In 2007, under the most powerful hybrid regime, hard drives from thirteen specific computers at the headquarters of what was then called the Central Bureau of Revenue in Islamabad were physically stolen. Not hacked. Not breached remotely. The drives were disconnected and removed. They housed the personally identifiable information of every registered taxpayer in Pakistan: National Tax Numbers (NTNs), names, addresses, and banking and financial histories. The entire database of Pakistan's tax system walked out of the building.
Benazir Bhutto, former Prime Minister, living in exile, called me shortly after. She told me the data of all taxpayers of Pakistan was gone with the wind. During that conversation, I learned something that stopped me cold: only 1.81 million taxpayers existed in that database. In a country of over 180 million people at the time, that was roughly one per cent of the population paying any direct tax at all. She wanted to change that, as none of the previous governments had been successful in doing that. Expanding the tax base was central to what she intended to do upon her return to Pakistan. She asked me to help.
I told her honestly that I had no idea about income tax. I was working in the United States at the time, leading the rollout of an automated Property Tax System in Michigan's largest county. Property Tax systems, yes. Income tax administration in Pakistan, no. But she was not asking me to become a tax expert. She wanted me to bring together economists who understood both the data and the policy, and she knew where to look. She suggested I discuss it with economists in my own family, people trained by the late Dr Mahbub ul Haq: my brother, at Oxford, and my sister, at Toronto. That conversation in 2007 was my introduction to Pakistan's tax problem. It was also, though I did not know it then, the beginning of a journey that would consume a significant part of my life.
Two months after her return to Pakistan, Benazir Bhutto was assassinated in the garrison town of Rawalpindi in December 2007. The conversation we had about the tax base and about what a properly documented Pakistani state could do for its citizens stayed with me.
Designing AI People Can Trust: A Researcher’s View On Secure Digital Infrastructure
In 2013, we estimated that if Pakistan's 3.5 million richest of rich tax evaders were brought into the tax net, the country could potentially collect at least US$3.5 billion in additional revenues, more than three times the amount Pakistan was then seeking from the IMF.
The Structural Problem
Before my tenure as NADRA Chairman, I served as General Manager of Networks at NADRA from April 2008. In that role, I was part of the team that helped build the technical infrastructure, not only for NADRA itself but for the Benazir Income Support Program as well, one of the largest targeted social protection schemes in the developing world. Routing cash transfers to the poorest women in Pakistan through biometric verification, ensuring the money reached real people rather than disappearing into corrupt intermediaries, was the kind of work that made the mission concrete. Digital systems were not ends in themselves. They were tools to deliver fairness to people the state had long ignored.
What I learned in that period also made the tax problem impossible to separate from the social protection problem. The same state that could not identify its taxpayers was also the state that could not find its poorest citizens without building an entirely new registry from scratch. Both failures had the same root: Pakistan had never seriously tried to document itself.
For decades, Pakistan's tax debate revolved around the same prescriptions: increase rates, impose new levies, or expand withholding taxes on electricity, fuel, telecom, and banking transactions. Governments repeatedly squeezed the documented economy because it was easier to tax those already visible than to identify those who remained outside the system. Yet Pakistan's fundamental problem has never been a shortage of taxable wealth. The problem is that much of that wealth remains undocumented, undertaxed, or politically protected.
Pakistan is not necessarily a poor country. It is a poorly documented economy. A nation of more than 240 million people cannot sustainably function when only a small fraction contributes meaningful direct taxes while large concentrations of wealth remain hidden in plain sight through undocumented commercial activity, underreported income, proxy ownership, and weak enforcement. The consequence is predictable: salaried classes and compliant businesses bear the burden, while the state increasingly relies on indirect taxation that disproportionately affects ordinary citizens.
Unearthing Commerce In Mohenjo-Daro: Ports, Agriculture And Ancient Infrastructure
Good governance is impossible without adequate state capacity to collect revenues. No country can sustainably finance infrastructure, social protection, education, healthcare, or economic resilience if the wealthiest segments of society remain outside the tax net while the poor effectively subsidise the rich. This is where digital governance fundamentally changes the equation.
First Tenure at NADRA: Finding the Money in 2013
During my first tenure as Chairman of NADRA, we began exploring whether integrated digital data could help Pakistan address its chronic tax underperformance. Working with the Federal Board of Revenue, we used data analytics and linear regression models to correlate visible indicators of wealth with tax behaviour. Linear regression was the right tool for that moment. Data availability was limited, and the computational infrastructure for more sophisticated approaches was not yet in place. We worked with what we had, and we were rigorous about it.
Even those early models produced startling findings. In 2013, we estimated that if Pakistan's 3.5 million richest of rich tax evaders were brought into the tax net, the country could potentially collect at least US$3.5 billion in additional revenues, more than three times the amount Pakistan was then seeking from the IMF. That exercise demonstrated something important: Pakistan did not necessarily lack taxable capacity. It lacked the institutional ability and political willingness to identify and tax it fairly.
At the time, the exercise drew international attention because it represented one of the first attempts in Pakistan to combine multiple state datasets to estimate tax liability through data intelligence rather than discretionary enforcement. Combined with evidence of identity (EOI) data (e.g. the CNIC and biometrics), it allowed tax authorities to identify patterns of fraud and take proactive steps to mitigate losses. But the first step in this regard was the digitisation of records and integrating the databases of the government’s revenue streams.
Launch Of AI-Based “Digital Defenders Pakistan” Portal To Strengthen Safe And Inclusive Digital Civic Space
The findings were presented. They were noted. And then a change of government arrived, and the initiative lost its institutional home. Evidence, on its own, is not enough. Without political continuity and institutional commitment, even compelling findings get filed away.
Second Tenure: Prime Minister Imran Khan and the AI Exercise
Years later, during my second tenure at NADRA (June 2021-June 2023), Prime Minister Imran Khan asked me to pick up where we had left off in 2013-14 and build a digital framework for structured data exchange and analytics between NADRA and the FBR. The objective was politically sensitive but economically essential: broaden Pakistan's direct tax base using digital intelligence and evidence-based targeting rather than arbitrary notices and harassment. The strategy, vetted by all stakeholders, was to develop and connect DPI in a way that it becomes a digital reusable asset, empowering citizens with an automated grievance redressal system.
This time, the technology had evolved considerably. During the first exercise, we had relied primarily on linear regression because both data availability and computational maturity were limited. By my second term, advances in artificial intelligence allowed us to move far beyond conventional statistical approaches. We decided not to rely on a single model. Three separate artificial intelligence approaches were independently applied: Light Gradient Boosting, Extreme Gradient Boosting, and Random Forest models. Three teams ran sprints using ‘Agile’ methodology. The objective was not merely to produce exciting numbers. It was to validate the integrity of the findings through methodological convergence. Each model was run independently. The teams did not share interim results. We structured it in this way for a reason. If three distinct analytical paths, using fundamentally different architectures, arrived at the same destination, the findings would be difficult to dismiss on methodological grounds.
Youth, Identity And SDG 17: Building Global Collaboration Through Dialogue
The first two models, Light Gradient Boosting and Extreme Gradient Boosting, produced results that were frankly shocking. They revealed the scale of Pakistan's hidden taxable capacity with a degree of precision that traditional tax administration had never previously achieved. When the findings were presented, Prime Minister Imran Khan was visibly encouraged and commended the effort. He was in a hurry and wanted to announce the result. I requested additional time. I did not want the reform to rest on one impressive presentation or one promising model. We wanted to walk through three entirely separate analytical paths and see whether they independently converged toward the same conclusion. Only then could we say with confidence that the findings were robust, defensible, and ready for policy implementation. He understood, and while walking out of his office, he remarked, “Remember, you are not doing it for me, but for millions of poor people. Their prayers are with your team. Tell them.”
When the Random Forest model ultimately produced results remarkably close to the earlier models, it confirmed what we had suspected. The outcome was not a statistical anomaly. It was evidence. The findings were presented formally to Prime Minister Imran Khan and Minister of Finance Shaukat Tarin. Both understood the scale of what had been found. The political and institutional weight of the exercise was clearly understood at the highest levels.
The Evidence: What the Models Found
And the evidence was overwhelming. NADRA assisted the FBR in identifying an estimated significant tax gap. We did that. At the same time, we recognised the need for realism and administrative feasibility. Rather than immediately targeting the full estimated universe, we advised the FBR to initially focus on approximately 11.97 million individuals (a small subset) who each appeared liable to pay at the very least PKR 35,000 annually in direct taxes.
Switching Off Citizenship: When Identity Becomes A Weapon Of The State
Even this narrower and more pragmatic first phase could have generated approximately PKR 419 billion, roughly over US$1.5 billion (as per the exchange rate then), without imposing additional burdens on already compliant taxpayers. This is if we suppose everyone just pays PKR 35,000. But the actual tax owed by 11.97 million individuals was coming around PKR 526 billion, approx. $2.6 billion. Again, more than our ask to the IMF at that time!
Building the Digital Highway
Importantly, this was not merely a data-sharing project. As part of a broader Digital Public Infrastructure vision, we were effectively building a secure digital highway, an X-Road-style interoperability layer between NADRA and the FBR, to enhance the state's ability to collect taxes intelligently and efficiently. The philosophy behind the initiative was not coercive surveillance. The objective was to reduce discretion, improve fairness, and empower citizens. We envisioned a digital public good through which individuals could review the information the state held regarding their tax profile, verify inaccuracies, and access an automated complaint and grievance redressal mechanism. Digitalisation was intended not merely to strengthen the state but also to strengthen citizen trust.
The project itself was completed. The data models had been validated. The tax gap had been quantified. The digital infrastructure for secure data exchange between NADRA and the FBR had been established. For perhaps the first time in Pakistan's history, policymakers possessed a credible, evidence-based roadmap for broadening the tax base without increasing tax rates on already compliant citizens.
The Resistance
None of this happened without friction. Anyone who has worked inside government reform in Pakistan understands that the obstacle is rarely technical. The technology is achievable. The data exists. The analytical capability can be built. What resists change is the ecosystem of interests that benefits from the status quo.
Why Kashmir Is More Than An Identity Struggle
From the 2007 data theft to 2026, almost two decades have passed, and the rent seekers embedded in the system are winning, not by defeating reform but by delaying it. Tax evasion at scale is not a collection of individual failings. It is an institutionalised arrangement. Large landowners, real estate developers, major traders, and agribusiness interests have operated for decades in deliberate opacity, protected by a political class many of whose members benefit from the same arrangement. When you build a system capable of making that opacity visible, you are not solving an administrative problem in isolation. You are threatening a power structure.
The pushback came in familiar forms. Delays in formalising data-sharing agreements between agencies to the extent that Prime Minister Imran Khan had to intervene to ask the President to promulgate an ordinance to share tax data with NADRA for the purpose of broadening the tax. Likewise, other pushbacks include jurisdictional disputes that materialised precisely when cooperation would have produced results. Informal pressure on officials who were moving too effectively. At no point did anyone say openly that the exercise should not succeed. They simply ensured it moved slowly enough that it could be outlasted. I had experienced the sharper version of this before, but it was a bit new for my team. The lesson from both of my experiences is the same: reform in Pakistan requires not only the right technology and the right evidence but also the personal willingness to absorb the cost of being effective.
Then Politics Intervened
Following the vote of no confidence against Prime Minister Imran Khan in April 2022 and the formation of a new government, momentum behind the initiative began to dissipate. The challenge was no longer technological. The challenge was implementation. Technology had identified the problem; political leadership now needed to act on the evidence. Instead, the initiative stalled after I left NADRA. Rather than building upon a completed indigenous public-sector platform, the government eventually chose to outsource the effort to a private company, effectively reinventing the wheel while continuing to negotiate tax-reform commitments under successive IMF programs. The same custodians who had failed to prevent Pakistan’s biggest data heist in 2007, when hard drives walked out of CBR headquarters, apparently had no qualms sharing sensitive taxpayer data to a private company. The so-called defenders of data security had, in the end, neither secured the data nor built on what their own institutions had produced. They outsourced both the failure and the solution.
A Journey Into Turkmenistan And The Resilient Baloch Identity
The irony was difficult to ignore. We had already built the digital highway. We had already identified the tax evaders. We had already quantified the revenue potential. Yet the country chose to start over. I resigned from my second tenure at NADRA in June 2023. The reasons were several, but they shared a common thread: the accelerating politicisation of digital public infrastructure and the institutional unwillingness to use the state's analytical capacity when that capacity pointed at the powerful.
It’s been three years since I resigned. We hear now that the government is hosting a tax data hub at an “undisclosed location”, plans to roll out a prototype in October 2026 in phases that could take years. The automated Citizen portal, empowering citizens with a grievance redressal system that we were developing, is now called “Faceless Tax System”. Old wine, new bottle. Reinventing the wheel is a new art now. Twice we found the money. Twice, we walked away. Far away from Pakistan, right into the lap of the IMF, with a begging bowl.
What the Numbers Show
The subsequent trajectory of Pakistan's tax base reflects this contradiction. Publicly available figures suggest that Pakistan increased the number of tax filers between 2022 and 2026, with filer counts rising from roughly four million active taxpayers to approximately seven or eight million. On paper, that appears to be substantial progress. But the numbers require careful interpretation. Many of these additions consisted of low-value or nil filers, often driven by documentation pressures, withholding penalties, and Active Taxpayer List restrictions rather than systematic expansion of high-income direct taxpayers. More importantly, even by 2026, Pakistan remained significantly below the number of actionable taxpayers identified through the NADRA-FBR exercise years earlier. That gap is analytically revealing. It suggests that while some progress occurred through digitalisation and compliance pressure, the full AI-driven tax broadening strategy was never implemented at scale. Either the identified individuals were not aggressively pursued, enforcement remained selective, institutional continuity weakened, or political resistance diluted implementation.
A New Diplomatic Identity For Pakistan
The Broader Lesson for Digital Public Infrastructure
Pakistan's unfinished tax transformation offers an important lesson not only for Pakistan but for every country investing in Digital Public Infrastructure. DPI is often discussed in terms of digital identity, payment systems, interoperability platforms, and citizen services. But its true value lies elsewhere. Properly designed, DPI enhances state capacity. It enables governments to deliver services more effectively, reduce leakages, improve targeting, strengthen accountability, and mobilise domestic resources without placing additional burdens on those who already comply with the law.
The NADRA-FBR initiative demonstrated that digital identity, trusted data exchange, and artificial intelligence can be combined to create a powerful public good capable of transforming tax administration. The technology worked. The models worked. The data worked. What remained unresolved was politics!
International institutions that fund DPI investment need to understand this. It is not sufficient to support the building of platforms. The harder question is whether the political conditions exist for those platforms to be used against the interests they most threaten. Without that assessment, billions in investment will produce elegant systems deployed only against the already-visible, leaving the fundamental structure of inequality untouched.
By early 2023, the Pakistani state already knew far more about taxable economic behaviour than it was willing to act upon. The evidence suggested that millions of additional taxpayers could be brought into the system and hundreds of billions of rupees mobilised without imposing new taxes on ordinary citizens. The lesson is therefore not about technology. It is about political will.
The Unfinished Promise
I tell this story with names and dates because I want it to be a record, not just an argument. The phone call from Benazir Bhutto in 2007, the stolen hard drives, the 1.81 million taxpayers, the linear regression models of 2013, the AI exercise of 2021 to 2022, the briefings to Prime Minister Imran Khan, the change of government, the resignation in June 2023, and then eventually institutional capture: these are facts, not abstractions. They form a continuous thread that runs from one leader’s vision for a documented, self-sufficient Pakistani state to an unfinished project sitting in a government data warehouse, waiting.
End Pakistan’s Power Paradox: Establish One National Distribution Company Under Public-Private Partnership
The tools exist. The evidence exists. The roadmap exists. The names of millions of immediately actionable tax evaders exist in DPI, waiting! From the 2007 data theft to 2026, almost two decades have passed, and the rent seekers embedded in the system are winning, not by defeating reform but by delaying it. Genuine tax reforms will have formidable political consequences, because they will redistribute power in society and disrupt the political status quo.
