When Punjab’s chief minister met Alibaba Cloud officials in Beijing this week, the message was that Punjab’s governance system is being integrated with artificial intelligence, that it is the first province to establish an AI policy, and that cabinet members have been given AI training. This is a welcome ambition. But there is a difference between announcing an AI policy and actually running public services on it. Across Pakistan, that difference is the real story.

The gap is not unique to Pakistan, but it is wider here than among regional peers. Pakistan possesses world-class foundational databases such as the National Database and Registration Authority (Nadra), yet has struggled to build an AI or advanced service-delivery layer on top of them.

Contrast this with India, where Aadhaar and DigiLocker have moved document verification onto active digital rails used by hundreds of millions. Riyadh’s Saudi Data & AI Authority serves as the national data and AI authority, publishing a maturity index that rigorously evaluates and ranks government entities. Even Bangladesh, closer to Pakistan’s socio-economic starting point, has embedded AI into its daily digital service delivery through the a2i programme.

Pakistan, by contrast, relies heavily on policy statements and isolated pilot projects. We lack a unified federal framework to drive adoption, a published maturity index to track progress, and clear accountability for actual deployment. These comparisons matter because they expose the missing piece in our approach. The deficit isn’t ambition; it is the institutional machinery required to convert that ambition into functioning public services.

AI systems are only as good as the information feeding them, yet much of Pakistan runs on paper, fragmented records, and inconsistent formats

The reasons for this lag are painfully familiar to anyone who has navigated a Pakistani government department. Data remains the primary bottleneck. AI systems are only as good as the information feeding them, yet much of our public sector still runs on paper, fragmented records, and inconsistent formats.

Without clean, interoperable data, AI projects inevitably stall at the pilot stage. Beyond data, there is a crisis of ownership. Initiatives are frequently announced from the top but rarely assigned to a single accountable owner with a budget, a deadline, and a mandate to deliver.

Then comes the procurement trap: public procurement rules (PPRA) are heavily biased toward physical hardware and construction, making the agile purchase of cloud capacity and AI models agonisingly slow for risk-averse bureaucrats. Finally, there is a severe skills gap. Training a cabinet is excellent optics, but it is not a substitute for building a cadre of mid-level officials who can operate, maintain, and challenge AI systems day to day.

Perhaps the least discussed, yet most critical, is the issue of accountability. Even as departments race to digitise paper records, they must simultaneously build governance frameworks for the algorithms that will eventually process that data. When a government deploys AI in revenue collection, welfare eligibility, health triage or law enforcement, it is making decisions that affect citizens’ rights and public money. Those decisions need to be auditable. Who can inspect the model? What data did it use? How is a wrong outcome detected and corrected?

In Pakistan, these questions have no clear answer, because the oversight framework for AI does not yet exist. This is not a technical afterthought. It is the difference between AI that builds public trust and AI that erodes it.

Ultimately, adoption is a policy challenge, not a technological one. Economically, public-sector AI can slash the cost of service delivery, plug leaks in welfare and tax systems, and save citizens countless hours navigating state bureaucracy. For a country struggling to attract foreign capital, a government that delivers services efficiently is an investment incentive in itself. Governance-wise, deploying AI without data protection, clear ownership, or audit trails poses an unacceptable risk to both citizens and state credibility. Both angles demand the same shift in mindset: AI adoption must be rigorously managed and measured, not merely announced.

The way forward requires structural reform rather than superficial tech fixes. While service delivery is largely provincial, Pakistan urgently needs a unified federal framework to set standards and prevent a fragmented rollout. We need a published maturity index that ranks departments and provinces on actual deployment, making progress publicly visible. We must overhaul procurement rules to flexibly buy cloud and AI services. We desperately need to pass the long-stalled data protection law. Most crucially, we need an independent oversight function to ensure that algorithms making decisions about citizens are audited just as rigorously as public finances.

None of this requires Pakistan to build frontier models or rival the world’s largest technology companies. It requires the harder, less glamorous work of making existing institutions deliver. Punjab’s AI policy and the Alibaba meeting are a start. The test is whether, a year from now, a citizen can point to a public service that runs measurably better because of AI, and whether an auditor can verify it. Until then, AI in Pakistan’s public sector will remain what it has been: a policy ahead of its adoption.

The writer is Director of Internal Audit & Risk Advisory with 21 years of Big 4 and GCC executive experience

Published in Dawn, The Business and Finance Weekly, September 7th, 2026