Documents4 min readBy Ahmed AkramPublished
Can you automate invoice entry without a template for every supplier?
If your team spends the morning typing supplier invoices into the accounting system, you've probably looked at OCR before and been put off by the setup: one template per supplier, and a new one every time a supplier changes their layout. It doesn't have to work that way anymore.
Short answer
Yes. Current document AI reads an invoice by meaning (this is the supplier, this is the total, these are the lines) rather than by fixed positions, so a new layout doesn't need a new template. What makes it safe is the checking around it: recalculated totals and tax, matching against supplier and purchase-order records in your ERP, duplicate detection, and a person reviewing anything the system isn't sure about before it's posted.
Why template OCR keeps breaking
Classic OCR turns the page into text and then relies on a template that says "the invoice number is in this box". That works for one layout. Then a supplier moves their logo, adds a column or sends a photo taken on a phone, and the template reads the wrong box without telling anyone.
So teams end up maintaining dozens of templates, or giving up and typing. Neither is what anyone wanted.
How template-free extraction works
A document AI model reads the whole invoice the way a person would and returns structured fields: supplier, invoice number, dates, line items, tax and totals, each with a confidence score. It doesn't care where on the page the total sits.
Capture
Invoices arrive by email, upload or scan, as PDFs or photos.
Read
The AI extracts supplier, dates, line items, tax and totals, in Arabic or English.
Check
Totals and tax are recalculated, and the supplier, PO and duplicates are checked against your ERP.
Review
Anything uncertain goes to a person, with the original next to the extracted fields.
Post
Approved invoices are created in your ERP or accounting system, ready for the payment run.
Getting the details into your ERP
Reading the invoice is half the job. The other half is writing to your ERP properly: creating a draft vendor bill through the system's API (Odoo, SAP Business One, Dynamics 365, Zoho and most others have one), linking it to the right supplier and purchase order, and leaving it in draft until someone approves it, at least for the first few months.
Where you have purchase orders and goods receipts, the system can match all three and flag the differences. That check is where a lot of the savings come from, because it catches overbilling that manual entry tends to miss.
| Manual entry | Template OCR | Document AI with checks | |
|---|---|---|---|
| A new supplier layout | The same typing as always | Needs a new template | Read without setup |
| Scans and phone photos | Slow | Often fails | Read, with review if unclear |
| Totals and tax checked | By eye | Rarely | On every invoice |
| Getting it into the ERP | Typed by hand | Export, then import | Draft bill created for approval |
Manual entry
- A new supplier layout
- The same typing as always
- Scans and phone photos
- Slow
- Totals and tax checked
- By eye
- Getting it into the ERP
- Typed by hand
Template OCR
- A new supplier layout
- Needs a new template
- Scans and phone photos
- Often fails
- Totals and tax checked
- Rarely
- Getting it into the ERP
- Export, then import
Document AI with checks
- A new supplier layout
- Read without setup
- Scans and phone photos
- Read, with review if unclear
- Totals and tax checked
- On every invoice
- Getting it into the ERP
- Draft bill created for approval
Is it worth it at 20 invoices a day?
Do the math with your own numbers. Take the invoices you handle in a month, multiply by the minutes each one takes to type and check, and you have the hours on the table. Long invoices with many lines take far longer than people guess.
Then test it before you believe anyone, us included. A pilot on last month's real invoices shows how many fields the system reads correctly and how many still need a person. That number decides whether it's worth it, not a vendor's accuracy claim.
Questions people also ask
How accurate is AI invoice extraction?
It depends on your documents: clean PDFs read very well, faded scans less so. We measure accuracy on a sample of your real invoices before go-live, and anything the system isn't sure about goes to a person instead of straight into your books.
Can it read Arabic invoices?
Yes. It reads Arabic and English invoices, including ones that mix both, and handles Arabic-Indic digits. We include Arabic samples from your own suppliers in the pilot.
Will it post wrong numbers to our accounts?
Not when it's set up properly. Invoices are created as drafts, totals are recalculated, and anything that fails a check waits for a person. Nothing is posted until it passes the rules you agree.
Related service
Send us last month's invoices and see what it reads.
Talk to usMore articles
- DocumentsIs your data ready for AI?A practical checklist: where your data lives, how current it is, who can see it, and what to fix first.Read the article
- BlockchainAI and blockchain by 2030Where AI and blockchain meet in real business use over the next five years, and which loud claims are safe to ignore.Read the article
- AI agentsAI in business by 2030Honest predictions for the next few years: what we are fairly sure of, what is uncertain, and what to do this year.Read the article

