Extract Invoice Data from Dropbox PDFs into Notion

What we're building

Finance drops vendor PDFs into a Dropbox folder. Make downloads each file, calls Cryvis, and creates a Notion database item with invoice headers. Useful when the AP tracker already lives in Notion instead of Sheets.

Dropbox: Watch Files (/Finance/Invoices/Inbox)
        |
        v
Dropbox: Download a File
        |
        v
HTTP: POST https://api.cryvis.com/v1/documents/invoice
  Authorization: Bearer <API_KEY>
  multipart: file
        |
        v
JSON: Parse JSON
        |
        v
Notion: Create a Database Item
  Title = invoice_number + seller.name
  Properties = dates, currency, amounts, parties
        |
        v
Dropbox: Move a File → /Finance/Invoices/Processed

What you need

• Make.com with Dropbox and Notion connections
• Dropbox folders: Inbox and Processed
• Notion database (full-page database) with properties listed below
• Cryvis API key — Invoice / docs
• Sample multi-page and single-page PDFs

Credits: 1 credit per PDF page or image.

Create the scenario

1. Dropbox — Watch Files (Trigger)
2. Dropbox — Download a File
3. HTTP — Make a Request
4. JSON — Parse JSON
5. Filter — success is true
6. Notion — Create a Database Item
7. Dropbox — Move a File (archive)

Configure source

Watch Files

• Path: /Finance/Invoices/Inbox
• Watch: files only (not folders)
• Limit: low while testing

Download a File

• Map file path / ID from Watch Files
• Ensure binary output is available to HTTP

Edge case: Dropbox Paper files and .docx must not hit the invoice endpoint. Filter:

Name ends with .pdf OR .png OR .jpg OR .jpeg OR .webp

Shared team folders: use the connection that owns the team space, or the Watch will see nothing.

Configure Cryvis HTTP

Do not URL-encode the file. Do not send the Dropbox share link as text — Cryvis needs the bytes, not a URL string.

Why Download is mandatory: Watch Files metadata alone is not enough. The HTTP multipart part must be the file content. If you map a path string into file, Cryvis receives text, not a PDF, and extraction fails.

After a successful response, keep the Dropbox path and content-hash (if available) on the Notion item so you can prove which binary produced which extraction when auditors ask.

Process response

Primary fields for Notion properties:

• data.invoice_number
• data.invoice_date
• data.due_date
• data.currency
• data.seller.name, data.seller.tax_id
• data.buyer.name
• data.subtotal, data.tax_total, data.total_amount, data.amount_due

Notion title property cannot be empty. Build it as:

{{data.invoice_number}} — {{data.seller.name}}

If invoice_number is missing, fall back to Dropbox filename + date.

line_items is an array — Notion cannot store it cleanly in one property without JSON text. Either:

• Store line_items as a Code/JSON property (long text), or
• Split rows in a child database with an Iterator (line items guide)

This tutorial stores a truncated JSON string in Line Items (raw) for audit only.

Configure destination

Notion database properties:

Date properties: if Cryvis returns YYYY-MM-DD, map directly. If Make treats them as strings, wrap with a date parse function in the mapper.

Move a File after create: destination /Finance/Invoices/Processed/{{filename}}. Fail the move only after Notion succeeds so a Notion outage does not archive unread files.

If Processed already contains the same filename, Dropbox Move may fail. Prefix with a timestamp or invoice number:

/Finance/Invoices/Processed/{{data.invoice_number}}_{{filename}}

That also makes manual lookup easier when finance digs through the archive.

Filters and error handling

• Filter: MIME / extension before HTTP.
• Filter: HTTP status 2xx and success true before Notion.
• Error Handler on HTTP: create a Notion item in a separate Extraction Errors database (or send Slack). Include Dropbox path and response body.
• Duplicate check: Notion has no native unique constraint. Use a Data Store keyed by seller.name|invoice_number or search Notion for an existing item before create (Search Objects → Filter → Create only if empty).

Test

1. Upload a known PDF to Inbox.
2. Run once; confirm Notion title and Amount Due.
3. Confirm file moved to Processed.
4. Upload a .txt — Filter should skip Cryvis.
5. Upload a 3-page PDF — expect 3 credits (Credits); Notion still gets one database item for the whole document (credits track pages, Notion items track documents).

Notion-specific pitfalls

• Select vs text for Currency: if Currency is a Select, every ISO code you encounter (USD, EUR, INR) must exist as an option or Notion rejects the write. Prefer rich text until the set stabilizes.
• Date timezone: Notion dates are date-only for invoice fields; do not convert 2024-03-15 through a timezone that shifts the calendar day.
• Title length: very long vendor names + invoice numbers can hit Notion limits — truncate vendor to 80 chars in the title formula if needed; keep the full name in the Vendor property.
• Concurrent Watches: two scenario runs on the same file (manual + scheduled) can create duplicate Notion pages before Move completes. Use a short Data Store lock keyed by Dropbox file ID around the HTTP→Notion section.

Production considerations

• Notion API rate limits: batch large backfills overnight, not every 1 minute Watch on a cold folder of 500 PDFs.
• Property type mismatches (text into Number) fail the Notion module — coerce amounts with parseNumber / formatters.
• Permission: the Make Notion connection needs insert access on the database, not only the parent page.
• Retention: Processed Dropbox folder lifecycle rules (delete after 90 days) keep storage down; Notion remains the system of record.

Next steps

Sheets instead of Notion: Extract invoice data. Gmail intake: Automate invoice processing. HubSpot vendors: Extract vendor details.

Invoice API · extractInvoice docs · Make.com hub.