Auto-categorisation

Every transaction filed, no 'what was this?' moments.

Finn reads each transaction, scores the likely category, and applies it. It learns your patterns and flags anything it isn't confident about.

Free for 14 days, no credit card

accounts-os.com/transactions
CategorisationLive
Why did you file this as Software?
FRecurring monthly charge from Notion, matches your last 8 payments to them.
FFinn categorised all 34 transactions in this import.
Live

In short

AccountsOS auto-categorises every transaction against the UK GAAP chart of accounts using Claude, with a confidence score. High-confidence matches apply automatically; lower-confidence ones surface for review. Combines with bank rules for fully automated categorisation of recurring payees.

60+
UK GAAP categories
Claude
Powered
Learns
Your patterns

Everything it does

How it works

  • Reads counterparty, amount, description
  • Scores likely categories
  • Applies high-confidence matches
  • Surfaces low-confidence for review
  • Learns from corrections

Combines with bank rules

  • Set a rule once, applies forever
  • Pattern + amount + description triggers
  • Layered with AI categorisation
  • Audit trail for every decision

Capabilities at a glance

AI categorisation powered by Claude

60+ UK GAAP categories

Confidence scoring

Auto-apply at high confidence

Surfaced review queue for low confidence

Learns from your corrections

Combines with bank rules

Per-trading-entity chart of accounts

How It Works

1

Transaction lands

Bank import, expense, bill: anything that needs a category.

2

AI scores categories

Reads counterparty, amount, description, history.

3

Apply or queue

High-confidence applied; low-confidence queued for one-click review.

4

You correct, AI learns

Your corrections feed forward. Future similar transactions get it right.

Real-world Use Cases

Bank import day

60 transactions imported. 50 auto-categorised at high confidence. 10 in the review queue, all cleared in 2 minutes.

Recurring AWS

AI categorises AWS as IT/Software every time without a rule. After 3–4 confirmations it's near-100% confident.

Why founders pick this over the spreadsheet

Stop categorising the same payee twice

Confidence scores tell you what to actually look at

Learns your patterns so accuracy improves over time

Frequently Asked Questions

How accurate is it?

~85–95% on transactions with clear vendors. Lower for ambiguous ones (e.g. PAYPAL: could be any merchant). Confidence scoring tells you what to review.

Does it learn?

Yes. Every correction you make feeds forward. The AI gets more accurate with use.

Can I override?

Yes. Change any category in one click. Bulk-edit hundreds at once.

What about bank rules?

Layered. Bank rules apply first (set once, applies forever); AI categoriser handles anything not covered by a rule.

Ready to try Automatic Categorisation?

Set up in minutes. Finn works in the background on your books, wherever you're based.

Free for 14 days, no credit card, cancel anytime