The challenge
Every month hundreds of documents arrived from suppliers and customers —invoices, purchase orders, delivery notes and contracts— as PDFs, photos or scans in different formats.
A team typed them into the ERP by hand. It was slow, caused errors in amounts, dates and document numbers, and contract information (terms, penalties, renewals) could not be searched without opening each file.
The solution
We implemented a flow that receives the documents, applies OCR and uses language models to extract the key fields into a defined schema: supplier, tax ID, dates, amounts, taxes, line items and, for contracts, parties, term and conditions.
Each document is validated automatically against business rules: tax ID format, totals that add up, consistent dates and a match with purchase orders.
Documents that fail validation or have low confidence go to a human review queue; the rest are loaded straight into the ERP with a link back to the original file.
Results
- Less manual data entry and fewer capture errors.
- Invoice and contract data that can be searched and analysed.
- Full traceability from each data point to its source document.
- The team focuses on reviewing exceptions, not copying information.