Skip to content

Bank Transaction Categoriser

Example prompt: "Each day, read new transactions from our bank export sheet. Categorise each one using our chart of accounts. Flag anything unusual or over 500 pounds and send me a summary on Slack."

How to automate bank transaction categorisation with GloriaMundo

The Problem

Bookkeepers and small business owners spend a disproportionate amount of time categorising bank transactions. Every day brings a new batch of debits and credits — software subscriptions, supplier payments, client refunds, card charges — and each one needs to be assigned to the correct account code. Most transactions follow predictable patterns (the same vendor, the same category, month after month), yet someone still has to review and tag them manually. When this work piles up, month-end reconciliation becomes a scramble through weeks of uncategorised entries.

How GloriaMundo Solves It

We build a daily workflow that reads new transactions from your bank export in Google Sheets. An LLM step examines each transaction description, amount, and payee, then assigns a category from your chart of accounts (which you maintain in a separate sheet or tab). For common recurring transactions — your hosting bill, your accountancy retainer — the categorisation is near-instant. For anything the model is unsure about, or any transaction exceeding your review threshold, the workflow flags it for manual review. A code step compiles the results into a daily summary: how many transactions were auto-categorised, how many need attention, and any anomalies. The summary is posted to Slack so you can review exceptions without opening the spreadsheet. Glass Box preview lets you see every proposed category before it is written back to the sheet.

Example Workflow Steps

  1. Trigger (scheduled): Runs daily at 9:00 AM.
  2. Step 1 (integration): Read new uncategorised transactions from the bank export sheet in Google Sheets.
  3. Step 2 (integration): Read the chart of accounts and any vendor-to-category mapping rules from a reference sheet.
  4. Step 3 (llm): For each transaction, analyse the description and amount against the chart of accounts and assign a category. Flag low-confidence matches.
  5. Step 4 (conditional): If the transaction exceeds the review threshold (e.g. over £500) or confidence is low, mark it for manual review instead of auto-categorising.
  6. Step 5 (integration): Write the assigned categories back to the bank export sheet for auto-categorised transactions.
  7. Step 6 (integration): Post a daily summary to Slack — number categorised, number flagged, and any anomalies.

Integrations Used

  • Google Sheets — bank transaction data source and destination for categorised entries; also holds the chart of accounts reference
  • Slack — daily summary with flagged items and anomalies

Who This Is For

Bookkeepers, fractional CFOs, and small business owners who manage their own books and want to reduce the time spent on transaction categorisation — particularly those processing 50+ transactions per month across multiple accounts.

Time & Cost Saved

Manual transaction categorisation takes roughly 1-2 minutes per entry. A business processing 100-200 transactions per month can expect to save 2-5 hours monthly, with the bulk of savings coming from recurring transactions that follow predictable patterns. Month-end reconciliation becomes significantly faster when most entries are already categorised. The workflow uses integration, LLM, and conditional steps, costing a few credits per daily run.