M Abdullah bin Zubair Hayat
Pakistan’s debt problem is usually told as a story about borrowing. It is just as much a story about not knowing. Official figures differ by definition. The Ministry of Finance put total public debt at Rs86.72 trillion at the end of June 2026, with the debt-to-GDP ratio improving to 68.3 percent from 70.6 percent. The State Bank’s wider measure of debt and liabilities reached Rs99.6 trillion, and servicing it cost Rs12 trillion in the last fiscal year. Both can be correct. The difficulty is that a citizen, a legislator or an investor must work out which perimeter is being quoted. A US State Department review noted that only limited information on debt obligations, including major state-owned enterprise debt, is publicly available. A state that cannot state its own obligations plainly cannot manage them.
The same blindness shapes how money is spent. A July 2026 World Bank assessment found that district funding does not appear to be linked to poverty levels or to indicators such as education, healthcare or infrastructure needs. Reported gaps are large, with Quetta receiving 475 percent more per capita than the rest of Balochistan and Karachi, the smallest gap, still 178 percent higher than other districts. The institutions meant to correct this are not doing so. The Bank describes provincial finance commissions as largely inactive, and local governments’ share of total government spending has fallen from around 10 percent in 2005 to around 4.7 percent in 2024.
Results are poor where the money is supposedly aimed. The World Bank notes that provinces raised education budgets substantially between FY09 and FY23, yet enrolment and literacy indicators stagnated or declined in some regions. A Business Recorder summary of the latest education financing report counts 25.37 million out-of-school children, with 77 percent of ten-year-olds unable to read and understand a simple text. The same report adds that Pakistan still lacks enough disaggregated data on spending by gender, locality, disability and school type, which makes it hard to tell whether girls or rural children receive a fair share. In Khyber Pakhtunkhwa, 94 percent of the FY2025-26 education budget is recurrent, leaving 6 percent for development.
This is the case for data science, and it is a professional case before it is a technological one. Allocation is an estimation problem. If transfers are meant to follow need, a formula can be built from poverty rates, out-of-school counts and facility inventories, then compared with what districts actually received. Gaps between the two can be ranked and published. Outcomes are an inference problem. Districts with similar budgets and very different results can be matched, and quasi-experimental methods can estimate whether a grant raised attendance or merely coincided with a rise. Risk is a prediction problem. Government guarantees stood at Rs4.283 trillion at the end of June 2026, with the power sector accounting for around 56 percent, and exposures of that size call for continuous monitoring, not annual surprise.
None of this works without linked, open and standardised records. Treasury ledgers, school registers, clinic inventories and learning assessments sit in separate agencies and formats. Joining them at the district level is the unglamorous foundation. Publishing the result lets independent researchers test official claims, which is how evidence becomes accountability.
The limits should be stated plainly. Administrative data can be inflated by the agencies that report it, so sample-based verification by an independent body is essential. Models trained on records of the already-served can reproduce the neglect of the unserved. Personal data needs statutory protection before any national platform is built. And no model answers to the public, so every model-assisted decision still needs a named official who can be questioned.
Two reforms would do most of the work. Every major federal and provincial programme should be approved with a written evaluation design that names its outcome measures and comparison group in advance. And the federal and provincial finance departments should publish a single reconciled debt and liabilities statement each year, with the definitions side by side. A state that can show what it owes, what it spent and what it achieved can be argued with. A state that cannot will keep borrowing to repeat the same mistakes.
The writer is student of Data Science in Public Policy at the London School of Economics (LSE)