The first thing the AI told me was something my mother had been saying for years, but it was supported by better statistics. I was overspending on food that I didn’t prepare. When it was returned, the total was ₹14,200 over a period of thirty days. Zomato, Swiggy, and the little cash-only bakery close to my workplace that managed to appear in my UPI history. I thought it might be six or seven thousand. When you see an unexpected number sitting at a spotless little table without arguing with you, there’s a certain kind of silence that descends.
Mostly out of annoyance, I began the experiment on a Sunday night. On the first, my salary arrived, and by the twenty-third, the account appeared exhausted. I’m just exhausted, not empty. So I put three months’ worth of statements in a folder, showed it to Claude, and asked the obvious question. Where is the money being spent? The response was well-organized, courteous, and nearly surgical. Additionally, a magazine subscription that I last read in 2023 was flagged.
| Headngs | Details |
|---|---|
| Experiment Name | The 30-Day AI Money Audit |
| Duration | 1 March 2026 – 30 March 2026 |
| Tools Used | Claude, ChatGPT, Microsoft Copilot |
| Monthly Income Range | ₹1,00,000 – ₹1,20,000 (post-tax) |
| Accounts Reviewed | 2 bank accounts, 1 brokerage, 2 credit cards, 1 UPI wallet |
| Total Transactions Logged | 412 |
| Hidden Subscriptions Found | 9 |
| Money Reallocated or Saved | ₹47,300 |
| Time Spent Per Week | Roughly 90 minutes |
| Reference Reading | OECD report on household financial literacy |
| Verdict | Useful, occasionally uncomfortable, not a replacement for judgement |
I wasn’t surprised by the categories. Categories are visible to anyone with a banking app. Beneath, it was the patterns. Although it seems obvious in retrospect, the AI observed that I spent more time on weekends after difficult workdays, something I had never noticed before. It observed that although my investment SIPs were technically active, lifestyle creep was subtly canceling them out during those same weeks. It observed that I was paying for two cloud storage plans over the course of two emails with a sort of indifferent precision. little things. It’s the kind of thing that a human accountant might overlook because the numbers are too tiny.
I started doing daily check-ins by the second week. Not very dramatic. I would enter the day’s transactions into ChatGPT, request a one-line synopsis, and flag anything that seemed out of the ordinary. Speaking with a chatbot about a ₹380 taxi ride was peculiar. It’s even stranger that the chatbot began indirectly influencing my behavior. During the first week, I cancelled two subscriptions. The AI informed me that rivals in my pin code were providing the same speed for less, so by the third, I had renegotiated my internet plan. The email was even written by it.
There were boundaries, and they were evident. The AI was unable to advise me on whether to hold a specific stock that had been stagnant for several months. It hedged, recommended that I talk to an advisor registered with SEBI, and provided me with a list of questions to think about instead. In a sense, honest. I would have benefited from more than a certain response. It was almost comforting to see it refuse to play oracle. The idea that the tools function best when they cease pretending to be intelligent is gradually gaining traction throughout this entire AI era.
I hadn’t changed by the end of the month. On a Friday, I still placed an order for biryani. However, I was aware of the true cost. No single revelation led to the discovery of the ₹47,300 that was reallocated. It was the result of forty minor adjustments, each based on something the AI saw before I did. It remains to be seen if that persists. Habits change over time. Silently, salaries rise. I might be wondering where it all went by July. But the spreadsheet tells me, at least for the time being. And I almost believe it.