Key Details
The World Bank Policy Research Report, Raising Revenue Right: A Roadmap for Domestic Resource Mobilization, uses India to illustrate both the potential and limitations of digital technology in tax administration.
Stage of tax administration | India-related evidence |
|---|---|
Identifying taxpayers | India has experience linking tax credentials with its national digital-identification system, helping expand and improve taxpayer registries. |
Making transactions visible | Research on the 2016 demonetisation found a substantial increase in cashless payments, accompanied by higher reported sales and tax payments. |
Detecting suspicious firms | A machine-learning model using Indian administrative data found that more than half of the firms it flagged did not exist at their registered addresses. |
Improving detection | The model performed better than conventional rule-based methods in identifying potentially bogus firms. |
Converting detection into revenue | A separate India example found that a highly accurate fraud-detection model did not raise revenue because it was insufficiently integrated into day-to-day administrative practices. |
Protecting taxpayers | The report calls for human oversight, explainable decisions, monitoring for bias and effective rights to challenge automated findings. |
Digital Systems Have Expanded What Tax Authorities Can See
Tax administrations perform three connected functions: identify taxpayers, verify their liabilities and collect the amount due. India’s digital identity, electronic payments, GST data and administrative databases have expanded the information available for the first two functions.
The report cites evidence that demonetisation accelerated cashless payments and was associated with increases in declared sales and tax payments. Greater use of recorded transactions can reduce the scope for concealing economic activity, especially when payment and transaction data can be checked against tax filings.
Yet digitalisation does not eliminate evasion. Taxpayers may change other elements of their declarations after one discrepancy becomes visible, while fraudulent entities can exploit weaknesses in registration, invoicing or input-tax-credit systems.
Machine Learning Can Find Cases That Static Rules Miss
Machine-learning models can analyse tax returns, invoices, registration records and transaction histories together, identifying combinations of risk that fixed rules may overlook.
An Indian model cited in the report flagged firms for physical verification and found that over half were absent from their registered addresses. Its performance exceeded traditional rule-based screening, showing that machine learning can help direct limited inspection resources towards higher-risk cases.
The finding concerns detection accuracy, not proof that every flagged firm committed fraud. Physical verification, documentary checks and applicable legal procedures remain necessary before liability or enforcement can be established.
The Revenue Gap Appears After Detection
The report’s more consequential India lesson comes from another fraud-detection application: a highly accurate model failed to increase revenue because it was not adequately integrated into routine administrative work.
An alert produces fiscal value only if it moves through an operational chain:
Risk flag → case assignment → verification → assessment → enforcement or recovery → appeal and closure
A breakdown at any stage can leave a sophisticated model generating lists without changing compliance. Officials may lack clear responsibility for acting on alerts; cases may not be prioritised by recoverable value; field verification can be delayed; or detected liabilities may remain tied up in disputes.
Automation Also Creates Accountability Risks
The World Bank cautions that machine-learning systems can reproduce biases in historical data, misclassify taxpayers or produce findings that officials and affected businesses cannot readily interpret.
Tax automation therefore requires more than technical accuracy. Human review, documented reasons, data protection, error monitoring and accessible appeal mechanisms are essential when an algorithm contributes to audits, denial of claims or other coercive action.
Policy Relevance
India’s digital tax performance should be judged across the complete enforcement chain, not by the number of databases connected or alerts generated.
What should be measured | Why it matters |
|---|---|
Flags verified as genuine risk | Separates useful detection from false positives. |
Cases completed within defined periods | Shows whether models are integrated into administrative workflows. |
Tax assessed and actually recovered | Distinguishes notional demand from realised revenue. |
Decisions sustained after appeal | Tests the evidentiary and legal quality of automated targeting. |
Compliance time and refund speed | Ensures technology also improves the taxpayer experience. |
Errors across taxpayer groups | Helps detect whether automated systems impose unequal burdens. |
The report’s India evidence suggests that the next gains may come less from adding another analytical tool and more from establishing clear case ownership, usable evidence, timely field action and feedback between enforcement outcomes and model design.
Follow the Full Report Here: Raising Revenue Right: A Roadmap for Domestic Resource Mobilization