Invoice extraction guide
Purchase invoice OCR in South Africa
OCR can remove repetitive typing, but a valid claim still depends on the document, the transaction and the applicable rules.
What does purchase-invoice OCR do?
Purchase-invoice OCR reads an uploaded image or PDF and converts visible invoice fields into editable data. It can extract supplier and recipient details, invoice number, issue date, descriptions, totals and VAT. It cannot prove that the supply was genuinely for business or independently decide that input VAT is deductible.
Useful fields to extract
| Invoice field | Automated use | Human check |
|---|---|---|
| Supplier name, address and VAT number | Identifies the vendor and populates the record. | Confirm the number and name are legible and belong to the supplier. |
| Recipient details | Checks higher-value full-invoice readiness. | Confirm the business identity is correct where required. |
| Serial number and issue date | Supports duplicate detection and period allocation. | Check ambiguous characters and dates. |
| Description and quantity | Supports category suggestions and document checks. | Confirm the description reflects what was actually bought. |
| Value, VAT and total | Populates the financial fields. | Confirm the totals reconcile and whether VAT is included. |
OCR is transcription, not tax judgment
An invoice can be read perfectly and still fail a claim check. The supplier may not be VAT registered, the document may be missing required particulars, the purchase may be private, or input tax may be denied for the category. Conversely, a low-confidence character on a supplier address should not cause AI to invent the missing text.
A safer workflow separates the jobs: AI extracts; deterministic checks apply known rules; the business owner confirms purpose and corrections; the accountant reviews the result.
Handle invoice tiers before claiming VAT
SARS distinguishes between the supporting document needed for small purchases, abridged tax invoices and full tax invoices. The transaction value affects which fields need to appear. Recipient details are part of the full-invoice requirements, but should not be demanded from every low-value invoice.
A practical review flow
- Upload the original image or PDF without cropping required details.
- Compare the extracted supplier, date, totals and VAT to the document.
- Confirm the business purpose, category and any personal-use portion.
- Resolve missing invoice details instead of letting AI guess.
- Export the corrected record with the original for professional review.
SlipSaver’s approach
SlipSaver uses AI for extraction and a separate claim engine for treatment. Editing a stored slip or purchase invoice re-runs the engine, so the server—not a client-only display—remains responsible for the automated claim flags.