Help › Purchasing › Incoming Invoice OCR

Incoming Invoice OCR — A Guide from Scratch

Typing every supplier invoice line by line takes time and causes mistakes. Incoming Invoice OCR reads the invoice PDF, photo or e-Fatura/e-Faktur XML and prepares a filled-in purchase invoice draft for you. You only check it, correct it if needed and approve it.

1. What does it do?

1. Reads: the supplier invoice PDF, photo or e-Fatura/e-Faktur XML is uploaded.
2. Prepares a draft: the header (supplier, invoice no, date, due date, amounts) and the items are extracted; every field has a confidence score.
3. You check it: you correct the fields and the stock matches.
4. You approve it: a normal purchase invoice is written. The final number comes from the AF document series; withholding, VAT and amounts are calculated on the server along the invoice's normal path.

2. How is the document read?

Reading goes from the most accurate to the weakest source; the first path that works is used:

1. XML — if the file itself is an e-Fatura (UBL) or e-Faktur XML, or a UBL attachment is embedded in the PDF, it is read directly. Confidence 99%.
2. PDF text layer — the text of computer-generated PDFs is read with Poppler pdftotext; table columns are found from the headings.
3. Image OCR — a photo or scanned PDF is read with the built-in Windows OCR engine. If there is no engine, a "manual entry" draft opens: the document image stays alongside and you fill in the fields.

Limits: the file can be at most 10 MB; PDF, JPG, PNG or XML; in a multi-page document the first 20 pages are processed. Processing that takes longer than 30 seconds continues in the background; the document appears in the list as Processing and the draft opens by itself when it is done.

3. Uploading an invoice

In the dock the window sits in the Purchasing group under the name Incoming Invoice OCR. Drag and drop the file onto the window or use the Choose file button. If the same file is uploaded a second time, no new draft is created; the existing draft for that file opens. In the Documents list on the left you can filter drafts by status (Processing, Awaiting approval, Approved, Rejected, Could not be read) and search by invoice no, supplier or file name.

4. The draft screen

The document itself is on the left and the information read from it is on the right; you check by comparing them side by side.

• Document image: the PDF page or photo; you can move between pages. Show recognized text shows the plain text the program extracted from the document.
• Supplier: the account card is found automatically from the VKN/NPWP on the document. If the same tax ID exists on several account cards, the one closest to the name on the document, or the one with the most recent purchase, is suggested and its confidence is shown in yellow; if needed you pick another card from the search.
• Invoice details: supplier invoice no, invoice date, due date, tax ID, currency, subtotal, VAT total and grand total. A field you correct gets the "corrected" mark.
• Item table: the description on the invoice, the stock card picker (candidates and search), quantity, unit, unit price, discount %, VAT %, amount and reading confidence. You leave a line out with Remove and bring it back with Undo; + Add item adds a line that was not read.
• Totals strip: the grand total calculated from the items and the grand total on the document side by side; it says "Totals match" or "do not match".
• Warnings: situations such as line arithmetic not matching, VAT-inclusive price converted to exclusive, unit converted to the main unit, withholding on the document, or a possible duplicate invoice are listed above the draft.

5. Stock card matching

For each item the stock card is searched in this order; the first match is used:

1. Supplier template — a match you approved before (see section 8).
2. Supplier's product code = stock code — if the code on the line is exactly a stock code.
3. Part number — if a part number in the description (e.g. 0CW398051) exists on a stock card.
4. Previous purchases — similarity to the descriptions of items bought from the supplier before.
5. Name similarity — similarity to the stock name; it is accepted only if the best candidate clearly stands out from the second, otherwise the line stays empty.

A stock card is required on every item: approval is refused while any line has no card selected. Quantity and price are converted to the card's main unit (e.g. 2 boxes on the invoice, 1 box = 12 pcs on the card gives 24 pcs and the per-piece price); a warning appears on the line when a conversion is made.

6. Open purchase order suggestion (3-way matching)

The supplier's open purchase orders are compared with the draft's items; a matching order is suggested in a box. Badges:

• Coverage % — how much of the draft's items are in the order.
• More than the order balance — the quantity on the invoice exceeds the quantity left on the order.
• Price difference — the price on the invoice differs from the order price.

If Invoice from order is selected, the order's remaining quantity is reduced, the order link is written to the invoice, and the actual prices on the invoice are written to it; any price difference appears in the Purchasing › Invoice matching queue and is decided there. If it is not selected (Write without order), a normal purchase invoice not linked to an order is written.

7. Approve, reject and delete

The Approve → purchase invoice button writes the invoice and opens the invoice card. "Supplier invoice no: …" is written to the invoice description, so you can also search by the supplier's own number. If any field is missing or wrong before approval (supplier, invoice date, at least one item, a stock card and a quantity above zero on every item), approval is refused and the gaps are shown.

• Reject — the draft is closed with a rejection reason; no invoice is written.
• Delete — deletes the draft and the uploaded file. An approved draft cannot be deleted; you must delete the invoice first.
• Read again — processes the document from scratch (with the current supplier template); your manual corrections are deleted, so it asks for confirmation.

8. Supplier templates

When you approve a draft, the program learns automatically for that supplier. On the same supplier's second invoice the matches come from the template and confidence goes up. The rules learned:

• Description → stock: which stock card a line description on the invoice corresponds to.
• Supplier code → stock: which stock code the supplier's product code corresponds to.
• Unit mapping, VAT-inclusive price (whether the supplier writes prices including VAT) and the invoice no pattern.

The rules are shown in the window's Supplier templates tab after choosing a supplier; rules can be added or deleted there. Changing a template requires approval permission.

9. On the phone (Garden + Steps)

On screens under 7 inches the window opens as a three-step wizard; there is also a Drafts list at the top:

1. Capture — take a photo of the invoice with the camera or choose a PDF/XML. The browser requires an HTTPS address for the camera; ask your administrator for the HTTPS address provided for phones.
2. Check — supplier, invoice details and items are shown card by card; you correct and save.
3. Approve — if you have approval permission, the purchase invoice is written; otherwise the draft stays as awaiting approval.

10. "Queued" in Pulse

The Queued section of the Pulse main screen shows the number of invoice drafts awaiting approval; tapping it opens the window. If a draft has been waiting for more than 24 hours or a document could not be read, the row is shown in red.

11. Permissions

Under User management › Incoming invoice OCR there are two permissions:

Approval also requires the user's purchase invoice entry permission; the invoice is written through the normal invoice path with the user's own session. Role templates: Purchasing and Accounting get both permissions, Finance only the first.

12. Common problems

• "No image OCR" badge: the server has no image OCR engine; the PDF text layer and XML are still read, photos open as manual entry drafts. Tell your administrator.
• Totals do not match: usually a line is missing or was read twice. Compare line by line with the document image; add the missing line with + Add item and leave the extra one out with Remove.
• Turkish character errors in photos (ı, ş, ğ, İ): the Turkish OCR language pack is not installed; it can be installed with the command in section 2. The PDF text layer does not have this problem.
• The most accurate result: if the e-Fatura PDF has an XML attachment inside (most e-Fatura portals add one), that is read directly and the result comes with 99% confidence; if possible, ask the supplier for the XML or a PDF with the XML attached.

Related guides: Purchasing · Document Series · Document Entry · Mobile Interfaces