How accurate is HIBR ERP's AI Receipt OCR?

On our UAE-specific training set, total-line accuracy is above 98%, vendor-name accuracy above 95%, and VAT-amount accuracy above 97% for receipts above AED 10. Thermal receipts faded by sunlight are harder; HIBR flags low-confidence extractions for human review rather than guessing.

Does the OCR read Arabic receipts properly?

Yes. The model is trained on bilingual Arabic + English UAE receipts and handles Arabic numerals (٠-٩), Western numerals, mixed-language receipts, and right-to-left layout correctly. Most UAE supermarkets, petrol stations, and restaurants print bilingual receipts, and HIBR reads both.

What happens if the AI gets a field wrong?

Every extraction shows the source image alongside the proposed journal entry. You can correct any field in one tap before posting; the correction is captured as feedback that improves accuracy on your future receipts.

AI Receipt OCR — HIBR ERP | Arabic + English Receipt Scan

Try the AI Receipt OCR live

Upload a UAE receipt (Arabic, English, or both). The AI extracts vendor, date, total, VAT, and category in seconds, ready to post.

Open the live demo →

Free. No login. Drop in a JPG or PNG of any receipt.

Anyone who has ever done UAE bookkeeping knows the shoebox problem: a glove compartment full of petrol-station receipts, an envelope of Lulu and Carrefour receipts, dozens of Talabat and Deliveroo PDFs from client lunches, and one day each month spent typing them all into the accounting system, line by line, to get the input VAT recovery the FTA actually allows. HIBR ERP's AI Receipt OCR is the entire end of that workflow. Snap a photo with your phone, or drop a JPG/PNG/PDF into the upload box, or forward an email receipt to a dedicated HIBR inbox. The model — Claude Vision plus a custom UAE-receipt training set — extracts the vendor name, the date, the total, the VAT amount, the line items, and the category, posts a balanced journal entry to the right Chart of Accounts code, attaches the original image to the entry, and tags it for the VAT 201 return. Three seconds, no typing.

How it works

Snap, upload, or forward. Three capture channels: the HIBR mobile app (camera tap), the web UI (drag-and-drop a JPG/PNG/PDF), or a dedicated email address (forward the receipt; HIBR ingests the attachment). Multi-page PDFs are supported; multi-receipt single images are auto-split.
Bilingual Arabic + English recognition. The model handles pure Arabic, pure English, mixed-language (most UAE supermarket receipts), Arabic numerals (٠١٢٣٤٥٦٧٨٩), Western numerals, and right-to-left layouts correctly. Thermal print, faded receipts, and crumpled receipts all work; severely damaged receipts surface as low-confidence for human review.
Field extraction. Vendor name (matched to the supplier master if it exists, created as a new supplier if not), date, subtotal, VAT amount, TRN if present, total, currency, and line items. Categorisation is done by the AI Categorizer using your Chart of Accounts.
Auto-journal posting. A balanced double-entry journal is created automatically — debit the expense account, debit input VAT, credit cash or the payment-method account. Attached to the journal: the original receipt image and the extraction confidence score.
VAT 201 tagging. Each receipt is tagged with the VAT recovery treatment (recoverable, partially recoverable, non-recoverable per Cabinet Decision 52/2017 Article 53 entertainment rules). Non-recoverable items (e.g., entertainment of UAE residents beyond reasonable hospitality) are flagged automatically.
Continuous learning. When you correct an extraction, the correction is captured as feedback for your own profile — HIBR learns that your "Costa Coffee" line should always go to "Staff refreshments" rather than "Client entertainment", and applies the rule going forward.

UAE compliance details

AI Receipt OCR is designed to satisfy the FTA's input-VAT-recovery evidence requirements:

Federal Decree-Law 8/2017 on VAT (Article 53) and Cabinet Decision 52/2017 (Article 53) — input VAT is recoverable only against a valid Tax Invoice. HIBR's OCR captures and attaches the original receipt image as the supporting evidence, retained in the immutable Document Vault.
Cabinet Decision 52/2017 (Article 60) — defines the Simplified Tax Invoice (typical retail receipt) content. HIBR's extraction maps to these fields and validates them.
Federal Decree-Law 28/2022 on Tax Procedures (Article 4) — record retention. Receipt images and the journal entries are retained 7 years on Lite/Pro and 10 years on Enterprise.
Cabinet Decision 52/2017 (Article 53(2)) — input VAT on entertainment of non-employees is not recoverable. HIBR flags entertainment expenses against employee versus non-employee classification.
Federal Decree-Law 47/2022 on Corporate Tax (Article 32) — limits the deduction of entertainment expenditure to 50% for CT purposes. HIBR applies the 50% cap automatically on entertainment-classified receipts.

Tier availability

AI Receipt OCR is included on every tier. Lite (199/month) includes 100 receipts per month — enough for most solo founders. Pro (499/month) includes 1,000 receipts per month. Enterprise (14,990/year) is unlimited receipts plus a dedicated multi-employee submission portal (each employee submits their own receipts to a shared queue for approval).

Use case

Dubai-based event production company, 4.2M revenue, six staff regularly out on shoots. Before HIBR, the finance manager spent the last week of every month typing receipts from a Google Drive folder that the six staff dumped into. Average backlog: 300+ receipts. With HIBR Pro and the staff-submission portal on Enterprise (they upgraded for this single feature), each staff member photographs receipts on the day; HIBR extracts and posts within seconds; the finance manager only reviews exceptions. Average finance manager time on receipts dropped from 22 hours per month to under 2.

Tier mapping. Lite for solo founders. Pro for typical SMBs. Enterprise for businesses with multiple employees submitting receipts. See full pricing.

Frequently asked questions

What happens to receipts the AI can't read confidently?

They land in a "Needs review" queue with the original image, the partial extraction, and the confidence breakdown per field. You complete the missing fields and post manually. Clear, well-lit receipts post cleanly; only a small share — typically faded thermal paper, heavy handwriting, or crumpled images — needs a manual touch.

Can the OCR handle receipts that don't carry a TRN (small vendors, taxi receipts)?

Yes — those are treated as non-VAT-recoverable purchases. The journal is still posted for the expense, the receipt is still archived for CT-deductible-expense evidence, and the input VAT line is omitted (or posted to a non-recoverable account, depending on your CoA).

Does the OCR work for invoices from suppliers, not just retail receipts?

Yes. The same OCR pipeline reads supplier Tax Invoices (full A4 PDFs), simplified Tax Invoices, and proforma invoices. For full Tax Invoices, the OCR also validates the supplier's TRN against the FTA registry (see TRN Validation) before posting.

Open the live demo

Drop in a UAE petrol-station receipt, a supermarket receipt, or a restaurant bill — Arabic, English, or both — and watch the extraction happen.

Open the live demo →

Runs in your browser. Free.

The shoebox problem, solved in three seconds per receipt

Snap, upload, or forward. Bilingual Arabic + English. Auto-journal posting with VAT-201 tagging. Beta launches October 2026.

Join the waitlist →