Most invoice automation comes down to the same five steps: catch the invoice where it arrives, extract the fields, check them, write them to your system of record, and send anything unusual to a person. This guide builds that flow three times, in Zapier, Make and n8n, using the PDF Vector invoice actions that each platform actually offers. It also covers the JSON Schema to use, the validation step most tutorials skip, and what each invoice costs in credits.
What the PDF Vector integrations can do with invoices
The PDF Vector app is available on Zapier, Make and n8n. For invoices, each platform exposes the same three operations under slightly different names:
| Operation | Zapier action | Make module | n8n node (resource: Invoice) | Credits per page |
|---|---|---|---|---|
| Parse | Parse Invoice | Parse an invoice | Parse | pro 6, max 10 |
| Extract | Extract Data From Invoice | Extract structured data from an invoice | Extract | nano 6, mini 10, pro 14, max 18 |
| Ask | Ask About Invoice | Ask an invoice | Ask | nano 6, mini 10, pro 14, max 18 |
- Extract is the one to use for automation. You give it a prompt and a JSON Schema, and it returns
datain exactly that shape, so later steps can map fields reliably. - Parse returns the invoice as Markdown. Use it when you want to archive a readable copy or pass the text to another AI step.
- Ask answers one question, such as "What are the payment terms?" It is useful for a quick lookup, not for structured data.
All three take the invoice either as a public URL or as a file from an earlier step, and default to the auto model. PDF Vector provides actions only, not triggers, so every flow starts with a trigger from your email, storage or form app.
The invoice schema
Every flow below uses the same schema. Keeping it identical across tools makes it easy to switch platforms later.
{
"type": "object",
"properties": {
"vendorName": { "type": "string" },
"vendorTaxId": { "type": "string" },
"invoiceNumber": { "type": "string" },
"invoiceDate": { "type": "string", "description": "YYYY-MM-DD" },
"dueDate": { "type": "string", "description": "YYYY-MM-DD" },
"currency": { "type": "string", "description": "ISO 4217 code" },
"subtotal": { "type": "number" },
"tax": { "type": "number" },
"total": { "type": "number" },
"lineItems": {
"type": "array",
"items": {
"type": "object",
"properties": {
"description": { "type": "string" },
"quantity": { "type": "number" },
"unitPrice": { "type": "number" },
"amount": { "type": "number" }
}
}
}
},
"required": ["vendorName", "invoiceNumber", "invoiceDate", "total"]
}
A matching prompt: "Extract the vendor, invoice number, dates, currency, subtotal, tax, total and every line item. Dates as YYYY-MM-DD. Amounts as plain numbers without currency symbols. If there is no tax line, use 0."
You can build or edit the schema visually in the free JSON Schema Editor and paste the result into the JSON Schema field of any of the three tools.
Flow 1: Zapier, from email attachment to Google Sheets
Goal: every PDF invoice sent to invoices@yourcompany.com becomes a row in a spreadsheet, and anything that fails validation goes to Slack.
- Trigger: Gmail, New Attachment. Limit it with a search such as
to:invoices@yourcompany.com has:attachment filename:pdf. - Action: PDF Vector, Extract Data From Invoice. Connect your PDF Vector API key. Map the attachment to the File field (or use Document URL if your trigger gives a link). Paste the prompt into Extraction Prompt and the schema into JSON Schema. Leave Model on Auto to start.
- Action: Code by Zapier, Run JavaScript. Map
subtotal,tax,total,invoiceNumberandvendorNamefrom step 2 as inputs, and run the check below. - Paths by Zapier. Path A continues when
validistrue; path B when it isfalse. - Path A: Google Sheets, Create Spreadsheet Row. Map the extracted fields to columns.
- Path B: Slack, Send Channel Message. Include the vendor, invoice number and the
reasonfrom step 3, plus a link to the email.
The Code step:
const num = (v) => (v === undefined || v === null || v === "" ? null : Number(v));
const subtotal = num(inputData.subtotal);
const tax = num(inputData.tax) ?? 0;
const total = num(inputData.total);
const reasons = [];
if (!inputData.invoiceNumber) reasons.push("missing invoice number");
if (!inputData.vendorName) reasons.push("missing vendor");
if (total === null || Number.isNaN(total)) reasons.push("missing total");
if (subtotal !== null && total !== null && Math.abs(subtotal + tax - total) > 0.01) {
reasons.push(`subtotal ${subtotal} + tax ${tax} != total ${total}`);
}
output = { valid: reasons.length === 0, reason: reasons.join("; ") };
The check uses scalar fields on purpose. Zapier flattens arrays such as lineItems into separate line-item fields, which is fine for mapping into a spreadsheet but awkward to sum in a Code step.
Flow 2: Make, from a Google Drive folder to an accounting sheet
Goal: invoices dropped into a shared Drive folder are extracted, checked line by line, and recorded with one row per line item.
- Google Drive, Watch Files in a Folder. Point it at your "Invoices / Inbox" folder.
- Google Drive, Download a File. Map the file ID from module 1.
- PDF Vector, Extract structured data from an invoice. Map the downloaded data to File and the name to File Name. Paste the prompt and schema.
- Router with two routes:
- Valid route with a filter:
totalexists,invoiceNumberexists, andsubtotal + taxequalstotal(use Make'sround()andabs()functions to allow a 0.01 difference). - Review route as the fallback.
- Valid route with a filter:
- Valid route: Iterator over
data.lineItems, then Google Sheets, Add a Row for each line item, carrying the invoice number and vendor on every row. - Review route: Slack, Create a Message (or email) with the invoice details, and Google Drive, Move a File to an "Invoices / Needs review" folder.
Make's Router and Iterator make line-item checks easier than in Zapier. If you also want to check that line items add up to the subtotal, add an Array aggregator or use sum(map(data.lineItems; "amount")) in the filter.
Flow 3: n8n, from an inbox to your accounting system
Goal: the same flow, self-hosted or on n8n Cloud, with the validation written in code.
- Gmail Trigger (or Email Trigger (IMAP)) with attachment download turned on. Note the binary property name it produces, for example
attachment_0. - PDF Vector node. Resource Invoice, operation Extract. Set Input Type to Binary Data and Input Binary Field to the property from step 1. Paste the prompt and schema. Optionally map the email message ID into Document ID so usage can be traced back to the message.
- Code node (JavaScript, "Run Once for All Items"):
const round2 = (n) => Math.round(n * 100) / 100;
return $input.all().map((item) => {
const inv = item.json.data ?? {};
const reasons = [];
const items = Array.isArray(inv.lineItems) ? inv.lineItems : [];
const lineSum = round2(items.reduce((s, li) => s + (Number(li.amount) || 0), 0));
const tax = Number(inv.tax) || 0;
if (!inv.invoiceNumber) reasons.push("missing invoice number");
if (typeof inv.total !== "number") reasons.push("missing total");
if (typeof inv.subtotal === "number" && Math.abs(round2(inv.subtotal + tax) - inv.total) > 0.01) {
reasons.push(`subtotal + tax != total (${inv.subtotal} + ${tax} vs ${inv.total})`);
}
if (items.length && typeof inv.subtotal === "number" && Math.abs(lineSum - inv.subtotal) > 0.01) {
reasons.push(`line items sum to ${lineSum}, subtotal is ${inv.subtotal}`);
}
return { json: { ...inv, valid: reasons.length === 0, reasons, credits: item.json.credits } };
});
- IF node on
valid. - True branch: write to your system: a Google Sheets append, a QuickBooks Online node to create a bill, or an HTTP Request to your own API.
- False branch: a Slack message or a task in your tracker with the
reasonsarray.
Since version 2.1.13 of the n8n node, API errors are routed to the node's error output when you set "On Error" to continue using the error output, so you can send failed extractions to the review branch too.
Validation rules worth adding
The arithmetic checks above catch most extraction errors. These catch the rest:
- Duplicate invoices. Look up
vendorNameplusinvoiceNumberin your sheet or accounting system before inserting. Suppliers resend invoices, and email rules sometimes fire twice. - Dates. The invoice date should not be in the future, and the due date should not be before the invoice date.
- Currency. Flag anything not in the currencies you pay in.
- Known vendors. Compare the vendor name or tax ID against your supplier list. New vendors deserve a human look, both for accuracy and for fraud.
- Thresholds. Route invoices above an amount you set to approval, even when they pass every check.
If an invoice fails the arithmetic checks, a reasonable first step is to run Extract again on a higher model (pro or max) before sending it to a person.
What it costs
Credits are charged per page, so a typical one-page invoice costs:
- Extract or Ask: 6 credits on nano, 10 on mini, 14 on pro, 18 on max.
- Parse: 6 credits on pro, 10 on max.
With auto, you are billed for the tier it selects. The free plan's 100 credits per month cover 5 single-page invoices on max, or 16 on nano, which is enough to build and test a flow. Basic includes 3,000 credits for $25 per month, and Pro includes 100,000 credits for $97 per month. See pricing for annual rates. Requests that fail are not charged. Remember that Zapier tasks and Make operations are billed separately by those platforms.
Developers who would rather call the API directly can use the same operation over REST at POST https://global.pdfvector.com/api/invoice/extract, described on the Invoice Extract page.
FAQ
Which PDF Vector action should I use to automate invoices?
Use Extract (Zapier: Extract Data From Invoice; Make: Extract structured data from an invoice; n8n: Invoice, Extract). It returns JSON that matches your schema, so later steps can map fields reliably.
Does PDF Vector have a trigger in Zapier, Make or n8n?
No. PDF Vector provides actions. Start your flow with a trigger from Gmail, Outlook, Google Drive, Dropbox, a form tool or a webhook, and pass the file or its URL to PDF Vector.
Can I process scanned or photographed invoices?
Yes. Image files such as PNG and JPG are supported on the pro, max and auto tiers. Validate the totals, since OCR errors usually affect digits.
How do I handle multi-page invoices?
Send the whole file. Credits are charged per page, and the extracted lineItems array covers all pages. Check that the line items add up to the subtotal to confirm nothing was skipped.
How much does it cost to extract data from an invoice?
Invoice Extract costs 6 to 18 credits per page depending on the model tier. A one-page invoice on mini costs 10 credits.
Do I need to write a JSON Schema by hand?
No. Use the free JSON Schema Editor on pdfvector.com to build one visually, then paste it into the JSON Schema field.
Sources (checked 2026-11-11): https://www.pdfvector.com/integrations/zapier, https://www.pdfvector.com/integrations/make, https://www.pdfvector.com/integrations/n8n, https://www.npmjs.com/package/n8n-nodes-pdfvector



