Blind Spots in Pakistan’s Data Governance

By Roshaan E Ali

Pakistan is building a framework to govern, protect and make better use of its data. But who is assessing whether that data actually represents Pakistan?

The Ministry of IT and Telecommunication has released the draft of the National Data Governance Policy 2026, while the National Artificial Intelligence Policy was approved in July 2025. The policies recognize the importance of data standards, quality and inclusion towards a digital economy and society.

The unresolved question, however, is what mechanism will ensure whether the datasets determining consequential AI systems are sufficiently representative of the people and realities they are intended to serve. Terms such as “fairness,” “ethical” and “inclusive” risk becoming little more than politically correct buzzwords devoid of accountability, unless they are supported by clear standards and a defined mechanism for assessing compliance.

AI systems depend heavily on the data they use and are trained on, and there are consequences if we fail to address the practical implications of the fundamental gaps and limitations within our datasets. Neither policy clearly establishes measurable benchmarks or a mandatory review mechanism to determine whether datasets are sufficiently representative and fit for consequential decision-making.

If data misses important dimensions of representation, such as populations from certain provinces, languages or forms of economic activity, among others, those omissions can become embedded in the systems built on it. The quality and representativeness of data can therefore directly shape how accurately these systems understand, respond to and serve the people and realities they are designed to address.

Existing issues in datasets already show why these questions matter. A 2025 World Bank assessment of Pakistan’s National Socio-Economic Registry (NSER), one of the country’s most important databases for identifying vulnerable households, found that while 84 percent of households were registered, approximately 2.2 million households among the poorest 40 percent remained excluded.

Being excluded from the registry has previously meant being excluded from income support, disaster relief, mother and child nutrition programmes and education stipends.The assessment also identified disparities in coverage across provinces, barriers to registration, and concerns around outdated information and data accuracy.

Without oversight, if such data is analyzed by AI systems or used to inform AI-driven decision-making, these gaps could potentially be automated, and negatively impact populations particularly in underserved areas and informal sectors of the economy. Citizens whose needs and circumstances are not comprehensively captured in digital records may be invisible to systems shaping access to healthcare, education and other government services.

Similarly, policies based on recorded economic activity may disproportionately reflect formally documented businesses while overlooking, for example, a woman running an unregistered home-based business. This risks further marginalizing citizens and economic activity that are already underrepresented in national datasets.

This also creates a risk of a negative feedback loop where populations and segments better documented receive better-targeted services and policies, which generates even more data about them whereas those who are poorly documented receive less, remain underrepresented, and become even harder for data-driven systems to see.

With AI, incomplete national datasets will translate into socio-economic inequalities faster than we think.

Without definitive benchmarks for assessing whether datasets are sufficiently representative and fit for their intended use, we risk making the already underserved and underrepresented even less visible. What lies ahead is the task of deliberating on creating equitable and measurable national standards for data collection and data quality, including clear measures of coverage, completeness, accuracy, timeliness and representativeness that reflect Pakistan’s geographic, political and socioeconomic realities.

Pakistan can set a stronger precedent in AI and data governance by establishing an accountable mechanism to assess and address gaps in representation before they are embedded in consequential systems. The policies should also ensure that the public and private sectors cannot deploy high impact AI systems—which can affect millions of lives—without demonstrating that the data on which they rely or use meets clearly defined standards appropriate to their purpose.

A country can build excellent systems, but with incomplete data, it will still produce incomplete governance.