Injaz Digital Become a partner

Automation

Reading invoices and bills of lading with AI, not manual entry

Updated 16/10/2026 · 6 min read · Injaz Digital team

The short answer

AI reading understands an invoice or bill of lading the way an accountant does: it extracts the supplier, items, quantities, tax, container numbers and HS codes from a PDF or phone photo, then checks that subtotal plus tax equals the total before entry. Injaz AI by Injaz Digital builds it on your email, WhatsApp and existing accounting system.

In import and wholesale companies the same question keeps coming: “is there a program that reads the invoice from the photo and enters it into the system?”. Invoices arrive as a PDF from a supplier, a phone photo from a rep, and a scanned bill of lading from the freight agent, and the accountant copies them all by hand. Injaz AI by Injaz Digital ends that copying and leaves the accountant the checking and the decisions.

Why does classic OCR fail on Arabic invoices and phone photos?

Classic OCR fails because it reads characters without understanding the document: it turns the image into text, then relies on a fixed template that says where each field sits. If the invoice layout changes, the table is nested, or the photo is skewed, the columns get mixed. Arabic makes it harder: connected letters, Arabic and Hindi numerals on the same invoice, and item names in two languages on one line.

CaseClassic template OCRAI reading
A new supplier with a different layoutA new template by handUnderstands fields by meaning, not position
A nested item table or one spanning two pagesCut or merged linesRebuilds each line: item, quantity, price, tax
A skewed phone photo or shadowWrong numbers with no warningA confidence score per field, weak ones to review
Arabic and English in one documentReversed word orderReads both and keeps each name as written

What does AI extract from an invoice and a bill of lading?

AI extracts every field accounting or clearance needs from the document, into a structured record that goes straight into the system. Microsoft's documentation for its invoice model, for example, says it extracts key fields and line items from invoices, sales orders and purchase orders, whether phone-captured images, scans or digital PDFs.

  • From the supplier invoice: supplier name and VAT number, invoice number and date, items, quantities and unit prices, tax and the total.
  • From the bill of lading: shipper and consignee, ports of loading and discharge, container numbers, cargo description, package count and weight.
  • From the commercial invoice and packing list: each item's HS code, country of origin and value.

The HS code is the Harmonized System run by the World Customs Organization: a six-digit code for over 5,000 commodity groups, used by more than 200 countries and economies as the basis for customs tariffs. Extracting it correctly from the invoice saves the clearance team looking it up by hand.

How does the system check before it enters any number?

The system checks the document with arithmetic and logic rules before entry, and any doubt goes to the accountant rather than being guessed. These rules are what Injaz AI builds into every reading flow:

  1. The line subtotal before tax plus the tax equals the invoice total; otherwise the invoice stops.
  2. Quantity times unit price equals each line's value.
  3. The supplier's VAT number exists in your supplier records, and the invoice number hasn't been entered before, so no invoice is recorded twice.
  4. Container numbers on the bill of lading match the packing list and the commercial invoice.
  5. Every low-confidence field or unclear line is shown to the accountant next to the document image, and entered after approval.

Which documents does customs require, and how are they read together?

Saudi Arabia's Zakat, Tax and Customs Authority asks importers to present three documents before import: the commercial invoice, the bill of lading, and the certificate of origin (not needed when the country of origin is clearly established). Reading them together is the real value: the invoice quantity is compared with the packages on the bill of lading and packing list, and any difference is caught before the shipment arrives, not at clearance. That comparison is part of what Injaz AI builds for import and export companies.

Where do documents arrive, and how are they connected to accounting?

Documents arrive from four places, and the automation picks them up from all of them: the invoices mailbox, WhatsApp messages, a shared Google Drive or OneDrive folder, and supplier portals. On WhatsApp in particular they must be saved at once: Meta's documentation states that media IDs in webhook notifications expire after 7 days, so a photo that isn't saved is lost.

After reading and checking, Injaz AI enters the data into your accounting software or ERP through its API: a draft purchase bill, an entry against the purchase order, or a completed shipment file. The original document image stays attached to the entry, so the auditor can return to it at any time.

How does Injaz AI build document reading on your systems?

Injaz AI builds document reading on the systems and channels you use today, starting with your most frequent document. We collect a sample of your real invoices and bills of lading, tune the reading and checks on them, and run it alongside manual entry until the results match; then staff move from entry to review. The flow runs on your server, with accounts in your company's name.

  • What disappears: copying invoices and bills of lading line by line, looking up HS codes, and re-entering duplicate invoices.
  • What the accountant keeps: reviewing and approving weak fields and exceptions.
  • What management gets: every entry attached to its document image, and a log of who reviewed it and when.

Why choose Injaz AI by Injaz Digital to automate your business?

  • We connect the systems you already have as they are: your CRM, accounting, spreadsheets, WhatsApp and invoicing, without replacing them.
  • We automate repetitive work end to end, so your team is freed from manual entry, reports and moving data around.
  • Human review where it matters, a log of every step, and accounts and permissions in your name.
  • From Injaz Digital: over 15 years in the Arab market, with offices in Cairo, Dubai and Riyadh.

Talk to us about automation

FAQ

Does it read handwritten invoices?

It reads many of them, but handwriting is less accurate, so its fields always go to the accountant for review before entry.

Does an e-invoice need reading?

In ZATCA's integration phase an e-invoice is issued as XML or PDF/A-3 with embedded XML, so we read its data directly without OCR. AI reading is for paper invoices, photos and PDFs from suppliers abroad.

What happens to a document the system can't read?

It goes to the accountant with a clear reason, and nothing from it is entered. No guessing on numbers.

Where are documents stored?

Where they already are: your email, folders and system, with the flow running on your server.