AI and data analytics are changing the fight against tax fraud in India

A company issues invoices and transfers tax credits to another business, even though no genuine transaction may have taken place.  

The paperwork appears legitimate, but the underlying business activity is missing.  

As money and tax credits move through multiple entities, identifying where the fraud began and who ultimately benefited becomes more difficult. 

Speaking to GovInsider, Nitesh Kumar Singh, a state tax officer with the Department of State Taxes, Government of Uttar Pradesh, India, says the tax administration is becoming increasingly data-driven as businesses move their operations and records online. 

“Taxation is no longer just about compliance,” he says, pointing to the growing importance of analysing invoices, payments, tax returns and other digital records to identify irregularities. 

While digital records give authorities more information to examine, the volume and complexity of transactions also make it harder to distinguish genuine business activity from tax evasion attempts, he adds. 

From fake invoices to interconnected business networks

Authorities have traditionally relied on tax returns, audits and investigations to identify irregularities.

Digitisation has changed the scale and complexity of the challenge. 

Fraudulent transactions can be embedded within extensive networks of businesses, with individual entities appearing legitimate when examined in isolation. 

One of the problems tackled by Singh’s tax authority is the creation or misuse of shell companies to generate invoices without genuine transactions.  

“Shell companies represent a significant systemic threat as vehicles for money laundering, hawala channels (an informal system of money transfer based on trust), terror funding, and illegal cross-border activities,” he says.  

These invoices may be used to claim or transfer input tax credits (ITCs), allowing businesses to obtain tax benefits to which they are not entitled. 

Fraud begins even before a business starts operating.  

According to Singh, identity details such as Aadhaar and Permanent Account Number (PAN) can be misused during registration, potentially allowing a business to be established using the details of someone who is unaware of it. 

Another problem tackled by his tax authority is legitimate businesses generating fake invoices without the actual movement of goods.  

Detecting such cases requires authorities to look beyond individual documents.  

Investigators may need to establish whether the businesses exist in practice, whether transactions are genuine and how the resulting tax credits move between entities. 

The challenge of analysing real-time data 

Beyond data collection, Singh distinguishes between the role of static and dynamic data in identifying fraudulent activity. 

Static data includes relatively stable information, such as business registration details. Dynamic data is generated as transactions occur, including invoices, payments and information about the movement of goods. 

While static data is comparatively straightforward to manage, dynamic data presents a more complex analytical challenge.  

Transactions take place continuously, generating large volumes of information that may need to be compared across different systems. 

Artificial intelligence (AI) and analytics can help to quickly identify unusual patterns, but their effectiveness depends on the availability of relevant records and the ability to distinguish meaningful discrepancies from routine variations in business activity. 

For example, a mismatch between an invoice and an e-way bill—the document used for the movement of goods under the GST framework—may warrant further examination.  

Such discrepancies are leads for further investigation, but not proof of fraud.
 

Singh cautions that their significance must be assessed in context. 

Following the connections between companies 

A company may appear compliant when its tax returns are examined independently.  

However, its connections with other entities could reveal a different picture. Several businesses might share directors, contact numbers or other identifying details, while transactions and tax credits move between them. 

Network analysis helps authorities map these relationships and identify patterns that may not be apparent from individual records. 

Singh points to graph-based analytical methods as one way to examine connections between companies.  

This approach allows investigators to examine whether several businesses are connected, whether their transactions reflect genuine commercial activity and whether ITCs are moving through a chain of entities without a legitimate underlying supply. 

Establishing these connections is only the starting point. Investigators must still determine the nature of the relationships and verify whether the suspected irregularities amount to actual wrongdoing. 

When a mismatch is not enough 

Despite the promise of advanced analytics, Singh cautions against relying entirely on automated systems. 

A discrepancy between tax returns does not necessarily mean a business has committed fraud.  

Differences may arise from reporting errors, timing issues or misunderstandings about a transaction, he explains.  

The design of the analytical model is critical. Singh emphasises the need to establish appropriate parameters and logical relationships between data points before using AI to identify potential fraud. 

If the underlying rules are incomplete or incorrectly designed, the system may produce misleading results. 

Consider how ITC works under the Goods and Services Tax (GST) regime.  

Businesses can generally offset eligible tax paid on purchases against the tax payable on sales, subject to applicable conditions. This mechanism helps prevent cascading taxation across the supply chain. 

But if the credit is claimed without a legitimate underlying transaction, the mechanism can be exploited. 

Comparing information reported by businesses can help authorities identify inconsistencies.  

These records in India range from GSTR-1, which records details of outward supplies, while GSTR-3B is used to report tax liabilities and claim eligible ITCs. GSTR-2B provides an auto-drafted statement of ITC information that businesses use when assessing their claims. 

A mismatch or an unusually high ITCs' claim is a risk indicator, not conclusive evidence of wrongdoing. Eligibility must be assessed against the applicable rules and the circumstances of the transaction. 

The next step is verification: establishing whether the transaction occurred, whether the credit was legitimately claimed and whether further action is justified. 

The limits of automation 

The growing use of AI in tax enforcement raises a broader question about the relationship between automated detection and human judgement. 

AI can help process large volumes of data, recognise patterns and prioritise cases for examination.  

But it cannot, by itself, establish the full context of every transaction or determine that every suspicious pattern constitutes fraud. 

While sound modelling is essential, Singh says that tax authorities must define how information should be collected, linked and analysed, and ensure that the parameters reflect relevant compliance requirements. 

Poorly designed models may flag legitimate businesses, create unnecessary compliance burdens or divert investigators' attention away from more serious cases. 

Turning data into reliable evidence 

Effective tax enforcement now depends on the ability to connect information, identify patterns and investigate relationships across complex business networks. 

As India's economy becomes more digital, the task is not merely to collect more information.

It is to turn that information into reliable evidence—and use it to distinguish genuine business activity from attempts to exploit the tax system.