How AI Can Automatically Categorize Your Transactions

The Manual Categorization Problem

Categorizing transactions manually is slow, repetitive data entry. It's also error-prone:

• Unclear merchant names are hard to place
• Subscriptions get forgotten and left uncategorized
• Refunds and returns don't always get matched to the original purchase

How AI Categorization Works

1. Reading Transaction Descriptions

A language model reads transaction descriptions and uses context to guess a category:

• "UBER *TRIP" → Transportation (not "Uber Eats" = Food)
• "SPOTIFY" → Subscription
• "AMZN MKTP US*AB123C" → Shopping (Amazon marketplace code)

2. Recognizing Patterns

The model can pick up on:

• Recurring charges (subscriptions)
• Merchant code patterns (e.g., "SHOPEE*" = e-commerce)
• Payment processors vs. actual merchants

3. Using Context

Details such as the amount and how often a charge repeats can also help separate similar-looking transactions.

Which AI Does Bills AI Use?

Bills AI uses OpenAI (gpt-4o-mini) to extract and categorize transactions. There is no model selection. We have not measured categorization accuracy, so we do not publish an accuracy figure. AI can miscategorize transactions, especially unclear merchant names, transfers and payments.

Beyond Basic Categorization

Bills AI also tries to provide insights from your statement:

1. Subscription Detection

Recurring charges can be flagged for you to review.

2. Spending Anomalies

Unusual transactions can be highlighted so you can check them.

Getting Started with AI Categorization

1. Download one or more bank statements
2. Upload them to Bills AI
3. Review the categories against your original statement
4. Use the results to see where your money goes