Invoice Batch Validator: PDF & Image Invoices to Checked JSON + CSV

Turn a batch of supplier invoices (PDFs, scans or photos) into structured JSON and CSV, and let the Actor recompute the arithmetic so you only look at the invoices that don't add up.

What you can do

• Extract every invoice field: vendor, customer, invoice number, dates, currency, every line item, subtotal, discounts, shipping, tax and total.
• Catch math errors automatically: line quantity × price, line sums, tax rates, totals and amount due are recomputed and compared with what is printed.
• Catch misreads: every extracted amount is cross-checked against the recognised document text, so a misread number is flagged instead of silently accepted.
• Enforce your rules: expected currency, allowed VAT rates, duplicates within the batch.
• Get spreadsheet-ready files: invoices.csv (one row per invoice) and line-items.csv (one row per line item).
• Automate it: call it from the API, a schedule, n8n, Make, Zapier or an AI agent via Apify MCP.

Checks and flags

Flag	Severity	Meaning
LINE_MATH_MISMATCH	error	quantity × unit price ≠ line amount
LINE_TAX_MISMATCH	error	line tax ≠ line amount × line tax rate
LINE_SUM_MISMATCH	error	line amounts do not add up to the subtotal
TAX_RATE_MISMATCH	error	VAT summary: base × rate ≠ tax amount
TAX_BREAKDOWN_MISMATCH / LINE_TAX_SUM_MISMATCH	error	tax lines do not add up to the tax total
TOTAL_MISMATCH	error	subtotal − discount + shipping + tax ≠ total
AMOUNT_DUE_MISMATCH	error	total + previous balance − paid ≠ amount due
LINE_VALUES_NOT_AS_PRINTED	error	an extracted quantity, price or amount does not match the numbers printed in that table row
VALUE_NOT_IN_DOCUMENT	error	an extracted subtotal, tax total, total or amount due does not appear anywhere in the document text
DUPLICATE_IN_BATCH	error	same vendor, invoice number and total as an earlier invoice in the run
CURRENCY_MISMATCH	error	currency differs from expectedCurrency
UNEXPECTED_TAX_RATE	error	a tax rate not in allowedTaxRates
DUE_DATE_BEFORE_INVOICE_DATE, INVALID_INVOICE_DATE	error	date problems
MISSING_INVOICE_NUMBER, MISSING_INVOICE_DATE, MISSING_VENDOR, MISSING_TOTAL	error	required field not found
MISSING_CURRENCY, NO_LINE_ITEMS, NOT_AN_INVOICE, INVOICE_DATE_IN_FUTURE, INVALID_DUE_DATE	warning	worth a look
PAGE_CAP_REACHED, TEXT_TRUNCATED	warning	the document was longer than the limits below

Amounts are compared with a tolerance of amountTolerance (default 0.01), plus half a cent per unit on quantity × price to allow for rounded unit prices. An invoice with any error flag gets status: "needs_review"; otherwise status: "valid".

How to use

1. Put direct links to your invoice files in Invoice URLs — one invoice per file (PDF, JPG, PNG, WEBP, GIF, TIFF or AVIF). The links must download without a login: an Apify key-value store record, a pre-signed S3 link or a public file host all work.
2. Optionally set the expected currency, the allowed VAT rates and the tolerance.
3. Start the run. Each invoice becomes one row in the Output tab; the CSV files are linked there too.

A one-page invoice takes about 5–10 seconds; up to 3 invoices are processed in parallel.

Field	Required	Description
invoiceUrls	yes	Direct, publicly downloadable links. Up to 1,000 per run; repeated links are processed once.
maxPagesPerInvoice	no	1–5, default 5. Only the first N pages of each PDF are read.
expectedCurrency	no	ISO code such as EUR.
allowedTaxRates	no	Tax/VAT rates in percent, e.g. 20, 10, 5.5, 0.
amountTolerance	no	Default 0.01.
includeOcrText	no	Adds the recognised page text to each item for auditing.

Pricing

You pay per processed invoice. Failed invoices (unreadable, not downloadable) are free.

Event	Price
Invoice processed (up to 5 pages)	$0.05

Example run	Cost
1 invoice	$0.05
100 invoices (a month of supplier bills)	$5.00
1,000 invoices (maximum per run)	$50.00

Before processing, the Actor checks that your run's maximum cost covers the whole batch ($0.05 × number of invoices). If it doesn't, the run stops without charging anything and tells you the amount to set.

Available on paid Apify plans.

Input example

{
  "invoiceUrls": [
    "https://example.com/invoices/INV-100.pdf",
    "https://example.com/scans/receipt-7.jpg"
  ],
  "expectedCurrency": "USD",
  "allowedTaxRates": ["10"]
}

Output example

One dataset item per invoice:

{
  "index": 0,
  "sourceUrl": "https://example.com/invoices/INV-100.pdf",
  "status": "needs_review",
  "flagCodes": ["LINE_MATH_MISMATCH"],
  "flags": [{ "code": "LINE_MATH_MISMATCH", "severity": "error", "message": "Line 1: 1 x 1 = 1.00, but the line shows 100.00." }],
  "vendorName": "Example Supplies Ltd.",
  "invoiceNumber": "INV-100",
  "invoiceDate": "2026-09-15",
  "currency": "USD",
  "total": 110,
  "amountDue": 610,
  "invoice": {
    "document_type": "invoice",
    "invoice_number": "INV-100",
    "invoice_date": "2026-09-15",
    "due_date": "2026-10-15",
    "currency": "USD",
    "vendor": { "name": "Example Supplies Ltd.", "address": "1 Sample Street, Springfield", "tax_id": null },
    "customer": { "name": "Demo Customer Inc.", "address": "99 Test Avenue, Riverside", "tax_id": null },
    "line_items": [{ "description": "Consulting services", "quantity": 1, "unit_price": 1, "amount": 100, "...": "..." }],
    "subtotal": 100, "tax_total": 10, "total": 110, "previous_balance": 500, "amount_due": 610, "...": "..."
  },
  "checks": { "lineItemsSum": 100, "lineTaxSum": null, "computedTotal": 110, "computedAmountDue": 610, "impliedTaxRate": 10 },
  "pagesProcessed": 1,
  "charged": true,
  "error": null
}

The run's key-value store also gets:

• invoices.csv — one row per invoice (header fields, totals, status, flag codes)
• line-items.csv — one row per line item, keyed by source URL and invoice number
• OUTPUT — run summary with counts and links to both CSV files

Files that cannot be downloaded or read come back with status: "failed" and an error message, and are not charged.

Limits

• Up to 5 pages per invoice. Longer PDFs are cut after page 5 (or after maxPagesPerInvoice) and flagged PAGE_CAP_REACHED. Split multi-invoice PDFs into one file per invoice.
• About 30,000 characters of recognised text per invoice are used for extraction; beyond that the invoice is flagged TEXT_TRUNCATED.
• Up to 1,000 invoices per run, 3 processed in parallel.
• Links must be directly downloadable without a login.

Good to know

• Values are read as printed; the Actor never "corrects" a number, it flags it. Always review needs_review invoices before paying them.
• Handwritten or very low-resolution scans may be read less reliably — the grounding checks will usually flag them.
• Download links to the CSV files expire with your plan's storage retention — save the files you need.
• Questions or a missing check? Open an issue on the Actor page.