Every week, someone on your team opens a supplier invoice, squints at the totals, and retypes them into the accounting system — supplier name, date, each line item, the VAT, the grand total. It's slow, it's error-prone, and it's the kind of work nobody enjoys. A transposed digit here, a wrong tax code there, and the month-end reconciliation turns into a hunt.
BIZA's AI document capture removes the retyping step. You upload a PDF or a photo of the invoice, the system reads it, you check what it found, and you post it to accounts payable. The document does the data entry; you do the reviewing. This post walks through exactly what happens between the upload and the posted purchase entry.
What "capture" actually means
When people hear "AI reads my invoice," they often picture optical character recognition (OCR) — software that turns the pixels of a scanned page back into text. OCR is part of it, but it's the easy part. The hard part is understanding: knowing that the number near the bottom labelled "VAT 15%" is the tax amount and not a line item, that "Riyadh Steel Trading Est." is the supplier and not a product, and that the table in the middle is a set of line items each with its own quantity, unit price, and total.
That's what capture does. It takes the raw text and structure of the document and maps it onto the fields BIZA needs to create a purchase entry: supplier, invoice number, invoice date, currency, each line, the tax breakdown, and the totals.
Step 1 — Upload
From the Procurement or Accounts Payable area, you choose New purchase entry → Capture from document and drop in a file. BIZA accepts:
- PDFs (the most common — supplier invoices emailed as attachments)
- Photos taken on a phone (JPG/PNG), including slightly skewed or shadowed shots
- Scans
A clear, flat image helps, but you don't need a perfect scan. A photo taken on a phone in decent light is usually enough.
Step 2 — Extraction
Once uploaded, the document is sent for extraction. Within a few seconds you get back a structured result. Conceptually, the data BIZA pulls out looks like this:
{
"supplier_name": "Riyadh Steel Trading Est.",
"invoice_number": "INV-2026-0481",
"invoice_date": "2026-06-10",
"currency": "SAR",
"line_items": [
{ "description": "Rebar 12mm", "quantity": 40, "unit_price": 95.00, "line_total": 3800.00 },
{ "description": "Binding wire 20kg", "quantity": 6, "unit_price": 45.00, "line_total": 270.00 }
],
"subtotal": 4070.00,
"vat_rate": 0.15,
"vat_amount": 610.50,
"total": 4680.50
}
Notice the VAT is captured as a rate and an amount, separately from the subtotal. For Saudi invoices this matters: the 15% standard rate has to land in the right place so your input VAT is recoverable and your return reconciles cleanly. Capture reads the tax line off the document rather than recalculating it, then BIZA checks the arithmetic — if the printed VAT doesn't match the rate applied to the subtotal, it flags the difference instead of silently "fixing" it.
Step 3 — Review
This is the step that keeps the process honest. The extracted data lands in a review screen, side by side with the original document, so you can compare what the AI read against what's on the page.
A few things you'll typically check:
- Supplier match. BIZA tries to match the extracted supplier name to an existing record. If it's a new supplier, you can create one on the spot; if there are two similar names, you pick the right one.
- Line items. Each line maps to a quantity, unit price, and total. Multi-line invoices come through as multiple lines — you're confirming, not retyping.
- Account coding. You assign each line (or the whole entry) to the right expense or inventory account. BIZA can suggest coding based on the supplier's history, but the choice stays yours.
- VAT. Confirm the tax code and that input VAT is being claimed where it should be.
Anything the AI was unsure about is highlighted so your eyes go there first. The goal isn't blind trust — it's letting you spend ten seconds verifying instead of two minutes typing.
Step 4 — Post
When the entry looks right, you post it. That creates the purchase entry and the matching accounts-payable liability, the input VAT is recorded for your return, and — if the line is an inventory item — stock levels update. The original file stays attached to the entry, so the source document travels with the transaction. When an auditor or a colleague asks "where did this number come from?", the answer is one click away.
Where it helps most, and where to stay careful
Capture earns its keep on volume and repetition: the recurring utility bill, the regular materials supplier, the stack of receipts after a site visit. The more consistent the format, the faster the review.
It's worth a closer look when:
- The document is a poor photo — blurry, cropped, or heavily creased. Garbage in still means a careful review.
- The invoice is handwritten or uses an unusual layout. Extraction may need more correction.
- The amounts are large or unusual. Treat the review step as a real control, not a formality.
The honest framing is this: AI capture turns invoice entry from typing into checking. You stay in control of every posting — but you stop being the keyboard.
The bottom line
Going from a photo of a supplier invoice to a posted purchase entry used to be a few minutes of careful typing per document. With capture, it's an upload, a quick review, and a post — often under a minute, with the source document filed against the transaction for good measure. For a small team processing dozens of invoices a week, that's hours back, and fewer typos to chase at month-end.