Guide · AI
How to automate invoice processing with OCR, extraction and review
Reviewed
Short answer
To automate invoice processing in PDFBetter, create a workflow from the “Invoices: OCR → extract → review → export CSV” preset, choose the extraction schema with your invoice fields, and set it to start when a document is moved into a folder. Each invoice filed there is recognized, extracted, held for a person to check the values against their page quotes, and exported as a CSV once approved. Nothing is approved for you unless you allow it.
On this page (7)
What you need
An extraction schema for your invoices (create one in the Extractions dashboard first), a folder for incoming invoices, and a workflow slot: 1 saved workflow on Free, 20 on Pro, unlimited on Business.
1Your PDF
2Pick the fields
3Rows with page evidence
Set up the workflow
- 1
Open Workflows and start from the Invoices preset
The preset adds four steps: recognize text (OCR), extract, review and export.
- 2
Choose the extraction schema
Pick the schema with your invoice fields. Set the OCR languages your invoices use.
- 3
Set the trigger to a folder
Choose the folder that starts this workflow. Runs can also be started by hand on a document or by a connected agent.
- 4
Pick the export
CSV by default; JSON or Excel also work. Download is the default delivery; webhook delivery is available on Business.
- 5
Create the workflow
Edits later apply to new runs; runs already in progress finish with the steps they started with.
Review and approve
Runs that reach the review step show “Needs approval”. Open the run to see each extracted value with the page it was found on, values marked “Not in document”, and anything to check. Correct a value if needed, then approve to continue or reject to stop the run. Exports use the approved, corrected values.
A review step waits until someone decides; there's no automatic timeout. An agent can approve only if you turned on agent approval for that specific step.
When something fails
Temporary problems such as a busy AI provider or a timeout are retried up to three times with growing pauses. An exhausted allowance, a missing document or invalid settings stop the run with the reason; after fixing it, use Retry to continue from the failed step.
What it costs
The workflow itself is free to run. OCR uses OCR pages (20 a month on Free, 1,000 on Pro), and extraction uses 1 AI credit per page read. A two-page invoice therefore uses two OCR pages and 2 AI credits.
Tools used in this guide
- PDF to JSONChoose the fields you need, run it on one PDF or a whole batch, and check each value against the quote and page it came from before exporting.
- OCR PDFUpload a scan, pick its language, and download a PDF that looks the same but lets you search, select and copy the text.
- PDF to CSVTurn tables in a PDF into CSV files ready to import into a spreadsheet, database or accounting tool.
- PDF to ExcelPull the tables out of a PDF into an Excel workbook you can sort, filter and sum.
Questions people ask
Does it post invoices to my accounting software?
Not directly. It produces a CSV, JSON or Excel file to import, or, on Business, sends a signed webhook with a link to the file that works for 24 hours, which your own integration can pick up.
Does it check that the invoice totals add up?
No. It extracts what is printed. Add your own checks in your spreadsheet or accounting system.
Can I skip OCR for digital invoices?
Yes, remove the OCR step if all your invoices are digital PDFs. Keep it if any arrive as scans; extraction on a scan without OCR finds little or nothing.
Can steps run in parallel or branch?
No. Steps run one after another in the order you set, with no branches or loops.
Are extracted values stored in the run history?
No. The run keeps ids and counts; the values live in the extraction record you open, correct and export.