Transaction categorisation
Automatic transaction categorisation to your chart of accounts, with your rules deciding first.
Every line on a receipt, invoice or bank statement needs an account. Pileform proposes one from your own chart of accounts: your categorisation rules run first, then what it has learnt about that supplier, and an AI suggestion only when neither answers. Learning is kept per client, and nothing posts until you confirm it.
Automatic transaction categorisation gives every receipt, invoice and bank line an account from your own chart of accounts. In Pileform your categorisation rules decide first, then remembered supplier codes, and an AI suggestion is made only when neither applies and only above a confidence floor. Learning stays per client company, and nothing posts until you confirm it.
Why categorisation should not start with a guess.
Most automatic categorisation starts with a model’s guess and asks you to correct it. For a practice that is backwards. You already know that the electricity supplier goes to utilities and the fuel card to motor expenses, and a guess that ignores what you know produces the same correction every month. One client’s habits must also never leak into another client’s coding. Categorisation is the step between reading the pile, which is invoice automation, and posting it.
Pileform runs your rules before any AI step. Description-pattern, SKU-pattern and supplier-default rules decide first, and they adjust from your corrections. Next comes counterparty memory: each supplier’s aliases, VAT number and default debit and credit codes, remembered per company. Only when neither applies is an AI suggestion made, and only above a confidence floor. Bank lines follow the same idea in tiers, set out under bank statement automation. The chart itself can be seeded from a Cyprus default template or synced from Xero or QuickBooks, so the codes proposed are the codes you post to.
How a line gets its account.
Your chart, per company
Each client company has its own chart of accounts: seeded from a Cyprus default template, synced from Xero or QuickBooks, or built by hand. A deleted account is soft-deleted, so the history behind it stays intact.
Your rules first
Description patterns, SKU patterns and supplier defaults are evaluated before anything else. A rule you wrote always comes before a suggestion.
Then memory, then a suggestion
A supplier already known for this company gets its remembered codes. Only a line nothing else explains gets an AI suggestion against the full chart, and only above a confidence floor; below it, the line is left for you.
Your correction counts
Confirm, edit or reject in the review queue. Rules adjust from your corrections and a supplier’s confidence rises as you confirm, for that company only.
The order it decides in.
Each proposed account records where it came from, and the posting history keeps that record.
- 1. Your rules
- Description-pattern, SKU-pattern and supplier-default rules, adjusting from your corrections
- 2. Counterparty memory
- Each supplier’s or customer’s aliases, VAT number and default debit and credit codes, with a confidence that rises as you confirm
- 3. AI suggestion
- Only for a line nothing else explains, mapped against your full chart, and only above a confidence floor
- Bank transactions
- Transaction-type defaults, then a fuzzy match to known counterparties, then an AI suggestion against the chart, then you
- Audit trail
- What suggested each account, its confidence and the action taken, kept in the posting history
Who needs categorisation that stays put.
Practices with many clients
Twenty clients, twenty charts of accounts, twenty independent rule sets. One client’s coding never shapes another’s.
Bookkeepers taking on a client
Sync the chart from Xero or QuickBooks, write down the few rules you already know, and let the first month’s corrections shape the second month’s proposals.
Businesses starting from the Cyprus template
Begin with the Cyprus default chart of accounts and let the supplier memory fill in as the bills come through.
Transaction categorisation, answered.
Only as a last resort. Your categorisation rules run first and a remembered supplier’s default codes second. An AI suggestion is made only for a line neither covers, only above a confidence floor, and it is still a proposal that waits for you in the review queue.
No. Rules, counterparty memory and the chart of accounts are all kept per company, so twenty clients under one practice login are twenty independent rule sets, never one blended guess.
Yes. Sync it from Xero or QuickBooks, seed it from the Cyprus default template, or build it by hand. Every proposal is a code from that chart, which is the chart you post to.
In tiers of their own: defaults by transaction type, then a fuzzy match against counterparties the company already knows, then an AI suggestion against the full chart, then a person. A bank line matched to an invoice carries the VAT already decided on that invoice.
Edit it in the review queue before anything posts. The correction is recorded in the posting history and your rules adjust from it, for that company only.
Bring your chart and one month.
A free account comes with 30 free processing pages that do not expire and needs no card. Sync your chart of accounts, upload a month, and see which proposals came from your own rules.