Automation · Accounting Document capture and account coding with AI: what works reliably today and where people decide

All tasks of the department and the connection to your programs are on the page AI in accounting.

Where AI in accounting stands today

According to Eurostat, one in five enterprises in the EU with at least ten employees used AI in 2025, up from just under one in seven in 2024. AI is most often used to analyse written language, and that is the core of accounting: reading invoices, delivery notes, bank statements and emails and assigning them correctly.

Many accounting programs now come with their own AI features, for example for document recognition. But AI brings the greatest benefit where several systems come together: the inbox with the invoices, the ERP system with purchase orders and delivery notes, the accounting software and the bank.

What AI reliably takes over

What matters is the principle behind it: the AI prepares, people decide. Uncertain points are flagged, not silently skipped.

E‑invoicing as an accelerator

The more structured an invoice arrives, the less the AI has to guess. With an e-invoice in a format such as XRechnung, ZUGFeRD or via Peppol, amounts, tax rates and line items are already machine-readable in the file. The AI no longer has to recognise them, only check and assign them. With the EU package “VAT in the Digital Age”, structured e-invoices become mandatory for cross-border business within the EU from July 2030, and many countries already have their own obligations today. What applies in your country is covered in our guide to e-invoicing, and how incoming invoices are automated in the article Automating incoming invoices.

A typical process by example

A trading company receives several hundred incoming invoices a month, as PDFs by email, as e-invoices and occasionally on paper. The AI reads each invoice, looks up the matching purchase order and delivery note in the ERP system and compares quantity, price and payment terms. If everything matches, it creates a posting proposal with account and cost centre. If something differs, such as a higher price than ordered, the invoice goes to purchasing with a short explanation.

In the morning the bookkeeper sees a list: most invoices with a finished proposal for approval, a few with flagged deviations. This example describes a typical process, not a client project.

Proper bookkeeping and data protection

The rules for proper bookkeeping apply regardless of whether a person or an AI captures a document. In Germany, the GoBD describe what this requires: traceability, immutability, complete retention and a description of the procedure. In Austria, the Federal Fiscal Code governs retention, and other countries have comparable rules. For AI‑assisted accounting this means:

• Every proposal and every approval is logged, with time and person.
• Original documents remain stored unchanged, the AI works with copies and extracts.
• The procedure description explains what the AI does and where people check.
• A data processing agreement governs the handling of personal data, for example of sole traders on invoices.
• The data is stored on servers in the EU where possible.

Agree the introduction with your accountant or tax adviser. They know your bookkeeping and have to work with the result.

How to start

1. Choose one process: usually incoming invoices, because that is where most documents arise.
2. Measure today: how many invoices a month, how long per invoice, how many queries?
3. Tidy up master data: suppliers, accounts and cost centres should be current and unambiguous.
4. Run in parallel: for a few weeks the AI creates proposals while your team posts as before and compares.
5. Take over step by step: only when the proposals are right do you approve instead of posting from scratch. Payment matching or dunning follow afterwards.