How Penny Categorises Your Transactions

How Penny categorises your transactions

Penny, your AI bookkeeper, automatically categorises every transaction that comes in from your connected bank accounts. Here's how the system works and why it gets smarter over time.

The three-tier categorisation system

Penny uses a three-tier approach to categorise each transaction:

1. HMRC category — The top-level category that maps to your tax return. For example, "Travel and subsistence" or "Office, property and equipment". These are the categories HMRC expects to see in your quarterly submissions.

2. Subcategory — A more specific grouping within the HMRC category. For example, under "Travel and subsistence", you might see "Fuel", "Public transport", or "Parking".

3. Description — A plain-English description of what the transaction was for. For example, "Diesel for work van" or "Train to client meeting in Manchester".

This three-tier structure means your books satisfy HMRC's requirements whilst also giving you genuinely useful insight into where your money goes.

How the AI makes decisions

When a new transaction arrives, Penny considers several signals:

• Merchant name and reference — The payee name and any reference data from your bank. Penny recognises thousands of common UK merchants and suppliers.
• Transaction amount and pattern — Regular payments of the same amount often indicate subscriptions or recurring business costs.
• Your transaction history — If you've previously categorised a payment to the same merchant, Penny learns from that.
• Your industry — The business type you selected during setup helps Penny make smarter initial guesses. A payment to Screwfix means something different for a builder than for an office worker.
• Timing and context — Penny considers when the transaction occurred and what other transactions happened around the same time.

Confidence scoring

Every categorisation comes with a confidence score that determines what happens next:

• 95% and above — Penny auto-categorises the transaction. No action needed from you.
• 80–94% — Penny suggests a category and asks you to confirm.
• 50–79% — Penny asks you to choose from a shortlist of likely categories.
• Below 50% — The transaction is flagged for manual review, or referred to your accountant if one is connected.

Learning over time

Every time you confirm, correct, or manually categorise a transaction, Penny learns. This means:

• Corrections to recurring merchants are applied automatically to future transactions from the same merchant.
• Your personal categorisation patterns are remembered.
• Penny's confidence scores improve the longer you use Accounted.

Most users find that after two to three months of use, the vast majority of their transactions are auto-categorised with high confidence.