Reading and Booking QR-Bills & eBill/Peppol Automatically with AI

What this is about

In Switzerland, incoming invoices reach a fiduciary office today in three technically different forms: as a QR-bill (PDF or paper with a Swiss QR Code), via eBill (digital invoice directly in e-banking), and increasingly as a Peppol document (structured XML over the Peppol network). All three can be processed automatically – but with very different effort and data quality.

The Swiss QR Code data of the QR-bill is machine-readable and standardised: SIX (SIX Group AG) defines in the "Swiss Implementation Guidelines QR-bill" exactly which fields the QR code contains. In addition, the industry body Swico defines in a separate syntax specification how the optional "Billing information" field (S1 string) – with VAT breakdown, invoice number and payment terms – is to be populated. Since version 2.3 (in force since 22 November 2025) only the structured address is permitted; a transition period for legacy data runs until 30 September 2026. Version 2.4 was published in February 2026 and only enters into force with the SIC release of 13 November 2026 (valid from 14 November 2026); it introduces no technical changes for CHF invoicing, and version 2.3 remains valid until November 2027.

eBill and Peppol deliver invoice data already fully structured (no OCR needed). For classic PDF or paper invoices without a QR code, OCR remains the only way. AI connects these worlds: it reads the QR payload and Swico string precisely, fills gaps via OCR, and proposes a coding – which the human reviews.

Why this matters for fiduciaries

Document capture is one of the biggest time sinks in a fiduciary office and at the same time error-prone: typing amounts, assigning the creditor, setting the VAT code, choosing the account. With the QR-bill a substantial part of this data is already contained in the QR code – payee IBAN, amount, currency, reference and, if the biller supplies the Swico S1 string in the "Billing information" field, the VAT breakdown, invoice number and – derivable from the payment terms – the due date (field /40/ encodes discount days and the net payment period, from which the date is calculated together with the invoice date). Reading these fields correctly instead of retyping them eliminates typos.

The three QR reference types determine matching: QRR (27-digit QR reference, requires a QR-IBAN) is unique per invoice and therefore ideal for automatic matching; SCOR (Creditor Reference "RF…" with a normal IBAN) also works internationally; NON carries no reference and must be assigned via other attributes.

eBill and Peppol reduce manual effort further because no image recognition is needed. eBill is operated by SIX on behalf of the Swiss financial centre and counted over 3.5 million registered users at the end of 2024; according to the current ebill.ch platform the figure is now over 3.8 million (as of June 2026). Note: There is no general B2B/B2C e-invoicing mandate in Switzerland – e-invoicing is only mandatory in the B2G area for federal suppliers (federal administration) from CHF 5,000 contract value (since 2016); cantonal and municipal contracting authorities have separate rules.

How it works technically

1. Detect the source. The system classifies every inbound item: Peppol XML (UBL/CII, aligned to EN 16931), eBill record, PDF with a Swiss QR Code, or plain PDF/paper. Structured sources (Peppol, eBill) are ingested directly; they need no OCR.

2. Decode the QR code. For the QR-bill the Swiss QR Code is extracted from the PDF and the payload parsed per the SIX guidelines: IBAN/QR-IBAN, amount, currency (CHF/EUR), reference type (QRR/SCOR/NON), reference, payee with structured address, and – if present – the Swico string from "Billing information" (VAT rates, invoice number, and the due date derivable from the payment terms). These values are exact, not "guessed".

3. OCR only as a supplement. What the QR code does not contain (service description, line items, differing VAT details) is supplied from the PDF text by a language model. The AI thus fills gaps but never overwrites the exact QR fields.

4. Propose the coding. Based on creditor, text, amount and VAT code, the AI proposes an expense account and VAT key – learned from the respective client's prior bookings (own RAG knowledge base, no mixing of other clients' data).

5. Human reviews, system books. Every proposal goes to the clerk with a confidence level. Only after approval (human-in-the-loop) is it handed over via interface to Abacus, Bexio or Sage. Corrections flow back into the model.

Step-by-step workflow

1. 01Consolidate inbound channels: connect a Peppol access point, eBill reception and a shared mailbox/folder for QR-bill PDFs.
2. 02Classify each document (Peppol XML / eBill / QR PDF / plain PDF) and ingest structured sources directly without OCR.
3. 03For QR-bills, decode the Swiss QR Code: take over IBAN, amount, currency, reference type (QRR/SCOR/NON) and – if present – the Swico S1 string exactly; calculate the due date from the payment terms (field /40/) and the invoice date.
4. 04Supplement missing details (line items, service description) via OCR/language model from the PDF, without overwriting the QR fields.
5. 05Match the creditor against the client master and propose coding + VAT code from the client-specific learning base.
6. 06Present the proposal with confidence to the clerk for review (human-in-the-loop).
7. 07After approval, hand over via interface to Abacus, Bexio or Sage and feed corrections back into the model.

When it pays off

The benefit is greatest when you process many recurring incoming invoices with QR-bills – e.g. supplier, telecom, insurance or rental invoices with a stable creditor base. Here the system quickly learns the typical coding per client and supplier.

It also pays off when clients increasingly use eBill or Peppol: these documents are already structured and can be booked without OCR and without image uncertainty – the automation hit rate rises markedly.

For your clients' B2G suppliers (federal contracts from CHF 5,000) a clean Peppol/eCH pipeline is sensible too, because e-invoicing is mandatory there anyway and structured data is the standard.

When not to (or only cautiously)

With very low document volume or strongly varying one-off cases, the setup effort rarely justifies automation – manual capture is then faster.

Caution is needed with NON references and pure paper/scan documents without a QR code: the unique machine-readable reference is missing, matching relies solely on OCR and heuristics, and the error rate rises. Such documents should always be confirmed by a person.

Fundamentally: AI does not replace professional responsibility. Coding, VAT treatment and plausibility remain the fiduciary's responsibility. The structured reading is exact, the *coding proposals* are proposals – human approval is mandatory. This is not tax or legal advice.

FAQ

Related topics

Sources

1. SIX – QR-bill (Swiss Payment Standards) · 2026
2. Swiss Implementation Guidelines QR-bill v2.4 (Februar 2026, in Kraft 14.11.2026) · 2026-02
3. eBill – offizielle Plattform (SIX, im Auftrag des Schweizer Finanzplatzes) · 2026
4. Swico – QR-bill Supporting Material (Syntaxdefinition S1 / Validator) · 2026
5. Peppol BIS Billing 3.0 Specification · 2026
6. KMU-Portal Bund – QR-Rechnung einfach erklärt (SIX-Flyer) · 2026