AI-powered tax governance in India and its challenges

UPSC Prelims + Mains Study Note

1. At a Glance

• India's tax-to-GDP ratio averaged 16.36% during 2001–22, the lowest among all emerging and developing economies — making tax mobilisation a structural policy imperative. [4]
• India loses approximately 4.3% of tax revenues annually to tax evasion, making enforcement efficiency a fiscal priority. [4]
• The Income Tax Department's (ITD) Project Insight (PI) — launched 2017, operational 2019 — is India's flagship AI/data-analytics platform for tax administration; Project Insight 2.0 is the current upgrade cycle (tender floated December 2024). [1][3]
• This topic sits at the intersection of GS-II (governance, e-governance) and GS-III (Indian economy, resource mobilisation) — high-yield for Mains essays and case-study questions.

2. Why in the News

• India AI Impact Summit (February 2026): Global leaders and technology executives highlighted India's AI deployment in solving real-world problems, including tax governance. [4]
• Project Insight 2.0 tender (December 2024): Directorate of Income Tax (Systems) floated a new tender for Designing, Development, Implementation, Operations & Maintenance of Insight 2.0 — signalling a major scaling-up. [3]
• Income Tax Day 2025 (July 23, 2025): CBDT's commemorative document titled "A Journey of Digital Transformation" formally catalogued AI milestones in tax administration. [7]
• Gross Direct Tax Collections FY 2024–25: Provisional figure of ₹27.02 lakh crore (as on March 31, 2025), demonstrating the fiscal stakes of improved compliance infrastructure. [5]

3. Background & Evolution

Predecessors: Annual Information Return (AIR) system (pre-2017); Manual risk-based scrutiny; Statement of Financial Transactions (SFT) filing framework.

4. Core Static Facts

Project Insight (PI) — Key Parameters

Key Objectives of PI:

• Encourage voluntary tax compliance
• Reduce high-risk cases of potential tax evasion
• Make enforcement fairer and equitable
• Reduce officer discretion / prejudice in enforcement

CBDT AI/ML Tools Used: [2]

• Data matching
• Network analysis
• Pattern recognition
• Predictive analytics
• Text mining
• Forecasting and policy studies

Three Components of Project Insight: [4]

1. INTRAC — AI-powered analytical engine; 360-degree taxpayer profiling
2. Compliance Management module — tracks and nudges non-filers/under-reporters
3. Investigation Support — risk-scoring for scrutiny selection

Key Numbers:

• India's tax-GDP ratio (avg. 2001–22): 16.36% — lowest among peer economies [4]
• Annual tax evasion loss: ~4.3% of tax revenues [4]
• Direct tax-GDP ratio: 5.62% (FY14) → 6.11% (FY23) [6]
• Gross Direct Tax Collections FY 2024–25: ₹27.02 lakh crore (provisional) [5]

5. Multi-Dimensional Analysis

Economic

• India's persistently low tax-GDP ratio constrains public expenditure on welfare and infrastructure; AI-driven compliance widening is a critical lever. [4]
• Direct tax collections reaching ₹27.02 lakh crore (FY25) represent a multi-year high, partly attributed to AI-enabled scrutiny and nudge mechanisms. [5]
• Improved voluntary compliance reduces litigation and administrative cost per rupee of revenue collected.
• Cross-matching of GSTN and ITD data plugs revenue leakage from GST-income mismatch — a structural reform with compounding fiscal impact. [2]

Legal / Constitutional

• Faceless Assessment Scheme (Section 144B, Income-tax Act, 1961): Removes jurisdictional officer from scrutiny; cases allocated randomly across the country, minimising human bias and corruption. [5]
• CBDT's e-portal for tax evasion complaints provides a statutory-backed channel for citizen reporting. [8]
• Income Tax Act, 2025 (new Direct Tax Code under development): CBDT has sought stakeholder inputs on proposed rules, which will shape how AI outputs are used in assessment. [9]
• Right to privacy (Puttaswamy judgment, 2017): Aggregation of 360-degree taxpayer profiles raises proportionality concerns under Article 21 — not yet adjudicated in the tax-AI context.

Scientific / Technological

• INTRAC aggregates data from multiple sources to build 360-degree taxpayer profiles — among the most comprehensive government-run data-fusion exercises in India. [4]
• AI capabilities deployed: Pattern recognition, network analysis (detecting shell-company chains), predictive risk scoring, text mining (social media / property registrations). [2]
• Project Insight 2.0 will upgrade the architecture for scalability, real-time processing, and enhanced ML model governance. [3]
• Integration with Account Aggregator (AA) framework and DigiLocker is a prospective enhancement that would deepen data richness.

Ethical / Governance

• Algorithmic bias risk: AI models trained on historical enforcement data may embed past biases (e.g., sector-specific or geography-specific prejudice) and amplify them at scale.
• Lack of explainability (black-box problem): Taxpayers risk-flagged by AI have limited recourse if the model's reasoning is opaque — violating principles of natural justice (audi alteram partem).
• Data privacy and consent: India's Digital Personal Data Protection Act, 2023 raises questions about the lawful basis for aggregating third-party financial data without explicit taxpayer consent.
• Faceless assessment reduces corruption but introduces new risks — system gaming, coordinated misrepresentation across data sources.

Administrative

• Capacity gap: Tax officers need retraining to interpret AI risk scores; over-reliance on algorithmic outputs without human judgment can produce false positives.
• Federal coordination challenge: State-level stamp duty / registration data, municipal property records, and cooperative bank data are not uniformly digitised, creating data gaps that undermine the 360-degree profile.
• Litigation overhang: Aggressive AI-flagged scrutiny, if not well-calibrated, increases assessments-to-appeals ratio, loading ITAT and High Courts.
• Insider data access risk: Centralised taxpayer data held by INTRAC is a high-value target for insider misuse.

6. Recent Developments (Last 12–18 Months)

• December 2024: Directorate of Income Tax (Systems) floated tender for selection of Managed Service Provider (MSP) for Project Insight 2.0 — covering design, development, implementation, and operations. [3]
• FY 2024–25 (March 2025): Gross Direct Tax Collections touched provisional figure of ₹27.02 lakh crore, a record, reflecting improved AI-driven compliance infrastructure. [5]
• July 23, 2025 (Income Tax Day): CBDT released special document "Income Tax Day 2025: A Journey of Digital Transformation" summarising AI milestones in tax administration. [7]
• February 2026: India AI Impact Summit — global tech leaders cited India's tax AI deployment as a model for emerging economies. [4]
• 2025–26: CBDT sought stakeholder inputs on rules and forms under the proposed new Income-tax Act, 2025 (Direct Tax Code revision) — the legislative framework that will govern AI use in assessments. [9]

7. Prelims Hooks

1. Project Insight was launched by the Income Tax Department in 2017 and became fully operational in 2019. [4]
2. The technology implementation contract for Project Insight was signed with L&T Infotech Ltd. [1]
3. The analytical engine within Project Insight is called INTRAC (Income Tax Transaction Analysis Centre). [4]
4. India's average tax-GDP ratio during 2001–22 was 16.36% — the lowest among emerging and developing economies. [4]
5. India loses approximately 4.3% of tax revenues annually due to tax evasion. [4]
6. Faceless Assessment in income tax is governed by Section 144B of the Income-tax Act, 1961. [5]
7. India's direct tax-to-GDP ratio improved from 5.62% (FY14) to 6.11% (FY23). [6]
8. Gross Direct Tax Collections for FY 2024–25 stood at a provisional ₹27.02 lakh crore. [5]
9. Project Insight builds a 360-degree profile of each taxpayer by integrating internal ITD data with external sources including GSTN. [2]
10. CBDT's AI/ML tools include data matching, network analysis, pattern recognition, predictive analytics, text mining, and forecasting. [2]
11. Project Insight 2.0 tender was floated in December 2024 by the Directorate of Income Tax (Systems). [3]
12. The India AI Impact Summit where India's tax-AI progress was highlighted was held in February 2026. [4]
13. CBDT (Central Board of Direct Taxes) — not MCA or MEITY — is the nodal body overseeing AI deployment in direct tax administration. [2]
14. Project Insight aims to reduce officer discretion and prejudice in tax enforcement — not just evasion detection. [4]

8. Mains Relevance

GS Paper Mapping:

Plausible Mains Question Stems:

1. "Project Insight represents a paradigm shift in India's tax administration, but algorithmic governance without legal safeguards risks violating principles of natural justice." Critically examine. (GS-II/IV)

2. "India's persistently low tax-GDP ratio reflects structural weaknesses that technology alone cannot cure. Evaluate the potential and limitations of AI in tax governance." (GS-III)

3. "The Faceless Assessment Scheme and AI-driven risk profiling can reduce corruption but may create new asymmetries between the state and taxpayers. Analyse." (GS-II)

9. Related Topics to Study Next

10. Common Errors / Trap Areas

1. Wrong ministry: Project Insight is under Ministry of Finance / CBDT — NOT under MeitY or NITI Aayog (though AI policy broadly involves them).
2. Launch vs. operational date: PI was launched in 2017 but became fully operational in 2019 — examiners may test both; do not conflate.
3. Confusing Project Insight with TRACES/AIS: TRACES (TDS reconciliation) and the Annual Information Statement (AIS) are separate ITD portals; INTRAC is the back-end analytics engine of Project Insight — distinct from the taxpayer-facing portals.
4. Tax-GDP ratio confusion: The article cites 16.36% (total, 2001–22); the direct-tax-to-GDP alone is ~6% — aspirants often mix these two figures or confuse India's ratio with the OECD average (~34%).
5. Faceless ≠ Paperless: Faceless Assessment removes jurisdictional human interface but the process still involves document submission; confusing "faceless" with fully automated AI decision-making is a conceptual error examined in GS-IV ethics questions.