Overview

AI Budget Prompt

A personal finance coaching GPT system. Upload a bank statement CSV, and the AI auto-detects the bank format, categorizes every transaction, compares spending to the 50/30/20 rule, finds hidden subscriptions, and gives actionable quick wins.

What It Does

Acts as a conversational, judgment-free financial coach (Dave Ramsey influenced). Takes raw bank data and transforms it into categorized spending analysis with personalized recommendations for improving financial health.

Project Files

  • budget-gpt-condensed.txt — System prompt (the AI's instructions)
  • BudgetGPT.pdf — Documentation / reference guide
  • sample-bank-statement.csv — Example CSV for testing

Core Flow

Upload CSVBank statement
Auto-DetectBank format
Categorize6 categories
Analyze50/30/20 rule
RecommendQuick wins

Supported Banks

Chase Bank of America Wells Fargo Generic CSV
Getting Started

How to Use

Simple 3-step process: export your bank statement as CSV, upload it to the GPT, and have a conversation about your finances.

Step-by-Step

  1. Export Your Bank Statement Log into your bank (Chase, BoA, or Wells Fargo), navigate to transaction history, and download as CSV. Most banks have this under "Download" or "Export" options.
  2. Upload to ChatGPT Open the BudgetGPT custom GPT (or paste the system prompt into a new conversation). Upload your CSV file. The AI auto-detects the bank format from column headers.
  3. Have a Conversation Ask questions like "Where am I overspending?", "How much do I spend on subscriptions?", "Can I afford to save more?" The AI responds conversationally with data-backed insights.
💡
Tone: The AI is designed to be conversational, judgment-free, and encouraging. Think "supportive friend who's good with money" — not a stern financial advisor. Dave Ramsey-influenced but without the lectures.
Processing

CSV Processing

The AI auto-detects which bank the CSV came from by analyzing column headers, then normalizes the data into a standard format for analysis.

Bank Format Detection

Auto-Detection Logic
Chase Detected by column headers: Transaction Date, Post Date, Description, Category, Type, Amount. Negative amounts = expenses.
Bank of America Detected by: Date, Description, Amount, Running Bal. Single amount column with positive/negative values.
Wells Fargo Detected by: Date, Amount, *, -, Description. Asterisk and dash columns are Wells Fargo-specific markers.

Transaction Categorization

6 Spending Categories
IncomePaychecks, direct deposits, transfers in, freelance payments
Fixed ExpensesRent/mortgage, car payment, insurance, loan payments — predictable, same amount each month
Variable ExpensesGroceries, gas, utilities — necessary but amount varies month to month
DiscretionaryDining out, entertainment, shopping, subscriptions — wants, not needs
Savings/InvestmentsTransfers to savings, investment contributions, 401k
Debt PaymentsCredit card payments, student loan payments, personal loan payments
⚠️
Categorization is AI-based. The system uses transaction descriptions to categorize spending. It handles most cases well (e.g., "NETFLIX" = Discretionary/Subscription) but may occasionally miscategorize unusual merchants. Users can correct it conversationally.
Analysis

Analysis Features

Beyond basic categorization, the AI performs deeper analysis: 50/30/20 comparison, subscription detection, debt payoff strategies, and emergency fund calculations.

50/30/20 Rule Comparison

Compares actual spending percentages against the 50/30/20 framework:

  • 50% Needs — Fixed + Variable expenses
  • 30% Wants — Discretionary spending
  • 20% Savings/Debt — Savings + Debt payments

Shows where you're over/under in each category with specific dollar amounts.

Subscription Detection

Scans for recurring charges and identifies hidden subscriptions:

  • Streaming services (Netflix, Hulu, Spotify)
  • Software subscriptions (Adobe, Microsoft)
  • Gym memberships
  • App store recurring charges
  • Forgotten trials that converted to paid

Debt Payoff Strategies

When debt payments are detected, the AI explains two payoff methods:

  • Snowball Method — Pay smallest balance first (psychological wins)
  • Avalanche Method — Pay highest interest rate first (mathematically optimal)

Recommends the best approach based on the user's specific situation.

Emergency Fund Calculator

Based on monthly expenses detected in the CSV:

  • Calculates 3-month and 6-month emergency fund targets
  • Compares against current savings (if detectable)
  • Suggests monthly savings amounts to reach the target
Output

Recommendations

The AI delivers personalized, actionable quick wins based on the user's actual spending data. No generic advice — everything is tied to specific transactions.

Quick Wins Format

Recommendations are delivered as specific, actionable items tied to real data. Examples:

  • "You spent $47/month on 3 streaming services. Canceling Hulu (used least) saves $15.99/month = $192/year"
  • "You ate out 14 times this month ($620). Cutting to 8 times saves about $265/month"
  • "Your gym membership ($49/month) hasn't been used — no associated nearby transactions. That's $588/year"

The AI Will

  • Be conversational and encouraging
  • Cite specific transactions and amounts
  • Give dollar amounts for potential savings
  • Explain trade-offs, not just commands
  • Celebrate good financial habits it finds
  • Adjust tone to be judgment-free

The AI Won't

  • Shame the user for spending habits
  • Give generic "spend less" advice
  • Make assumptions about income level
  • Recommend specific financial products
  • Provide legal or tax advice
  • Store or retain financial data
💡
System prompt design: The entire personality and analysis framework lives in budget-gpt-condensed.txt. This can be pasted into any ChatGPT conversation or used as a Custom GPT system prompt. No API calls, no backend — pure prompt engineering.
🎉
Walkthrough complete! AI Budget Prompt is a pure-prompt personal finance coach. Upload a CSV, get categorized spending analysis, 50/30/20 comparison, subscription detection, debt strategies, and personalized quick wins — all from a conversational AI.