Overview
Money Tracker uses advanced AI to automatically extract transaction information from emails and uploaded documents. The AI agent analyzes text content, images, and PDFs to identify financial transactions with high accuracy.AI Engine
Money Tracker uses the Grok-4-1-fast-non-reasoning model from X.AI, optimized for:- Fast processing speed
- High accuracy in financial data extraction
- Multi-language support
- Image and document understanding
What the AI Extracts
The AI identifies and extracts these fields from emails and documents:Transaction Type
Determines if money was received (income) or sent (expense)
Merchant
The company or person involved in the transaction
Amount
The transaction value (numeric)
Currency
3-letter ISO code (USD, EUR, ARS, etc.)
Date
When the transaction occurred
Category
Automatic categorization (food, transport, etc.)
Description
Additional context about the transaction
How Email Extraction Works
Email Reception
When a new email arrives in your INBOX, Money Tracker receives a webhook notification
Content Preparation
The system extracts:
- Email body text (plain text preferred, HTML as fallback)
- Email subject and sender
- Image attachments (JPG, PNG, GIF, BMP, TIFF, WebP)
- PDF attachments with text extraction
Context Building
The AI receives:
- Your full name (to determine income vs expense)
- All email text content
- OCR text from images
- Extracted text from PDFs
AI Analysis
The AI model analyzes all content sources to:
- Determine if a transaction exists
- Extract transaction details
- Return a confidence assessment
Income vs Expense Detection
The AI uses sophisticated logic to determine transaction direction:Income Detection
- Sender keywords: “deposit”, “payment received”, “transfer from”, “salary”, “refund”
- Context clues: Your name appears as recipient
- Banking language: “credited to your account”, “incoming transfer”
Expense Detection
- Sender keywords: “purchase”, “payment to”, “charge”, “debit”, “withdrawal”
- Context clues: Your name appears as sender/payer
- Banking language: “debited from your account”, “outgoing transfer”
Currency Detection
The AI identifies currency using:Explicit Currency Codes
Explicit Currency Codes
ISO 4217 codes like USD, EUR, ARS, GBP are recognized immediately
Currency Symbols
Currency Symbols
$ (USD), € (EUR), £ (GBP), etc. are detected and converted to ISO codes
Contextual Inference
Contextual Inference
If currency is ambiguous (e.g., just ”$”), the AI uses:
- Email sender location/domain
- Language of the email
- Country-specific formatting
Supported Currencies
Supported Currencies
- USD (US Dollar)
- EUR (Euro)
- GBP (British Pound)
- ARS (Argentine Peso)
- MXN (Mexican Peso)
- BRL (Brazilian Real)
- CLP (Chilean Peso)
- COP (Colombian Peso)
Category Classification
The AI automatically categorizes transactions into 11 categories:| Category | Examples |
|---|---|
| Salary | Payroll, wages, income deposits |
| Food | Restaurants, groceries, food delivery |
| Transport | Uber, gas, public transit, parking |
| Entertainment | Movies, concerts, subscriptions, games |
| Housing | Rent, mortgage, utilities, repairs |
| Health | Doctor visits, pharmacy, insurance |
| Education | Tuition, courses, books |
| Services | Subscriptions, memberships, professional services |
| Clothing | Apparel, shoes, accessories |
| Investment | Stocks, crypto, retirement contributions |
| Other | Anything that doesn’t fit above |
Document Upload Processing
When you upload a receipt or invoice:Image Processing
Image Analysis
The AI directly processes the image using vision capabilities:
- Reads text from the receipt/invoice
- Identifies structured data (amounts, dates, merchant names)
- Understands layout and context
PDF Processing
Multi-Language Support
The AI processes emails and documents in multiple languages:- English
- Spanish
- Portuguese
- And more…
The AI preserves the original language of merchant names and descriptions - it does not translate them
AI Confidence and Accuracy
Confidence Scoring
While individual transactions don’t show confidence scores, the AI internally evaluates:- Clarity of transaction information
- Presence of required fields (amount, merchant, date)
- Consistency of data
Quality Factors
Accuracy is highest when emails contain:- Clear merchant names
- Explicit amounts with currency
- Specific dates
- Transaction-related keywords
- Structured formatting (tables, invoices)
False Positives
The AI is designed to minimize false positives by:- Requiring explicit transaction indicators
- Skipping promotional emails
- Ignoring advertisements
- Filtering out email signatures and footers
Fallback Processing
If the AI service fails, Money Tracker uses keyword-based fallback:- Searches for transaction keywords: “purchase”, “payment”, “charge”, currency symbols
- If found, creates a basic transaction with:
- Merchant: Email sender
- Amount: 0 (requires manual update)
- Category: “uncategorized”
Email Discard Logic
Emails without transactions are “discarded” to improve efficiency:- Stored in a
discarded_emailstable - Won’t be re-processed during future email imports
- Prevents duplicate analysis
- Reduces AI processing costs
Discard Reasons
- “No transaction detected” - AI didn’t find financial data
- “Invalid transaction: amount is 0” - Detected transaction but amount was zero
- “AI processing failed” - Technical error during analysis
- “User discarded transaction” - You manually deleted the transaction
AI Processing Metrics
The Dashboard shows AI activity stats:Emails Processed by AI
Total number of emails analyzed
Transactions Found
Emails that contained valid transactions
Emails Skipped
Emails discarded (no transaction found)
Best Practices
- Keep your Gmail inbox organized: The AI only processes INBOX emails
- Don’t delete transaction emails: Let the AI process them first
- Provide your full name: Helps the AI determine income vs expense
- Upload clear documents: High-quality images and PDFs improve accuracy
- Review auto-extracted transactions: Verify category and merchant names
Related Features
- Gmail Integration - Email monitoring setup
- Transactions - Manage extracted transactions
- Subscriptions - AI-detected recurring transactions