Experiments AI Personal Finance

<aside> 🗒️

Experimented with Google Sheets’ built-in =AI() function to automatically categorize my personal finance transactions, intentionally constraining the LLM to my own defined category system.

</aside>

Background

Personal finance requires discipline. Every week, I sit down to review my bank transactions and categorize each line based on my designed category system. This lets me recognize spending patterns (ie. where my money goes, any seasonal trends, etc). It does feels tedious and is time consuming.

Over the years, I’ve been incrementally improving my personal finance tracker on Google Sheets, like refining formulas and dashboards to extract more information from my data. So this time I wondered, “why not let AI handle the tedious categorization work instead?”

I’ve tried some apps with built-in categorization, but noticed that:

• They can invent categories that do not match my spending philosophy / how I group things
• Can misclassify edge cases, especially ambiguous ones (like grocery stores that sells a bunch of different things)

Question to answer:

Can an AI meaningfully reduce manual effort while preserving some human control and trust?

Hypothesis

Yes, maybe! An LLM might be able to reliably map messy transaction descriptions to my pre-defined categories, provided that:

• Categories are constrained and explicit (not open ended)
• AI is used only for classification (not interpretation)

Process & Implementation

Initial Set up

1. Exported raw bank transactions to .csv