The problem
Thousands of small Saudi firms must now integrate with ZATCA’s Fatoora platform, but their inbound paperwork is still paper: supplier invoices photographed in warehouses, stamped PDFs, mixed Arabic-English line items. Somebody has to turn that pile into structured fields.
The build
A single API endpoint. Send it an invoice photo or PDF; it returns buyer, seller, VAT number, and line items mapped to ZATCA Phase 2 fields as JSON. Intake over WhatsApp and email for shops with no IT. Two people assemble it from an OCR API, a field-mapping model, and the Fatoora sandbox for validation.
Why now
In March 2025 Mistral released its OCR API at roughly $1 per 1,000 pages with native multilingual support — Arabic table extraction became a commodity call. The same month, ZATCA’s 22nd wave pulled every firm above SAR 1 million turnover into Phase 2 integration, with compliance due by December 2025. Cheap Arabic OCR met a hard regulatory deadline.
First customer
Independent accounting offices in Riyadh and Jeddah onboarding wave-hit SMEs. They already do this conversion by hand and lose money on every rejected filing — a per-invoice API they can resell to fifty clients is an easy yes before it is finished.
Challenges
ZATCA rejects malformed invoices, so anything below near-perfect accuracy becomes a support treadmill. Handwritten Arabic and stamp-covered scans still break extraction. And this is a feature, not a company: every accounting suite adds it as a checkbox within two years, so the window is a wedge, not a moat.