What a matching rule is
A matching rule is a hard instruction the engine checks on new bank transactions that are not linked to an invoice automatically: if the transaction looks like X, then categorise it as Y (or: ignore it). Rules are evaluated in priority order, and the first match wins.
Rules sit at the very top of the pecking order:
- Your rules: always win.
- Learned knowledge: what the system has learned from your corrections.
- AI suggestion: the model's own guess for a bank transaction, shown as an accept-chip and never applied on its own.
Before you ask for a rule: you probably don't need one
For the classic cases ("AWS is always Infrastructure", "Stripe payouts are always Revenue"), category learning often covers it without any setup: choose the same category for a counterparty's bank transactions three times and it is suggested on that counterparty's new transactions, for you to accept. It does not apply to invoices, and it never books on its own. A rule is the way to make a category automatic.
Rules earn their keep for the cases learning can't express:
- Ignore rules: e.g. permanently ignore a specific internal counterparty so it never shows up as open work.
- Overrides: when a vendor must land in a category different from what you usually pick elsewhere, no exceptions.
How to get a rule today
There is currently no rule-builder screen in the app. Rules are configured on the backend, per company. If you have a case that learning doesn't cover:
- Write to [email protected] with the condition ("counterparty contains …", "description contains …", "amount between …") and the action (category or ignore).
- We set the rule up for your company and confirm.
- From then on it applies to new transactions that match its condition and aren't linked to an invoice automatically.
Troubleshooting
A rule seems to fire on the wrong transactions. Rules match exactly what their condition says, so tell support the transaction that was hit and we adjust the condition or its priority.