Something strange happened this tax season. Chartered accountants started getting panicked calls from clients who had already filed their returns. "I need to file an ITR-U," they'd say. "Urgently." 

The ITR-U is the updated return form which is basically a way to fix mistakes after you've already submitted your taxes. It's usually filed by people who forgot to include some income or made genuine errors. But this year, 40,000 taxpayers suddenly needed to file corrections, withdrawing ₹1,045 crore worth of deductions they had claimed. 

That's not forgetfulness. That's fear. 

The Income Tax Department has deployed something new. An AI system that could cross-reference every claim against multiple data sources in real-time. When someone claimed HRA deductions under Section 10(13A) but their bank statements showed no rent payments, the machine flagged it instantly. When donation claims under Section 80G didn't match actual bank transfers, the AI caught it. When loan interest deductions were claimed but no EMI payments existed, the system knew that. 

For years, tax filing had been a game of creative interpretation. People would claim deductions they weren't entitled to, knowing that manual audits were rare and took months. The odds of getting caught were low. The penalties, while severe on paper, felt distant and unlikely. 

The AI changed these odds overnight. 

Suddenly, every return is being scrutinized. Every claim is being verified against actual financial behavior. The system didn't care about your tax advisor's clever strategies or your careful documentation. It looked at what you actually did with your money, not what you said you did. 

The panic seemed swift and widespread. Tax advisors who had built businesses around "maximizing deductions" found themselves dealing with clients who were terrified of prosecution. The panic makes sense. The penalties for fraudulent claims are severe, going up to 200% of the tax owed, interest rates reaching 24% annually, and potential imprisonment for up to seven years under Section 276C. 

Here’s an interesting read by ICAI: https://ai.icai.org/articles_details.php?id=61 

But here's why this is particularly interesting to me. While we do not have many details on what parameters were being used for fraud detection by the IT department, the potential plug of existing fraud tools who have mastered behaviourial fraud detection is massive. This could potentially catch complex schemes involving multiple entities, shell companies, circular transactions, and more.  

AI can potentially call out the kind of sophisticated tax evasion schemes that would take human auditor months to uncover. Every transaction, every bank statement, every TDS record that would flow into the machine for analysis can be used to prevent fraud. But that’s for the future.  

Bringing our attention back to the story; what emerged was a clear picture: people lie on their tax returns. A lot. But their financial behavior doesn't lie. The AI could see through carefully constructed narratives and look at the underlying reality. 

Now, imagine if this same approach was applied to lending. 

Traditional lending is built on trust. Borrowers provide income statements, employment letters, bank statements. Lenders review these documents, run credit checks, and make decisions based largely on what borrowers tell them. But what if lenders stopped trusting what people said and started analyzing what they did? 

The technology exists. There are bank statement analyzers like BankConnect with fraud detection capabilities and many alike. The same AI plugged into BSAs can cross-reference tax claims against bank statements that could cross-reference loan applications against actual financial behavior. The same system that flags suspicious deductions could flag risky borrowers before they default. 

For personal lending, this could mean analyzing spending patterns instead of stated income. Tracking actual salary deposits instead of trusting employment letters. Cross-referencing bank statements against tax filings and transaction histories to build a complete picture of financial reality. 

For corporate lending, the implications are even more interesting. The tax AI can trace complex relationships between entities, track money flows across organizations, and detect suspicious patterns.  

The shift from stated preferences to revealed preferences could transform financial inclusion. Traditional credit scoring excludes people with limited credit history, but AI can analyze alternative information sources to assess creditworthiness.  

So, the IT department's success this tax filing season is a pretty decent demonstration of what's possible when you build the right systems and deploy AI effectively.  

Here’s a callout for the lenders –  

The technology exists to move beyond traditional credit assessment to real-time behavioural analysis.  

The infrastructure can be built to handle millions of applications simultaneously.  

The insights are there for those willing to look beyond what borrowers claim, and to what they do. 

The tax department has made the move, when will you? 

 Until next time.    Cheers,  Srijan Nagar