AI accounting: From invoices to a reliable month-end close for Swiss SMEs

A month-end close becomes reliable when documents are retrievable, responsibilities are clear, and unresolved items remain visible. AI accounting can support Swiss SMEs in the preparatory work from document arrival to reconciliation. It does not remove the need to assess posting, VAT treatment or closing evidence.

For a business in Zurich, Zug or elsewhere in Switzerland, the useful starting point is not “automate everything”. Start by making the document flow, approval path and controls clear. Technology can then surface suggestions and exceptions, while employees and fiduciary professionals assess the cases that require business context.

Document capture is not the same as OCR, rules or AI

The terms are often used as if they meant the same thing. Document capture means that an invoice, receipt or PDF reaches a defined intake channel, such as a shared mailbox, scanner or mobile upload. A photographed receipt still needs a link to a transaction, supplier and accounting period.

OCR, or optical character recognition, extracts visible text from an image or PDF. It may read an invoice number, date, amount or IBAN. It does not decide reliably whether expenditure is business-related, which VAT code applies, or whether the invoice was paid. A misread date remains an error when transferred into a field.

Rules apply explicit instructions. A known supplier might be routed to a cost centre, for example, or invoices above a limit may require an additional approval. When exceptions are documented, rules are easy to review.

AI-enabled functions can add pattern-based suggestions, such as likely accounts or supplier matches. A suggestion is not an accounting conclusion. New suppliers, split supplies, unusual costs and uncertain tax situations should go to a review queue. This is how AI accounting can help triage work without pretending to replace judgment.

Build the creditor approval before payment or posting

A sound accounts-payable process separates capture, substantive review and payment approval. That protects against both incorrect postings and payments that cannot be connected to a genuine delivery. An invoice can be visible in the ledger without being approved for payment.

A practical approval matrix reflects value, expense type and responsibility. The person who ordered or received the goods or service confirms what was received. A second role checks supplier details, payment information, due date and, where relevant, the purchase order or delivery note. Changed bank details warrant independent confirmation; a single invoice is not enough.

Creditor approval checklist

• Is the original document, or a traceable electronic document, available?
• Do supplier, invoice number, service date, amount and currency agree with the order, contract or delivery note?
• Has the service been received, and does the cost centre or project assignment make sense?
• Is this a potential duplicate, for instance because the supplier, number, amount or date combination is already present?
• Has the proposed VAT code been reviewed, and is the approving person recorded under the agreed authority limit?

The workflow can be digital. What matters is that the approval remains traceable by person, date and decision. The Swiss Ordinance on Keeping and Preserving Accounting Records requires changes to be detectable and calls for documented responsibilities, processes and infrastructure so that records and supporting documents can be understood. [4]

Treat duplicates, VAT and bank reconciliation as distinct controls

Duplicate detection is valuable, but it is not a fraud or error guarantee. A system can flag identical invoice numbers or similar amounts. It cannot know whether two instalment invoices are legitimate or whether the same document returned with another date. Each match needs a documented decision: stop, merge, reject or explain.

VAT review is more than accepting a proposed rate. Check invoice details, the supply, tax period and whether the transaction belongs to input VAT, output VAT or a category without input tax recovery. The Swiss Federal Tax Administration (FTA) publishes current VAT practice online. For special situations, consult it instead of relying on an unreviewed automated suggestion. [5]

Bank reconciliation is a separate control layer. It connects a ledger payment with money that entered or left the business account. Use clear categories for unresolved movements: fees, batch payments, open customer items, supplier invoices not yet posted, or items needing clarification. Do not force a reconciliation through a blanket difference entry.

AI accounting at month-end: the reconciliation record matters

The decisive test is the month-end close. By an agreed cut-off, documents are collected, approvals followed up and bank movements matched. Then exceptions need expert review: unclear postings, missing evidence, unreconciled payments, unusual balances and VAT questions.

A brief reconciliation record makes this stage transferable. It does not need to be a lengthy report. It should name the period, responsible person, controls performed, open items with an owner and due date, and the decision to release or escalate. For cooperation with a fiduciary advisor in Zurich or Zug, this prevents a close from becoming a set of system suggestions that were silently accepted.

Handover checklist before the monthly review

• Documents for the period are captured, assigned and annotated where exceptions remain.
• The payables list contains only approved, open or deliberately blocked items.
• Bank accounts, cards and payment providers are reconciled; every difference has an explanation and owner.
• VAT-relevant entries and corrections have been reviewed; unresolved questions are marked rather than hidden.
• The reconciliation record identifies open items, supporting material and the handover date for the responsible professional.

Swiss businesses may retain records electronically when the link to the underlying transaction is assured and the records can be made readable at any time. Accounting records and supporting documents generally need to be retained for ten years. [3] That is why the evidence trail and archive should be designed at the start, rather than added shortly before the annual financial statements.

Review cloud and privacy controls deliberately

Accounting data often includes personal data. According to the Federal Data Protection and Information Commissioner (FDPIC), the Swiss Federal Act on Data Protection applies directly to AI-supported processing. Transparency about purpose, functionality and data sources is part of a responsible rollout. A data protection impact assessment may be needed where risks are high. [1]

Cloud use is not categorically prohibited. The FDPIC does stress that the company as controller must verify and contractually secure compliant processing. Review safeguards, sub-processors, documented instructions, deletion or return of data, and potential transfers abroad. [2]

Accounting also requires integrity, access protection, readability and logging. For changeable media, the GeBüV refers to integrity measures, verifiable storage timing, documented procedures and preserved information such as log files. [4] The suitable option depends on data, contract and process. Hosting claims do not replace a review.

A 30-day proposal, not a performance promise

Implementation need not start as a large programme. The following is a non-binding proposal for a limited pilot, such as incoming supplier invoices. It is not a commitment to faster closes or error-free results.

Days 1–5: Map document types, intake channels, roles, payment approvals, posting rules and archive questions. Choose a narrow scope and observations, such as exception types, rather than promising savings.

Days 6–12: Clean up mandatory fields, supplier master data, approval limits and duplicate criteria. Define which suggestions always go to a review queue and which circumstances must always be assessed manually.

Days 13–20: Test with a limited set of real documents. Compare OCR output with the source, review proposed accounts and VAT codes, and record variations. Check archiving, access rights and the cloud agreement against the business’s requirements.

Days 21–30: Run a trial bank reconciliation and prepare a reconciliation record. Review exceptions, update the working instructions and decide together whether to keep, adjust or stop the pilot. Expansion should come only after that decision.

Professional responsibility remains personal

AI accounting can sort, suggest and make unusual items visible. It cannot make a professional assessment of an exceptional case or take responsibility for a coherent monthly or annual close. That responsibility needs clear roles in the company and, where appropriate, review by a qualified fiduciary professional.

Global City Finance AG offers personal collaboration for financial accounting and annual financial statements. If you would like to connect digital accounting for SMEs with clear approvals and a traceable handover, see our services. To discuss your starting point, you can contact us or request a non-binding quote.

FAQ: AI accounting for Swiss SMEs

Is OCR already AI?
OCR primarily extracts text from documents. Software may also use rules or learning-based suggestions. In practice, what matters is which data is actually extracted and who reviews it.

Can a Swiss SME process invoices in the cloud?
It is not generally prohibited. First assess the data, provider, contract, safeguards, sub-processors and any international data transfers. [2]

Can a posting suggestion be booked straight away?
For recurring, clearly governed cases, a business may define a controlled workflow. New, unusual or VAT-relevant cases should remain in a review queue. Professional responsibility is not delegated to the suggestion.

What should I hand over for the month-end review?
Provide reconciled bank and payables lists, all marked exceptions, missing documents, VAT questions and a short reconciliation record. The professional can then review targeted items instead of reconstructing the process.

Sources