Automated Account Assignment & the SME Chart of Accounts: AI Booking Suggestions with Human-in-the-Loop

What is automated account assignment?

Automated account assignment refers to the machine-based mapping of a business transaction – such as a supplier invoice or a bank transaction – to a debit and a credit account, along with the appropriate VAT code. In Switzerland the basis is usually the Swiss SME chart of accounts (Kontenrahmen KMU), originally published by Karl Käfer in 1947 and last revised in light of the 2020 company-law reform (in force since 1 January 2023). It is published today by the association veb.ch – Verband der Experten und Berater in Rechnungslegung und Controlling, which has operated under the name SwissAccounting since 2023.

The SME chart of accounts is organised decimally into classes 0 to 9: class 0 fixed assets, class 1 assets (current assets), class 2 liabilities (both in the balance sheet), classes 3 to 8 in the income statement (3 operating income from goods and services, 4 cost of materials, merchandise, services and energy, 5 personnel expense, 6 other operating expense, 7 ancillary operating result, 8 non-operating, extraordinary, one-off or prior-period result and taxes) plus class 9 closing entries. The law prescribes no specific chart of accounts: Art. 957a CO requires the complete, truthful and systematic recording of business transactions and further principles of proper accounting (including clarity and verifiability); companies may freely choose their actual account structure as long as these principles are observed.

AI does not replace this legal obligation but supports it: it reads document data (amount, date, VAT, supplier), compares it with historical entries and proposes an account assignment. Approval remains with people.

Why this matters for SMEs and fiduciaries

Document capture and account assignment are among the most repetitive and at the same time most error-prone tasks in accounting. Every misassigned invoice affects the income statement, the VAT return and ultimately the financial statements. If an expense is posted to the wrong account or a VAT code is mixed up, corrections arise that become costly downstream – particularly with VAT, where input tax and output tax must be correctly separated.

For fiduciary firms, automated account assignment shifts the effort from pure data entry towards review and advice. Instead of booking each item manually, specialists check the AI's suggestions and intervene in special cases. The goal is not to hand off responsibility but to direct human working time to the points where judgement is genuinely required.

Important: accuracy figures from individual vendors (often in the range of 'over 90 per cent') are not guaranteed values. They depend heavily on document quality, industry and training history. Handwritten documents and complex VAT constellations noticeably lower the hit rate.

How AI booking suggestions are produced

In the first step the software extracts the document data – via OCR and increasingly via AI models – i.e. amount, date, supplier, invoice number and stated VAT. In the second step the system searches for similar, already-booked transactions of the same client and derives an account-assignment suggestion including a VAT code. The relevant rates have been unchanged in Switzerland since 1 January 2024: standard rate 8.1%, reduced rate 2.6% (e.g. food, medicines, printed matter without advertising character and – since 1 January 2025 – menstrual hygiene products) and special rate 3.8% for accommodation. The current ESTV list remains authoritative.

The feedback loop is decisive: if a specialist corrects a suggestion, that correction flows back into the suggestion logic. In supplier- or pattern-driven approaches this produces stable rules over time ('supplier X → account Y, VAT code Z'). In model-based approaches the corrections improve the statistical assignment.

Good systems clearly separate suggestion from posting: as long as no one approves, nothing is booked. This separation – human-in-the-loop – is the core of a controllable workflow and enables a traceable audit trail as expected by accounting law.

Workflow: from document to approved entry

1. 01Capture document: an invoice or bank transaction is imported; amount, date, supplier and VAT are extracted via OCR/AI.
2. 02Generate suggestion: the system matches the case against booking history and proposes debit/credit accounts and a VAT code based on the SME chart of accounts.
3. 03Review (human-in-the-loop): a specialist checks account, amount and VAT code; unusual or uncertain cases are flagged.
4. 04Correct & approve: the suggestion is adjusted if needed and then deliberately approved – only then is it booked.
5. 05Let it learn: the correction feeds back into the suggestion logic so comparable cases are assigned better in future.
6. 06Control: periodic spot checks and a review before VAT filing and year-end safeguard quality.

When AI account assignment pays off

The greatest benefit arises with high, uniform document volumes: recurring suppliers, subscriptions, expenses, bank statements with many similar transactions. Here the AI delivers reliable suggestions after a short learning phase because it builds on many comparable cases.

It is also worthwhile where consistent account assignment across time and people is required. Trained suggestion logic books the same matter the same way every time and thus reduces individual deviation. In client relationships with a clearly defined chart of accounts and stable processes, automation delivers its value fastest.

A prerequisite is a well-maintained data basis: a cleanly structured chart of accounts, correctly stored VAT codes and a sufficient history of approved entries the system can learn from.

When caution is warranted

For rare, complex or legally sensitive transactions the AI should not run unchecked. This includes mixed VAT rates on a single document, acquisition tax, margin taxation, period accruals, asset additions with capitalisation questions or intra-group settlements. Here professional judgement is required, and a statistical suggestion can mislead.

A training history that is too small or biased is also a warning sign: if the system learns from few or faulty entries, it reproduces their errors. If a wrong account assignment is repeatedly approved, the error becomes entrenched in the suggestion logic. Spot checks and periodic reviews therefore remain indispensable.

This is not legal or tax advice. Responsibility for proper bookkeeping and VAT accounting remains unchanged with the company or the mandated fiduciary, regardless of which tool generates suggestions.

FAQ

Related topics

Sources

1. veb.ch / SwissAccounting – Schweizer Kontenrahmen KMU · 2026-06
2. Verlag SKV – Schweizer Kontenrahmen KMU (Mattle/Helbling/Pfaff, Ausgabe Aktienrechtsrevision) · 2026-06
3. ESTV – Mehrwertsteuersätze Schweiz (Normalsatz 8,1 %, reduziert 2,6 %, Sondersatz 3,8 %) · 2026-06
4. Fedlex – Art. 957a OR, Buchführung (Seite erfordert JavaScript) · 2026-06