What AI Revealed About My Money That I Couldn't See Myself

I've always been good with numbers. Or so I told myself.

I have dashboards. Spreadsheets. A structured view of assets, liabilities, and monthly cash flow. I track CPF, equity positions, credit lines, and school fee schedules. I know my numbers.

What I didn't know was what my numbers meant about me.

There's a difference between tracking data and understanding it. Between knowing what happened and understanding why. My financial tracking was excellent at the former. It was doing nothing useful for the latter.

The experiment

As part of my AI audit across all life domains, I decided to paste three months of transaction data into an AI and ask it a deceptively simple question: What patterns do you see that I might be missing? I was looking for new insights or perspectives, I had not previosuly considerde myself.

Not "categorise this." Not "give me a pie chart." Not "tell me where I'm overspending." Those are tool-level prompts. They get tool-level answers.

I wanted the thinking-partner version. I wanted it to look at the data the way a sharp, neutral, unafraid advisor would, someone with no incentive to protect my ego.

What it found

The AI identified three patterns I hadn't consciously registered.

First: my discretionary spending spiked consistently in periods of high travel or professional stress. Not random spikes. Patterned ones. The months I was keynoting or traveling for work most heavily were the months I was spending most unconsciously. I was rewarding the performance of high output with financial decisions that quietly undermined the stability I was working toward. In the moment, I could easily justify these.

Second: my investment discipline was excellent in isolation, but I had a habit of treating credit facilities as a secondary buffer rather than an emergency option. The line between leverage as a strategy and leverage as avoidance was blurrier than I'd admitted to myself.

Third: the data showed a clear emotional dimension to my spending that mapped almost directly onto school fee cycles. The months where education costs peaked were the months where I was most likely to make small, comfort-driven purchases, nothing dramatic individually, but cumulative in a way that mattered.

None of this was catastrophic. But none of it was visible to me when I was inside the data, because the data didn't tell me the story. It just showed me the events. Simply by making me aware of these, it completely change my conversation with my money.

The distinction between data and story

This is the gap I talk about with organisations all the time. Companies have more data than they've ever had. They can tell you what happened with extraordinary precision. What most organisations, and most people, struggle to do is contextualise it. To see the narrative in the numbers.

AI is genuinely good at this when you ask it the right questions. The problem is that asking the right questions requires you to be willing to hear the answers.

That's not an AI challenge. That's a human one. And it's the most important one.

After my confidence with AI as my thinking partner for finances grew, I started asking new questions. Like, what if I could optimise the timing of my core financial asset purchase, my CPF IA and my Long Term EFT. These were already performing well for me. But if a few strategic inputs, my models were able to auotmate a weekly analysis of my financial strategy, and suggest simple recommendations like; hold, buy, sell, use leverage. It was optimising the relationship I have with my bank, with insights I'd never considered. The result, a staggering 70% better return rate on the exact same investments, just timed better.

How I Use AI Agents to Run My Finances Today

I don't manage my money manually anymore. I've built a set of AI-powered finance agents that monitor, optimise, and report across every dimension of my financial life, from daily cash flow to a leveraged USD brokerage position.

At the core is what I call my VGT Navigator, a Python-based system that runs weekly portfolio optimisation on my Citi Brokerage account, where I hold units of VGT (Vanguard Information Technology ETF). It monitors buy zones, volatility signals, and position sizing to inform tactical decisions without me needing to watch the market daily. It's not algorithmic trading in the traditional sense, it's structured decision support, built around data-backed conviction rather than gut feel.

Running in parallel is my Volatility Trader, which scans for short-term tactical opportunities on my Saxo account. Both systems operate independently but feed into a unified reporting layer that emails me summaries on a scheduled cadence. No dashboards I need to log into. It comes to me.

Beneath the trading layer, I run a comprehensive financial data pipeline that consolidates statements across Banks, Investments, Reitrement Account and credit cards into a single structured data model, updated monthly. This gives me a clean, auditable view of net worth movement, cash flow patterns, debt trajectory, and investment performance across every account I hold.

The goal isn't automation for its own sake. It's clarity and capital efficiency. I want idle cash working, high-interest debt shrinking, and investment positions sized with intention, all without the cognitive load of managing it manually.

The system isn't perfect yet. But it's already changed how I make decisions, and that's the point.

Practical AI: Let AI read your spending

Try this. Export or copy the last 3 months of your bank or credit card transactions (strip out account numbers for privacy, keep categories and amounts). Paste it into your AI of choice and use this prompt:

"Here are my last 3 months of transactions. I don't want you to categorise them, I want you to look for behavioural patterns. What does this data suggest about how I relate to money emotionally? What patterns are here that I might be too close to see?"

Then sit with what comes back. Don't defend it. Don't dismiss it. Just ask: Is this true? Today, I have a significantly more advanced workflow. I built a local agent that captures all financial files, such as bank statements, purchase receipts, CPF statements, trading activity, and more, as PDFs (since almost no banks in Singapore provide raw data). It strips the data, normalises it, then enriches all those useless descriptions before storing it in a markdown file, something that, in the AI world, is 95% more efficient. From there, my financial models can work their magic.