Automatic booking with AI: how it works for firms

What is automatic booking with AI?

Automatic booking with AI is processing purchase invoices, sales invoices and receipts into entries in the accounting software, with software taking over the reading, coding and recording. Your team no longer fills in every field, but only reviews what the AI cannot process with enough confidence. The work shifts from typing to checking, and from every document to the exceptions.

A correct booking takes more than reading an amount. It needs a linked supplier or customer, a general ledger account and VAT code per line, sometimes a cost centre or project, and a check that subtotal, VAT and total add up. The invoice date, invoice number and payment terms also have to be recorded correctly. Automatic booking with AI performs all of those steps, based on what happened earlier in the same administration and on the agreements your firm has recorded.

What is the difference between OCR, templates and AI recognition?

OCR (optical character recognition) turns a scanned or photographed invoice into readable text, but does not know what that text means. Older systems combine OCR with templates: for each supplier you define where the invoice number, date and amount sit on the page. On top of that you often configure per supplier which ledger account and VAT code the invoice should go to. It works, but takes a lot of setup and maintenance.

Templates work as long as nothing changes. When a supplier switches layout, a new supplier appears or an invoice deviates slightly, things go wrong or someone has to fix the template. AI recognition looks at the content and context of the document instead of fixed positions. That is why Autoboeker works without templates: you never create or maintain them, not even for suppliers you only see once.

How is an AI booking proposal created?

An AI booking proposal is built from the document itself, the history of the administration and the agreements your firm has recorded. Autoboeker follows the same path for every document, regardless of who books it or what the invoice looks like. Master data from the accounting software, such as ledger accounts, VAT codes and contacts, sets the boundaries: the AI only proposes accounts and codes that exist in the administration.

When proposing, the AI weighs previous bookings, your corrections, supplier behaviour, descriptions, amount patterns and frequency. On top of that come administration rules, global rules and AI instructions written in plain language, such as an agreement that fuel receipts always go to travel expenses. The result is a fully completed booking with a confidence score and an explanation of what the proposal is based on. Each document follows this path:

1. 1Receive: the document arrives via upload, a purchase or sales email address, the client app, WhatsApp or Connect.
2. 2Understand: supplier, date, invoice number, amounts, lines, VAT, ledger account, cost centre and optionally project are recognised.
3. 3Check: subtotal, VAT and total are recalculated, VAT numbers are checked against VIES, IBANs are verified and duplicate invoices flagged.
4. 4Propose: history, rules and instructions lead to a proposal with a confidence score and explanation.
5. 5Book: above the threshold directly into the accounting software, below it ready for review.

When does the AI book automatically, and when not?

The AI books automatically when the confidence score of the proposal is above the threshold you have set. If the score is below it, the proposal waits for review and a team member decides. A low score is not an error but a signal: for example a new supplier, an unusual amount or an invoice that could be booked in more than one way.

You choose per firm or per administration how far you go. There are three levels, and an obvious approach is to start cautiously and move up as the AI gets to know an administration. You can also use approvals to record who signs off on what, before or after booking, so that automatically booked invoices stay within your firm's process. The three levels are:

• Suggestions: the AI fills in everything, you click to approve.
• Smart confirmation: you only see the deviations.
• Fully automatic: high-confidence bookings are posted directly.

What booking logic can the AI apply?

The AI can choose more than one ledger account per invoice. Autoboeker books at line level and can split or summarise lines automatically, for example by merging lines that share the same ledger account and VAT code. That is useful for long invoices from wholesalers or web shops, where you do not want every item line in the books but do want a split by cost type and VAT rate.

More complex situations are supported too, such as allocation across multiple ledger accounts, a percentage split per booking, cost centres, cost units and projects, and accruals and deferrals (transitorische posten). New suppliers or customers are recognised and can be created automatically, payment terms are read or created and foreign currencies are detected, with optional automatic conversion. Recurring invoices and regular suppliers are recognised, so proposals for known costs stay consistent.

How do you stay in control of what the AI books?

You stay in control because every booking is explainable and traceable. With each proposal you see the confidence score and what it is based on, such as similar bookings, the VAT code and the ledger account. Proposals and processed bookings can always be edited by hand, and you can have a document processed again with an extra note for the AI if the proposal is not right.

Autoboeker also flags high amounts, unknown suppliers, deviating booking behaviour and missing information instead of booking them silently. Every action by a person or the AI is recorded in the audit log, so you can see afterwards who or what created or changed a booking. When you correct a booking, the correction is saved in the feedback history and you can apply it to similar bookings.