Arabic & English invoice OCR for UAE tax
Extract invoice data from Arabic and English PDFs and images into a structured, reviewable format. Mandatory UAE tax-invoice fields are flagged, duplicates are detected and low-confidence lines land in a review queue.
OCR is not e-invoicing
A PDF or scanned invoice processed by OCR is not a structured e-invoice under the UAE e-invoicing mandate. The e-invoicing regime requires issuance through an accredited service provider on the PINT-AE data model. OCR is useful for AP processing and VAT return preparation, especially during the transition period.
What the OCR extracts
Core fields
- • Supplier name, address and TRN
- • Customer name, address and TRN
- • Invoice number and date
- • Line descriptions, quantities and unit prices
- • Subtotal, VAT amount and total
- • Currency and exchange rate (where shown)
Compliance checks
- • Validate supplier TRN format (does not confirm active status)
- • Check mandatory UAE tax-invoice fields present
- • Flag simplified vs full tax invoice
- • Detect duplicate invoices by invoice number + supplier
- • Flag credit notes and link to original invoice where possible
Review and exceptions
Every extracted invoice comes with a confidence score per field. Low-confidence lines and failed compliance checks route to a human-review queue. We do not auto-post invoices to the ledger or auto-populate a VAT return without review.
Auto-approved
All mandatory fields present, TRN format valid, totals reconcile and confidence above threshold.
Review required
Field below confidence threshold, missing mandatory data, or duplicate match. Queued for human correction.
Unsupported layout
Where extraction fails altogether, the invoice remains in the queue with the raw document for manual processing.
Frequently asked questions
Can it read Arabic invoices?
Yes. Arabic and English invoices are both supported. Mixed-language invoices (Arabic header with English line items, for example) are common in the UAE and handled.
What must I review?
Anything in the review queue: low-confidence fields, missing mandatory data, duplicate matches and unsupported layouts. We publish a single-figure accuracy percentage only when it is tied to a tested dataset; you should validate on your own invoice sample during the trial.
How are duplicates handled?
Potential duplicates (same invoice number and supplier, close total) are flagged for human confirmation rather than auto-merged.
Where is invoice data processed?
See the security page for current hosting, subprocessor and residency details. We only publish what we can evidence.
Try OCR in a verified demo
Bring 10–20 of your own invoices (Arabic, English or mixed) and we will show the full extract-review-post cycle in real time.
Book demo