengthening and building upon this foundation will be essential for the next generation of sustainable development monitoring and for navigating an increasingly complex data and AI landscape.

Sustainable SDG monitoring depends on long-term investment, but financing has not kept pace, leaving many countries to meet expanding responsibilities with limited, fragmented or declining resources. A historic reduction in official development assistance (ODA) has made this challenge more urgent, exposing the vulnerability of data systems that rely heavily on short-term external support or project-based funding.

In 2025, ODA declined by 23.1 per cent from 2024, the largest annual drop ever recorded and one that pushed ODA back to levels seen at the start of the 2030 Agenda. Further cuts are expected in 2026. Recent surveys of national statistical offices in low- and middle-income countries, carried out by the Inter-Secretariat Working Group on Household Surveys, show that funding reductions are already affecting core statistical programmes of these countries, including SDG monitoring, household surveys, administrative data systems and dissemination platforms.

This financing shock arrives at a moment when the world faces increasingly complex and interconnected challenges, from climate change and conflict to inequality and demographic shifts. Meeting these challenges demands reliable, timely and inclusive statistics.

AI has the potential to support almost every stage of the statistical process. It can improve questionnaire design, assist interviewers, classify images and text, detect inconsistencies, automate quality checks and make complex statistics more accessible through natural-language interfaces. A recent meta-analysis of around 1,800 articles confirmed AI is already being applied across the full survey lifecycle.

Proportion of AI use in survey research tasks, 2019-2025 (percentage)

Source: Recreated based on chart in Buskirk et al (2026), More Parameters Than Populations: A Systematic Literature Review of Large Language Models within Survey Research.

But AI also introduces new risks. Large language models answer data-related queries, often relying on outdated or unverified information rather than official statistics. When reliable official data are unavailable, AI systems fill gaps through extrapolation, interpolation or synthetic generation, producing figures that may appear authoritative while masking significant uncertainty. Without trusted official statistics, misinformation can scale faster than ever before.

AI also raises governance questions: whose data are used, whose realities are represented and how trusted are those who control the systems themselves. The opportunity is not simply to adopt AI but to shape it in ways that reinforce the values that underpin official statistics: professional independence, transparency, methodological rigour, accountability and protection of confidentiality. But seizing that opportunity requires deliberate actions. Going forward, AI and future innovations for official statistics must be intentionally designed and governed and held to the same standards of public trust that statistical systems have been established over decades.

Official statistics must serve people, not technologies. A people-centred system begins with inclusion – ensuring that data production captures population diversity, allows meaningful disaggregation and gives communities opportunities to shape what is measured, how data are collected and how findings are used.

Yet many population groups are still underrepresented or absent from official data systems because they are difficult to reach, insufficiently covered by traditional sources or not prioritized in the measurement. For example, only 29 per cent of responding countries reported statistical programmes for internally displaced persons, compared with more than 70 per cent for persons with disabilities and youth. Even where participation mechanisms exist, they do not always translate into action. In 2024, high levels of community participation in rural drinking-water policy were reported in only 30 per cent of cases, despite the availability of broader engagement processes.

Technology already has the potential to reach those that traditional data sources miss through satellite imagery, mobile phone data and data integration. AI can help realize such potential. AI tools can analyse community feedback, complaints and social discourse at scale. Assisted by AI, dissemination tools can make statistics more accessible to non-specialists. And enhanced quality assurance can redirect capacity towards community engagement and other necessary areas of work.

Making people the guiding principle builds a foundation for AI governance that technical frameworks alone cannot provide. The question of whom statistics are to meant to serve must remain at the heart of every decision about AI adoption. Algorithmic shortcuts cannot replace engagement, and efficiency gains must not come at the cost of participation. National statistical offices (NSOs) that are committed to retaining these principles will be better placed to avoid the governance failures AI adoption can bring – not because they have a new rule book but because they have stayed true to what made official statistics trustworthy in the first place.

Data have little value if people do not trust them. For decades, official statistics has earned that trust by maintaining professional independence, transparent methodologies, high-quality and accessible data, and protecting confidentiality.

When official statistics are outdated, incomplete or difficult to access, misinformation can fill the void. Though statistical performance indicators (SPI) have demonstrated improvement, with scores rising from 59 to 70 between 2016 and 2023, they also reveal persistent gaps. Low- and lower-middle-income countries lag far behind high-income countries. While demand for trusted, timely and disaggregated data is growing, the countries with the greatest needs may face the steepest constraints as funding support declines.

AI introduces a new layer of challenges to this already uneven environment. First, AI systems learn from available data, and where data are missing, they may fill gaps with numbers that rest on weak empirical foundations. In populations that have historically been undercounted or completely left out, AI risks deepening those absences and masking the underlying invisibility rather than resolving it.

AI has also changed the calculus on confidentiality – a core commitment of official statistics. A 2023 review found that 93 per cent of the 167 national statistical laws included a confidentiality clause. That near-universal legal commitment was built in an era when anonymization techniques were robust against available re-identification methods. Such is no longer the case. AI has made it possible to re-identify individuals, even from aggregate statistics, by combining open official data sets with the vast commercial data ecosystem. Without first strengthening the technical and governance frameworks that protect confidentiality, feeding official data into AI models risks eroding the trust that was built over past decades.

AI also adds something genuinely new here: the public no longer reliably knows who is behind a statistic, and the institution responsible for an error may not be the one that actually produced it. NSOs must therefore assert their role as the verifiable source of record – documenting methods, quantifying uncertainty and labelling observed data as distinct from modelled output wherever statistics are used, not only where they are produced. Strengthening confidentiality safeguards should be a condition of AI-readiness, not an afterthought. The standard in both cases is the same: AI adoption should be held to the transparency and accountability the statistical system has spent decades earning, even when the system delivering the answer is no longer the NSO itself.

Statistical performance indicators (SPI), 2016-2023, by income status

As AI use surges, so too does the risk that countries will lose sovereignty over their own data ecosystems. Sovereignty has three reinforcing dimensions: control over statistical priorities and data; sustainable national financing; and strong, independent institutions able to coordinate the data ecosystem and govern AI use. In the AI era, all three dimensions become more consequential. Without control over data, financing or institutions, countries will struggle to govern AI adoption in the public interest.

In many countries, sovereignty over data is already limited. An analysis carried out in 2023 found that only around one third of low- and lower-middle-income countries manage their own microdata repositories. Financial independence is similarly constrained. In 2025, 135 countries had national statistical plans under implementation, but only 59 per cent were fully funded. In sub-Saharan Africa, just 15 per cent of countries reported full funding. Countries most dependent on external support for statistical production are also the least positioned to finance the governance infrastructure that responsible AI adoption requires. These figures point to a structural vulnerability: countries are expected to produce trusted, timely and increasingly AI-ready statistics, but many do not yet have full control over the data infrastructure or the sustained financing needed to do so.

Institutional sovereignty presents another challenge. Most countries legally mandate NSOs to coordinate national statistical systems, but this does not translate into the authority to govern the wider data ecosystem. This gap is particularly consequential in the AI era, when NSOs need to set standards, assess risks and shape how official statistics are accessed by others. In the 2021 survey on implementation of the Cape Town Global Action Plan for

Sustainable Development Data, only 17 per cent of responding countries reported being satisfied with their capacity to coordinate across the data ecosystem. In low- and lower-middle-income countries, the figure was only 8 per cent, suggesting institutional sovereignty remains weakest precisely where stronger coordination, governance and public-interest safeguards are most needed.

Sovereignty is the practical foundation on which countries can engage with AI on their own terms. Yet without control over their own data, sustained domestic financing and institutional authority, NSOs cannot audit AI systems, enforce safeguards or hold technology partners accountable. The result is a widening gap: countries with strong sovereignty will shape AI to serve the public interest, while those without it risk becoming consumers of systems built for someone else’s priorities.

Perception of the capacity of NSO to coordinate with the entire data ecosystem, by income group, 2021