Pakistan Raises Revenue Program – World Bank financed FBR initiative approx US $ 400 million – Cyber Attacks – Leakage of Taxpayers Data ?

The Federal Board of Revenue (FBR) has undertaken one of the largest ICT (Digitisation) modernisation projects in Pakistan’s public sector under the Pakistan Raises Revenue Program (PRRP) a World Bank-financed initiative worth approximately US $ 400 million. While this program was designed ostensibly to make FBR a “data-driven, AI-powered organization” a review of tender documents, hardware specifications and investment priorities reveal critical strategic gaps .

FBR’s Digital Transformation Roadmap in Jeopardy

FBR manages Pakistan’s two largest fiscal data ecosystems namely Inland Revenue Service (IRS) and Pakistan Customs which collectively generate billions of structured, semi-structured, and unstructured records daily. These datasets now exceed the threshold where conventional CPU (Centralised Processing Units ) – based infrastructure can sustain storage and analytics simultaneously.The Computational Implications for Combined IRS + Customs data exceeds 5–6 petabytes (current) with an annual growth rate of 35–40 %. Real-time ingestion from POS, e-invoicing and ports adds tens of millions of new rows daily.

Data Center – Lack of AI / ML Vision

FBR’s current data-center procurement demonstrates a fundamental absence of AI/ML vision. The investments made, though substantial, are architecturally misaligned with global standards and risk locking Pakistan’s revenue system into a 2010-era technological paradigm. To transition to a data-driven organization FBR must reframe its infrastructure around GPU (Graphic Processing Units) – enabled AI computing, institutionalise data science and model governance and evaluate all future procurements through the lens of analytical capacity and National Data Sovereignty.

AI or Technological Obsolescence

A data center can store information but only an AI infrastructure can interpret it. The transformation of FBR depends on taking that next decisive step. FBR’s transformation toward an AI-driven authority therefore hinges on its ability to process, integrate, and learn from this multi-source data in near real time. The current infrastructure being built is CPU-centric, not AI-centric.  Without AI acceleration, analytical turnaround for fraud, leakage and forecasting remains too slow for operational relevance leading to Technological Obsolescence. The newly setup data center will be outdated within 3 years, requiring costly retrofits for GPU integration and promote Analytical Inefficiency and any public claims of “AI-based revenue intelligence” will ring hollow when infrastructure cannot technically support the workload.

Investigate Procurement Deficiencies or Face Revenue Shortfall

Procurement Deficiencies may be investigated at appropriate level focusing on absence of AI/ML-ready hardware leading to a mismatch between public claims and technical reality and Procurement practices that emphasise cost and vendor convenience over strategic outcomes. The consequence is a high-cost infrastructure with limited analytical value , unable to support the advanced workloads required for predictive risk analysis, fraud detection and policy simulation. Evidence from Procurement Reviews and examination of PRRP-funded tenders on the FBR website (e.g., IDs 4245, 4249, 4284) indicates the focus remains on rack-mount servers, storage arrays and network equipment optimized for traditional database and virtualisation environments. Key configurations highlight Intel Xeon or AMD EPYC CPUs (32–48 cores, 2.3–2.5 GHz) and large memory banks (1– 4 TB DDR5 RAM) adequate for relational databases but unsuited for modern machine-learning tasks.

Glaring Procurement Deficiencies

Lack of Technical Evaluation Expertise is apparent as the tender evaluation process appears led by administrative teams with limited AI/data-center specialisation. No evidence of independent technical design review appears visible. Specifications appear Vendor-Driven as many hardware descriptions match vendor catalogues rather than open-standards design, hinting at vendor influence in shaping technical criteria. Under the guise of Short-Term Budget Utilisation Mindset procurement is managed , yet at no stage in the World Bank Funded Procurement project documentation apperas any indication whether due diligence for an AI infrastructure feasibility study was conducted before final procurement.

Missing also are vital GPU and AI-Specific Provisions as no Tender specifies GPU nodes (NVIDIA A100/H100, AMD MI300, etc.), Tensor or neural processing units, or MLOps orchestration frameworks. Even in high-value procurements such as the particularly Oracle Exadata X11M-2 racks and database appliance (Tender 4284), the configuration remains purely CPU-based with no mention of AI acceleration.

Cyber Attack – Identity Theft

Cyber Security – Without internal AI capacity, FBR remains reliant on external vendors or public-cloud processing , raising both cost and data-sovereignty risks. The 2021 FBR cyber-attack underscored the vulnerability of outsourcing core computing functions. Replicating that exposure in the age of AI would jeopardise the confidentiality mandated under Section 216 of the Income Tax Ordinance.

Mismatch – FBR’s AI/ML Vision & Reality

Strategic Implications Mismatch Between FBR’s AI/ML Vision and Reality- This absence of GPU or AI integration contradicts FBR’s Digital Transformation Roadmap, which positions AI as the foundation for compliance and forecasting and National AI Policy 2025 , which identifies the revenue sector as a priority domain for machine-learning adoption. In effect, millions of dollars of World Bank funds are being spent to reinforce legacy architectures incapable of supporting FBR’s declared digital vision. AI-driven Risk Management System (RMS) in Customs reportedly achieves “98 % accuracy” in valuation checks. The Lifestyle Monitoring Cell launched in 2025 applies ML algorithms to detect tax evasion via social-media analysis. However, these projects run on isolated cloud or third-party systems as FBR’s own infrastructure lacks the compute resources to host, train, or scale such models independently.

Ignoring Global Best Practice

Global best practice dictates AI data centers must integrate Scalable GPU clusters, High-bandwidth fabric (100–200 Gbps), Power density of > 20 kW per rack and AI-specific orchestration tools.Whereas  FBR’s current CPU-only infrastructure fails all four criteria. Workloads such as fraud-network detection or natural-language query of filings that run in hours on GPUs would take days or weeks on CPUs, wasting both time and energy.

National Revenue Intelligence

Proposals to prevent the World Bank Loan from being wasted include a Technical Audit Commission to independently analyse the AI Readiness Audit of all World Bank funded project data-center procurements .Architecture realignment is proposed to update design documents to include GPU nodes, high- speed interconnects and modular AI pods.Establish an FBR Data & AI Directorate to design ML pipelines, manage models and coordinate with PRAL and NADRA. Build a Sovereign AI Compute Fabric inside the FBR G-9 facility to ensure data confidentiality and self-reliance. Launch a Pilot High-Impact AI Projects through focusing on refund risk scoring, POS anomaly detection, customs valuation intelligence. Integrate a National Revenue Intelligence Grid combining Customs, Inland Revenue, SECP, SBP and provincial data streams. Adopt Global Standards through implementing of  OECD Tax 3.0 and IMF GovTech AI principles for ethical and explainable model governance.

 

By Nadir Mumtaz

Trademark (IPO)