ISLAMABAD: Pakistan’s tax authority is preparing to expand artificial intelligence checks on tax returns to identify undeclared income and assets, a senior official said this week, part of a push to increase revenues under an IMF-backed economic reform program.

Pakistan has struggled to bring more economic activity into its tax system, leaving a narrow base of taxpayers and limiting the government’s ability to finance public services without borrowing. Stronger collection is a central objective of its reform commitments to the International Monetary Fund.

The Federal Board of Revenue (FBR) plans to cross-check declarations against property, banking, vehicle and other records. Following a pilot involving a limited number of income tax and sales tax returns, it intends to apply the models to income tax filings for the year ended June 2026 and subsequent sales tax returns, a senior FBR official said.

“This time the system will leave nowhere to hide,” he told Arab News, requesting anonymity because he was not authorized to speak publicly.

“It does not get tired and it does not miss a line. Our campaign this year says ‘FBR Sab Janta Hai,’ (FBR knows everything). With this system, that is no longer just a slogan.”

The official said the models would check returns against tax-law requirements, compare declarations with those of similar taxpayers and identify inconsistencies with information obtained from other sources. A flagged discrepancy would require examination. It would not, by itself, establish fraud.

“Returns flagged by the AI could be selected for audit, face assessment proceedings, or trigger compliance nudges asking taxpayers to correct their filings,” the official said.

The push comes as the FBR extended the income tax filing deadline from Sept. 30 to Oct. 15 following requests from business organizations and tax practitioners’ associations. The extension applies to taxpayers whose returns for Tax Year 2026 were due on Sept. 30.

Around 5.5 million returns had been filed by Wednesday, compared with about four million by the same date last year, according to the official. Filing numbers do not establish how much additional tax has been collected.

Pakistan has set a federal tax collection target of Rs15.26 trillion ($54.8 billion) for the financial year ending June 2027, alongside efforts to strengthen enforcement and digitize tax administration.

CROSS-CHECKING FINANCIAL RECORDS

Faisal Sattar, chief executive of Pakistan Revenue Automation Ltd., the public-sector technology company supporting the FBR, said the authority had established the infrastructure needed to compare taxpayer information across multiple sources.

“FBR has very successfully established a data center where we are now in a position to collect all the required data, which is coming from many, many different sources,” he said.

“So obviously, we have the tax returns data that comes to us, but also from other sources, property data, vehicle data, travel-related data.”

Sattar described an ongoing process of developing analytical models and incorporating them into AI-based systems to identify discrepancies and potential concealment.

“So, you may not tell us something in your tax return. However, there are other sources of finding out,” he said.

“Then we can triangulate what’s happening on the property transactions, what’s happening in the banking data with what’s in your tax return.”

He said the volume of information made manual scrutiny impractical, while automated analysis could make the selection of cases more consistent.

The AI effort accompanies a separate overhaul of audits. On Sept. 25, the FBR announced a National Faceless Center in Islamabad, under which a computerized risk-based system would select cases and assign them automatically to officers, replacing direct dealings between taxpayers and particular officials.

Separate officers would conduct the audit, determine the tax assessment and review the work before an order was issued. The FBR says the arrangement is intended to reduce discretionary intervention and ensure more consistent treatment of taxpayers.

Sattar said the longer-term ambition was to automate parts of the process for resolving discrepancies, rather than simply identifying them:

“So our goal is that the AI systems come to a point where they also give you a settlement proposition, that if you’ve done this wrong, this is what you now need to do to put it right.”