AI bookkeeping: books that keep themselves

What happens to your data

1. Import. Excel/CSV statements from any bank, document registers, trial balances, PDF documents and receipt photos. The table header is found even when a file starts with a title and blank rows.
2. Understanding. Each transaction gets a category (revenue, rent, salary, taxes, transfers, etc.) with a confidence level. If your file carries your own category, it wins over any guess.
3. Postings. Double entry on Uzbekistan’s NAS chart of accounts: bank 5110, cash 5010, revenue 9010, VAT 6411/4410 and others. Every entry balances.
4. Matching. A payment is linked to its invoice, act, VAT invoice or receipt; a part payment reduces what is owed instead of closing the invoice.
5. Questions. Anything that cannot be decided with confidence goes to the Exception Engine.

Reading documents

From a PDF with a text layer Centa takes the number, date, parties with STIR, net amount, VAT, total, contract and due date — by their field labels, in Russian, Uzbek and English. The figures must prove each other: net + VAT = total only at a real rate. A field the document does not contain stays empty.

Learning from your decisions

When you choose a category for a counterparty, Centa remembers the rule and posts such payments itself next time. Rules are visible and editable in Company Brain: you can disable or change them.