AI-Powered Taxation — Project Insight's Gains and Governance Risks

🗞️ Why in News India’s Project Insight — the AI-driven tax analytics platform launched in 2017 by the Income Tax Department — has recovered over ₹11,000 crore in additional revenue since FY2020-21 through the NUDGE strategy of voluntary compliance prompts. The Hindu editorial analyses the platform’s achievements alongside growing concerns about algorithmic bias, data quality failures, privacy rights, and the absence of independent oversight mechanisms.

What Is Project Insight?

Project Insight is the Income Tax Department’s Big Data Analytics and AI platform for tax compliance and evasion detection:

• Launched: 2017 (by the Central Board of Direct Taxes — CBDT)
• Developed by: Infosys BPM (contracted by MeitY/CBDT)
• Core function: Creates comprehensive 360° taxpayer profiles by aggregating data from multiple sources and using machine learning to flag discrepancies

Data sources integrated into Project Insight:

The NUDGE strategy: Rather than immediately triggering scrutiny notices, Project Insight sends non-intrusive digital nudges — personalised compliance reminders via SMS, email, and the income tax e-filing portal — prompting taxpayers to self-correct discrepancies before formal proceedings. This voluntary compliance model is credited with generating ₹11,000 crore+ in additional revenue since FY2020-21 with minimal litigation.

The Achievements — What Project Insight Gets Right

1. Non-Intrusive Compliance Architecture

Traditional tax administration relied on random scrutiny selection or informant-based detection. Project Insight replaces this with risk-scoring — every taxpayer gets an AI-generated risk score based on income-expenditure mismatches. High-risk returns are flagged for scrutiny; low-risk returns pass through with no human intervention. This:

• Reduces harassment of legitimate taxpayers
• Focuses investigative resources on genuine high-risk cases
• Makes compliance predictable and rule-based

2. CASS (Computer-Aided Scrutiny Selection)

Since 2015, all scrutiny selections are made by CASS — an algorithm, not a tax officer. This eliminates the discretionary power that led to rampant inspector corruption in the pre-IT era. Project Insight feeds risk signals into CASS.

3. AIS (Annual Information Statement)

The Annual Information Statement (AIS) — an outgrowth of Project Insight’s data aggregation — shows taxpayers all information the tax department holds about them. This transparency enables self-correction before filing.

4. Revenue Impact

• ₹11,000 crore+ in voluntary compliance additions (FY2020-21 onwards)
• India’s tax-to-GDP ratio has improved: ~11% (FY24-25) vs ~10% in pre-GST era
• Direct tax-to-GDP: ~6.4% — still low vs OECD average ~15%

The Problems — What the Editorial Flags

1. Data Quality Vulnerabilities

Project Insight aggregates data from multiple sources — but those sources have quality problems:

• Property registration values in India are often under-reported (circle rate vs. market rate gap)
• Banking records may include inherited or gifted amounts that appear as unexplained credits
• Third-party reporting errors: If a bank or employer reports incorrect data to the tax department, the taxpayer gets a mismatch notice they must manually contest
• Result: False positives — legitimate taxpayers receive notices for apparent discrepancies that are actually data entry errors by third parties

2. Algorithmic Bias Risk

Machine learning models trained on historical tax data may encode historical biases:

• Models trained on past audit data (which skewed toward scrutiny of small traders, proprietorships, and salaried employees) may systematically under-flag complex structures (HNI trusts, offshore structures, multi-tier corporate arrangements)
• Survivorship bias: The model learns from detected evasion cases — but the most sophisticated evasion structures evade detection entirely and never enter the training data
• Certain communities or geographies that were historically over-policed may receive disproportionate AI-flagging — a form of proxy discrimination

3. Lack of Transparency — The Black Box Problem

Citizens receive notices based on AI risk-scoring but are not told:

• What data triggered the flag
• What algorithm was used
• What the risk score threshold is
• Why their case was selected vs. a similar case that was not

This opacity violates the principle of natural justice — the right to know the case against you — which is a component of Article 21 (Right to Life including right to dignity and fair procedure).

4. No Independent Oversight

There is no:

• Algorithmic audit of Project Insight by an independent body
• AI Ombudsperson for taxpayers who believe the AI system treated them unfairly
• Statutory right to algorithmic explanation under Indian law (unlike the EU’s GDPR Article 22 right against automated decision-making)

The CBDT/Income Tax Department is both the operator and the regulator of Project Insight — a structural conflict of interest.

5. Privacy and Surveillance Concerns

Project Insight’s 360° profile aggregates data across every major life domain (banking, property, investments, business) without requiring a court order or independent authorisation. This is:

• Substantively more intrusive than a conventional tax scrutiny
• Operating under Section 133C of the Income Tax Act (power to collect information) — not subject to judicial oversight before deployment
• In tension with the Privacy Judgment (K.S. Puttaswamy v. Union of India, 2017) — 9-judge bench held that informational privacy is a fundamental right under Article 21; state data collection must be proportionate, necessary, and legally grounded

The Editorial’s Recommendations

The Hindu calls for a governance framework for AI in taxation:

1. Algorithmic Transparency: Publish the decision criteria (not the model weights) used for scrutiny selection; give taxpayers a summary of why they were flagged

2. Independent Algorithmic Audit: Commission an annual audit of CASS and Project Insight by an independent body (CAG or a statutory AI regulator)

3. AI Ombudsperson: Create a dedicated redressal mechanism for taxpayers who contest AI-generated notices — separate from the existing faceless appeals system

4. Data Quality Accountability: Mandatory 30-day correction window for third-party data errors before AI-generated notices are issued

5. Proportionality Review: Ensure data aggregation is proportionate under the Puttaswamy proportionality test — not all data sources need to be integrated for all taxpayers

The Broader AI Governance Gap in India

India does not yet have a comprehensive AI regulation law. The Digital India Act (DIA) — the proposed successor to the IT Act, 2000 — has been in consultation since 2022 but not yet enacted. Key elements still missing:

• No statutory right to explanation for algorithmic decisions affecting citizens
• No AI-specific data protection provisions (the DPDP Act 2023 covers personal data broadly but not algorithmic accountability)
• No mandatory impact assessments for high-risk AI systems in government

Global contrast:

• EU AI Act (2024): Categorises AI systems by risk; high-risk systems (including those used in tax/welfare determination) require mandatory human oversight, transparency, and audit
• GDPR Article 22: Gives EU citizens the right not to be subject to solely automated decisions with significant effects
• US Executive Order on AI (2023): Mandates AI safety evaluations for government procurement

UPSC Relevance

Prelims: Project Insight (launched 2017, CBDT), CASS (Computer-Aided Scrutiny Selection), AIS (Annual Information Statement), Section 133C of Income Tax Act, K.S. Puttaswamy judgment (2017 — informational privacy under Article 21), NUDGE strategy, ₹11,000 crore additional revenue (FY2020-21 onwards), India tax-to-GDP ratio (~11%), EU AI Act (2024).
Mains GS2: AI in governance — Project Insight, algorithmic accountability, transparency in public administration, natural justice principles, right to privacy (Puttaswamy), CBDT, faceless assessment scheme, digital governance. GS3: Technology — AI risks and regulation, big data in taxation, India’s tax administration reforms, direct tax-to-GDP, Digital India Act.

📌 Facts Corner — Knowledgepedia

Project Insight — Core Facts:

• Launched: 2017 (Income Tax Department, CBDT)
• Developer: Infosys BPM (contracted by MeitY/CBDT)
• Function: AI-driven tax analytics — 360° taxpayer profiling, risk scoring
• NUDGE strategy: Voluntary compliance prompts (SMS, email, e-filing portal)
• Revenue recovered: ₹11,000 crore+ (FY2020-21 onwards)
• Legal basis: Section 133C, Income Tax Act (power to collect information)

Data Sources Integrated:

• PAN-Aadhaar, banking (Form 26AS/AIS), GST filings, property registrations, stock market transactions, foreign remittances, credit card spends, TDS returns

India Tax Statistics:

• Tax-to-GDP ratio: ~11% (FY24-25)
• Direct tax-to-GDP: ~6.4%
• OECD average tax-to-GDP: ~15%
• India’s income tax base: ~75 million returns filed (FY24-25)

Key Related Schemes:

• Faceless Assessment Scheme (2020): Random allocation of cases to officers; no face-to-face interaction
• CASS (Computer-Aided Scrutiny Selection): All scrutiny cases selected by algorithm since 2015
• AIS (Annual Information Statement): Shows taxpayers all data held about them; launched 2021

Privacy Framework:

• K.S. Puttaswamy v. Union of India (2017): 9-judge SC bench — right to privacy is fundamental under Article 21; data collection must satisfy: (1) legality, (2) legitimate aim, (3) proportionality, (4) procedural guarantees
• DPDP Act 2023 (Digital Personal Data Protection): Governs personal data; limited algorithmic accountability provisions
• Digital India Act: Proposed successor to IT Act 2000 — in consultation since 2022; not yet enacted

Global AI Governance Comparison:

• EU AI Act (2024): Risk-based AI regulation; high-risk AI (incl. government tax decisions) = mandatory human oversight + audit
• GDPR Article 22: Right not to be subject to solely automated significant decisions (EU citizens)
• US Executive Order on AI (October 2023): AI safety standards for government use

Other Relevant Facts:

• CBDT: Central Board of Direct Taxes — statutory body under MoF; administers Income Tax Act 1961
• MeitY: Ministry of Electronics and Information Technology — nodal for digital governance, AI policy
• Aadhaar-PAN linkage: Mandatory for all taxpayers (deadline extended multiple times; now enforced)
• Project Insight vs. GSTN: Both are big-data tax systems — GSTN for indirect tax, Project Insight for direct tax — integration is a next-phase plan

Sources: The Hindu, CBDT, PIB