The problem: too many invoices, too little time

An average Swiss fiduciary handles between 200 and 800 invoices per month per mandate. Each invoice requires reception, verification, accounting entry, payment matching and archiving. Using the traditional method — manually opening the envelope or PDF, keying in the data, assigning the account — an operator spends 4 to 7 minutes per invoice. Over 500 invoices, that is 40–58 hours of manual work per month, roughly 25% of a full-time employee's capacity.

The cost is not just in hours: transcription errors (wrong amount, incorrect VAT rate, wrong ledger account) generate corrections, rework and potential FTA penalties. The average time to fix an accounting error is 12–15 minutes — and on 500 invoices the typical manual error rate is 3–5%, meaning 15–25 additional corrections per month.

The good news: with a structured workflow based on OCR, AI categorisation and targeted validation, it is realistic to reduce processing time to 2–3 minutes per invoice and the error rate below 1%. This guide describes the operational process step by step, with concrete timings, tools and metrics.

The manual workflow: where time is lost

Before automating, it is useful to map the traditional process to identify bottlenecks. Here are the 5 typical stages of manual invoice management, with average time per invoice:

1. Reception and sorting

~1 min

The invoice arrives by post, email or supplier portal. The operator opens the envelope or PDF, checks that it belongs to the correct mandate and classifies it by type (supplier, expense, investment). On large volumes, sorting alone takes significant time.

2. Data entry

~2–3 min

The operator reads the invoice and manually enters: supplier, date, invoice number, gross amount, VAT rate, net amount, currency, order reference. Every field is an error opportunity: an '8' read as a '3', a misplaced decimal, a rate of 7.7% instead of 8.1%.

3. Account assignment

~1 min

The operator selects the correct ledger account from the chart of accounts (Kontenrahmen KMU), the cost centre and the VAT code. This requires knowledge of the chart of accounts and the specific mandate. For recurring invoices it is quick; for new suppliers or ambiguous categories it slows down considerably.

4. Approval and posting

~0.5 min

The entry is reviewed (often by the same operator), approved and posted in the system. In many fiduciaries there is no systematic dual-control process, which increases the risk of undetected errors.

5. Archiving

~0.5 min

The original document is archived — physically in a binder or digitally in a shared folder. Often without a standardised naming convention, making future retrieval slow and frustrating.

Total: 5–7 minutes per invoice. Over 500 invoices/month = 42–58 hours of manual work. At an average cost of CHF 65/hour for an accounting employee, that is CHF 2,700–3,800 per month for invoice recording alone — not counting error corrections.

Phase 1: OCR scanning and data extraction

OCR (Optical Character Recognition) is the first step to eliminating manual data entry. A good invoice OCR system does more than 'read text': it extracts structured data and maps it to accounting fields. Here is the 5-step flow:

Scanning or uploading

The invoice is scanned (desktop scanner, smartphone or network scanner) or uploaded directly as a PDF/image from the management system or email inbox. The system accepts PDF, JPEG, PNG and multi-page PDF. For Swiss QR-invoices, the QR code is decoded automatically without visual OCR.

Image pre-processing

The OCR engine straightens the image (deskew), adjusts contrast and brightness, removes noise and shadows. This step is critical for poorly scanned documents, faxes or photos taken in poor lighting. Good pre-processing can lift accuracy from 85% to 97%.

Key field extraction

The OCR identifies and extracts: supplier company name, address, VAT number (CHE-xxx.xxx.xxx), invoice date, due date, invoice number, gross amount, VAT rate, VAT amount, net amount, currency (CHF/EUR), bank details (IBAN), payment reference (QR-reference or creditor reference).

Automatic structural validation

The system checks internal consistency of the extracted data: does net + VAT equal gross? Is the VAT rate among those permitted (2.6%, 3.8%, 7.7%, 8.1%)? Does the IBAN have the correct format? Is the CHE number valid? Anomalies are flagged for human review.

Proposed booking entry

Validated data is used to generate a draft accounting entry: date, amount, supplier account, proposed expense account, VAT code. The draft is ready for human review or automatic posting if the confidence level exceeds the configured threshold.

Phase 2: intelligent AI categorisation

OCR extracts the raw data; AI interprets it. Intelligent categorisation goes beyond simple text recognition: it analyses the invoice context to suggest the correct accounting allocation.

Supplier recognition

AI identifies the supplier by matching company name, VAT number and IBAN against the existing master data. For recurring suppliers, it automatically applies the expense account used in previous invoices. For new suppliers, it suggests the most likely account based on the business category.

Ledger account mapping

Based on the invoice description, amount and supplier category, AI proposes the Swiss chart of accounts entry (e.g. 4000 Cost of goods, 6000 Rent, 6500 Insurance). The system learns from corrections: after 50–100 invoices, accuracy exceeds 95% for regular suppliers.

Cost centre assignment

For businesses with cost accounting, AI automatically assigns the cost centre based on configured rules (e.g. 'all Swisscom invoices → IT department') and learned patterns. The cost centre can also be derived from the order number or project indicated on the invoice.

Automatic VAT classification

AI determines the correct VAT code based on the rate stated on the invoice, the type of expense and the supplier's nature (Swiss/foreign). It correctly handles complex scenarios: reduced-rate VAT, exempt, reverse charge for foreign services, mixed invoices.

Confidence level and thresholds

Every AI proposal has a confidence score (0–100%). Thresholds are configurable: above 95% the entry can be posted automatically; between 80% and 95% it is presented for quick validation; below 80% it requires full manual review. This threshold approach balances speed and control.

Phase 3: targeted human validation

Automation does not eliminate human control — it makes it more efficient. Instead of checking every invoice from scratch, the operator focuses only on cases that need attention. Here is when human review is required:

New supplier not in master data

When the system finds no match in the supplier master data, it presents the invoice for supplier record creation: company name, address, IBAN, default ledger account. After the initial setup, subsequent invoices from the same supplier will be automatic.

AI confidence below threshold

Invoices with unusual layouts, poorly scanned documents or anomalous amounts may have a low confidence score. The operator sees the AI proposal alongside the original image and can confirm, correct or reject with a single click. Average validation time: 15–30 seconds.

Amount or VAT discrepancy

If net + VAT does not equal the total, or if the VAT rate does not match the expense type, the system flags the anomaly. The operator checks the original and decides: supplier error (credit note to be requested), OCR error (manual correction) or legitimate exception.

Invoices above amount threshold

For invoices exceeding a configurable amount (e.g. CHF 5,000 or CHF 10,000), the system always requires explicit approval regardless of confidence level. This ensures that significant expenses always pass through a manager's review.

Phase 4: automatic bank reconciliation

Once posted, the invoice must be matched to the corresponding payment. Automatic reconciliation compares recorded invoices with bank transactions imported via API/CAMT.053:

Exact match (1:1)

The simplest case: one invoice matches one bank payment for the same amount, with matching QR or creditor reference. The system automatically matches and closes the item. On invoices with Swiss QR-code, the automatic match rate exceeds 98%.

Cumulative payment (N:1)

A single payment covers multiple invoices from the same supplier. AI identifies the combination of open invoices whose total matches the payment and proposes the multiple match. The operator confirms with a single click if the proposal is correct.

Partial or excess payment

The payment does not exactly match the invoice amount: early payment discount applied, rounding, withholding. The system flags the difference and proposes the variance entry (discount account, exchange differences, rounding).

Unmatchable payment

Bank transactions without a clear reference (manual transfers, payments with generic descriptions). The system proposes likely candidates based on amount, date and supplier, but requires manual confirmation. These cases typically represent 5–10% of total volume.

Savings metrics: before and after automation

Realistic comparison for a fiduciary handling 500 supplier invoices per month, based on real AccountEX client benchmarks:

Metric	Before (manual)	After (OCR + AI)
Average time per invoice	5–7 minutes	2–3 minutes
Error rate	3–5%	< 1%
Monthly hours dedicated	42–58 hours	17–25 hours
Monthly cost (CHF 65/h)	CHF 2,700–3,800	CHF 1,100–1,600
Operational capacity	1 operator = max 500 invoices	1 operator = 800–1,200 invoices
Client visibility	Delayed monthly report	Real-time dashboard

7 practical tips for go-live

Start with a single pilot mandate of medium volume (200–400 invoices/month) to calibrate AI thresholds and train the team. After 2–3 months of run-in, extend to other mandates
Set up the supplier master data completely from day one: exact company name, VAT number, IBAN and default ledger account. Every configured supplier is a supplier that will no longer require manual intervention
Set conservative AI confidence thresholds at the start (90% for auto-booking) and lower them gradually as the system learns. Better to validate a few extra invoices than to correct errors afterwards
Ask your clients to forward invoices to a dedicated email address (e.g. [email protected]) instead of accumulating them in folders. Automatic import eliminates the manual upload step
Activate the bank connection (Open Banking API or daily CAMT.053 import) from day one: automatic reconciliation only makes sense if bank transactions are updated daily
Plan 30 minutes per day (not 4 hours on Friday) for validating queued invoices. Processing in small batches keeps the flow constant and avoids stressful backlogs
Use AccountEX to manage the entire workflow on a single platform: upload, OCR, AI categorisation, validation, posting and bank reconciliation — without manual exports between different systems

Simplify your Swiss accounting

AccountEX handles VAT, QR-invoices and bookings with AI. Start for free.

Start Free