Project Insight, Ten Years On: Revisiting India's Flagship Tax Surveillance Program

In 2016, in the aftermath of the Panama Papers leak and India's demonetisation drive, the Ministry of Finance unveiled Project Insight — a flagship, technology-driven initiative meant to plug the country's chronic tax-base problem by mining social media, transaction data, and personal records to catch evaders. I wrote about the project back then, before it had even launched, flagging concerns about its narrow individual-level focus, its reliance on a single private contractor, and its dependence on international cooperation frameworks that hadn't yet materialized. A decade later, with the government now in the process of retiring the original system in favor of "Project Insight 2.0," it's worth asking plainly: did it work?

What Project Insight Actually Became

To be fair to the project, it did get built, and it has been running continuously since 2017. Housed under the Central Board of Direct Taxes, it grew into a genuine data-warehousing and analytics platform, pulling in information from banks, mutual funds, property registries, credit card issuers, and the GST network to build what the department calls a "360-degree" profile of each taxpayer. Two dedicated centres — the Income Tax Transaction Analysis Centre and the Compliance Management Centralised Processing Centre — were stood up to process this data and generate automated compliance notices, largely without direct human contact between officials and taxpayers. Machine learning components were added over time to flag anomalies and risk-score cases for scrutiny. A public-facing Compliance Portal let taxpayers see and respond to discrepancies the system had flagged in their own filings.

By any measure of institutional survival, ten years is a long run for a government IT program. But survival isn't the same as success, and on the specific goals the project set for itself — widening the tax base, catching serious evasion, and doing so credibly and fairly — the record is considerably more mixed than the department's own promotional language suggests.

The Tax Base Barely Moved

The original justification for Project Insight was a startling statistic: as of 2016, only about 3.61% of India's population paid income tax. Nearly a decade later, independent analyses of government data put the number of people who actually pay tax (as opposed to merely filing a return) at roughly 2 crore individuals — about 1.5% of the population. Even accounting for differences in how these figures were calculated, the trajectory is not what a successful tax-base-widening program should look like. India's direct tax-to-GDP ratio has remained stuck in the range of 5–6%, among the lowest for a major economy, and the burden of funding the state continues to fall disproportionately on indirect taxes like GST — which weigh most heavily on people who were never Project Insight's intended targets in the first place.

If a ten-year, AI-driven surveillance and analytics program was supposed to move this needle in a meaningful way, it hasn't.

Detection Without Recovery

The department does point to real numbers on the enforcement side. In FY 2024–25 alone, officials say survey operations detected roughly ₹30,444 crore in undisclosed income. That sounds significant, and it is worth acknowledging as a genuine output of the broader analytics apparatus Project Insight helped build. But two things complicate the picture.

First, the number of survey operations that produced this detection actually fell compared to the previous two years — suggesting the system may be narrowing its enforcement activity even as it claims continued technological sophistication. Second, and more damningly, a 2024 report from the Comptroller and Auditor General found that the Income Tax Department's total outstanding tax demand had ballooned from roughly ₹10.44 lakh crore in 2016–17 (the year before Project Insight launched) to ₹19.35 lakh crore by 2021–22 — and that more than 97% of that outstanding amount was classified by the department itself as "difficult to recover." In other words, even where the system successfully identifies undeclared income or unpaid dues, the department's ability to actually collect on those findings has gotten worse, not better, over the Project Insight era. Detection without recovery is not a tax compliance program; it's an expensive way of generating numbers for press releases.

The Individual-Focus Problem Persists

My original critique centered on a structural flaw: Project Insight was built to track individual taxpayers through PAN-linked data, while the Panama Papers had shown that serious tax avoidance overwhelmingly runs through corporate structures, shell companies, and offshore vehicles — the exact kind of arrangement a firm like Mossack Fonseca specialized in building. Ten years on, that mismatch hasn't been resolved. Project Insight's architecture — bank transactions, property records, credit card spending, social media activity — remains oriented around catching the individual professional or small businessman who under-reports income relative to visible lifestyle, not the layered corporate and trust structures used for large-scale wealth concealment. Complex evasion involving shell companies and cross-border structures continues to be handled, when it's handled at all, through separate mechanisms like the Benami Transactions Act and international information-sharing — not through Insight's core data-mining engine.

One qualified improvement is worth noting: the Common Reporting Standard, the international automatic tax-information-exchange framework, has now been adopted by more than 100 jurisdictions, addressing the specific worry I raised in 2016 that FATCA-style cooperation would remain too limited to matter. That's a genuine, if partial, win — but it sits alongside a domestic architecture that still isn't well suited to catching the kind of sophisticated concealment the Panama Papers exposed in the first place.

A Quiet Admission

Perhaps the clearest signal of the project's limitations is what the department itself has done: it is retiring Project Insight. Internal CBDT communications from 2023 describe the "operations phase of the existing Project Insight" coming to an end by March 2025, with a replacement — Project Insight 2.0 — put out to tender in December 2024 for a new managed service provider to design and run essentially the same mission from scratch. Officials solicited feedback from field officers on the "deficiencies" of the existing system as an explicit part of that process. A program confident in its own success does not typically get scrapped and rebuilt after less than a decade; it gets expanded. The decision to start over, rather than iterate, is itself a form of institutional verdict on how the original Project Insight performed.

The Contractor Question

The original piece also raised a question about accountability: what oversight exists over the private technology contractor building a state surveillance system, especially when that contractor's own corporate structure and international operations could, in principle, be subject to the very scrutiny the system was meant to enable? L&T Infotech held the implementation contract for the original Project Insight for its full run. There's no public evidence that this arrangement was abused in the way I speculated it might be — but there's also little public evidence of independent audit or disclosure regarding the contractor relationship, the cost of the program over its decade of operation, or its actual cost-effectiveness relative to the revenue it helped recover. That opacity was a concern in 2016, and it remains one now that a second, presumably larger, procurement is underway for Insight 2.0.

The Verdict

Project Insight did not fail as a piece of software — it was built, deployed, and kept running for a decade, and it produced a genuine paper trail of flagged discrepancies and enforcement notices. But judged against the goals it was actually sold on — widening India's notoriously narrow tax base, catching the kind of sophisticated evasion the Panama Papers exposed, and doing so with recoverable results — the evidence a decade on is unflattering. The tax base has barely widened. The gap between detected and recoverable tax demand has grown dramatically. The program's core design still struggles with exactly the corporate and shell-company evasion it was partly built in response to. And the government's own decision to replace rather than extend the system is hard to read as anything other than a tacit acknowledgment that the original bet didn't pay off as promised.

Project Insight 2.0 may yet do better — it inherits a genuinely improved data ecosystem, a more mature international information-sharing regime, and presumably some lessons learned. But that next verdict will have to be earned. The first ten years of the "flagship" answer to India's tax evasion problem were, by the numbers the government itself has released, a disappointing return on a decade-long, technology-heavy bet.