expense-categorization

Expense Categorization

Automatically classify business expenses into accounting categories, assign department cost centers, and flag tax-deductible items from raw transaction data. This skill processes credit card statements, bank feeds, and expense reports to produce clean, categorized output suitable for bookkeeping, tax preparation, and spend analytics.

Workflow

1. Receive Expense Data

Accept transaction data as CSV, bank statement text, or structured records. Required fields: date, amount, and description or merchant name. Optional fields: card last four digits, employee name, department, receipt notes. Normalize date formats and currency to a consistent standard.

1. Parse Description and Merchant

Extract the merchant name from the transaction descriptor, stripping out authorization codes, location suffixes, and card network prefixes. Map common merchant name variations to canonical names (e.g., "AMZN MKTP US" → "Amazon", "GOOGLE *GSUITE" → "Google Workspace"). Use the merchant category code (MCC) when available as a secondary signal.

1. Classify Expense Category

Assign each transaction to a primary category based on merchant identity, MCC, description keywords, and amount patterns. Standard categories: Travel & Lodging, Meals & Entertainment, Software & SaaS, Office Supplies, Professional Services, Advertising, Utilities, Insurance, Shipping & Postage, Equipment, Training & Education, Miscellaneous.

1. Assign Department and Cost Center

Route each expense to the appropriate department based on the cardholder, project codes in the description, or pre-configured rules. Apply default department assignments for known merchants (e.g., AWS charges → Engineering, HubSpot → Marketing).

1. Flag Tax-Deductible Items

Mark expenses that qualify for tax deduction based on category and business purpose. Apply IRS rules for meals (50% deductible for business meals, 100% for company events), home office, vehicle mileage, and professional development. Flag items that need substantiation — receipts, business purpose memo, or attendee lists.

1. Generate Categorized Summary

Output a categorized transaction list with totals by category, department, and tax status. Highlight any transactions that couldn't be confidently categorized (confidence < 80%) for manual review.

Usage

Provide transaction data in any tabular format. Specify any custom category mappings, department rules, or tax jurisdiction if not US federal.

Example prompt:

Categorize these November credit card transactions for our 15-person startup. We have Engineering, Marketing, and Operations departments. Flag tax-deductible items. Here's the CSV: [paste transactions]

Examples

Example 1: Monthly Credit Card Categorization

Input (CSV):

date,amount,description,cardholder
2024-11-02,249.00,GITHUB TEAM PLAN,Sarah Chen
2024-11-03,1450.00,UNITED AIRLINES 0167823,Mike Torres
2024-11-03,189.50,MARRIOTT HOTEL CHI,Mike Torres
2024-11-05,42.67,UBER TRIP FKJE83,Mike Torres
2024-11-08,156.00,GOOGLE ADS 8834721,Lisa Park
2024-11-12,89.99,ZOOM VIDEO COMMS,Sarah Chen
2024-11-14,67.32,DOORDASH DASHER TIP,Operations
2024-11-18,4200.00,DELL TECHNOLOGIES,Sarah Chen
2024-11-22,350.00,COURSERA BUSINESS,Mike Torres
2024-11-25,23.45,STAPLES #1284,Operations

Output:

Summary:

Example 2: Flagging Misclassified Expenses

Input: Review existing categorizations for accuracy.

Output — Correction Report:

Impact: Reclassifying these 4 transactions shifts $1,179.00 across categories, affecting department budgets and tax deduction calculations. The Adobe correction reduces the Training & Education deduction by $599 and increases the Software & SaaS deduction by the same amount.

Best Practices

• Build and maintain a merchant-to-category mapping dictionary. Start with the 50 most common merchants in your transaction history and expand over time.
• Set a confidence threshold (recommended: 80%) below which transactions are routed for human review rather than auto-categorized.
• Review categorization accuracy monthly — track precision and recall by category to identify systematic errors.
• Always separate personal and business expenses before categorization. Flag transactions from personal merchants (grocery stores, streaming services) for review.
• Apply consistent rules for edge cases like meals during travel (Travel vs. Meals) and document the policy.
• Keep tax deductibility rules current with the applicable tax year — IRS rules change frequently for categories like meals, entertainment, and vehicle expenses.

Safety Boundaries

• Treat the output as analytical support, not individualized financial, tax, investment, or accounting advice.
• Preserve source data and expose assumptions, formulas, units, and reconciliation checks so a reviewer can reproduce the result.
• Do not initiate payments, transactions, journal entries, filings, or account changes without explicit user authorization.
• Require a qualified professional to review material decisions, regulated filings, or conclusions based on incomplete data.

Edge Cases

• Split transactions: A single purchase at a warehouse store may include both office supplies and snacks. If the receipt is available, split into separate line items with distinct categories.
• Foreign currency transactions: Categorize based on the merchant and purpose, not the currency. Record both the original currency amount and the converted amount. Watch for duplicate entries from currency conversion fees.
• Refunds and chargebacks: Match refunds to the original transaction and apply the same category as a negative entry. Don't create a new "refund" category — it distorts spend analytics.
• Recurring vs. one-time: Identify recurring charges (same merchant, similar amount, monthly cadence) and flag any that stop unexpectedly or change amount by more than 10%.
• Ambiguous merchants: When a merchant name maps to multiple possible categories (e.g., Amazon could be office supplies, software, or equipment), use the amount and cardholder's department as tiebreakers. Amounts under $100 from Amazon default to Office Supplies; over $500 default to Equipment.