The problem
Accounts payable teams still retype vendor, invoice number, dates, totals, and line items from PDFs and scans. The slow part is spotting outliers before someone approves payment. A shared inbox and a spreadsheet leave no run history when a vendor asks what happened to an invoice.
How the workflow runs
Extract invoice fields
The starter workflow sends the uploaded invoice to an extract agent. It pulls the vendor, invoice number, dates, currency, totals, tax, and line items into structured fields the rest of the run can use.
Write an approval summary
A generate step writes a markdown approval summary from those fields and calls out amounts that look like outliers. The summary stays attached to the run so the next person sees the same numbers.
Refine the canvas
The starter stops at the summary. On the visual canvas you can insert a human review pause, a purchase-order match, or a compliance check before anyone treats the summary as an approval.
What the hosted app shows
The hosted invoice desk has an upload, an invoice details form, and run status. A results page shows the latest run, a table of extracted fields, and the source document. Operators work the desk. They do not open the studio.
Who it is for
Finance operations and accounts payable teams that receive vendor invoices by email or upload and need a repeatable desk instead of a one-off script.
Questions
No. It extracts fields and writes an approval summary that flags outliers. Payment approval stays with your team. Add a human review step on the canvas if someone must sign off before the summary is used.
The hosted app accepts the documents your workflow is set up to read, including PDFs and scans. OCR uses extra credits when the file is an image rather than text.
Use the PO matching solution for that comparison. You can also add a validate step on this canvas if invoice and PO files arrive in the same run.