PDF extraction of material certificates

The challenge

The production process depends on material certificates. Every batch of material that enters the process arrives with a certificate from the supplier, and the certified values on it have to be recorded before the material can be used.

Those certificates arrive as PDF attachments in a shared mailbox. From there, the process was entirely manual: an admin operator opened each mail, saved the attachment to a folder, opened the PDF, and copied the certified values one by one into an Excel file. The values were manually compared against the internal spec sheets before getting loaded into the ERP system.

The process was slow and prone to errors, generating a lot of frustration among the admin operators. A mistyped digit in a certified value is not a cosmetic error in a production environment: it is a material specification that no longer matches the production requirements.

Our approach

1. Listen - We followed the process from the shared mailbox to the ERP load and collected a representative set of certificates across suppliers, to properly understand the process end to end.
2. Explore - We tested extraction against real certificates, established what could be read reliably with general rules and where supplier-specific handling was needed, and agreed what “validated” has to mean before data is allowed into the ERP.
3. Shape - We designed the Dataverse model to hold the extracted values alongside the original PDF, and designed a custom application around the admin operative’s own process requirements rather than around the data model.
4. Build - We delivered the end-to-end flow, from mailbox pickup through custom extraction agents tailored to each supplier’s certificate specifications to the ERP push, and handed over the supplier rules so the client can extend them as new suppliers come on board.

What we built

• An automated pickup that monitors the shared mailbox, detects incoming certificates and pulls the PDF attachment into the flow - no more saving files to a folder by hand.
• Custom extraction agents, built in Azure AI Foundry and called from the flow, that read the certificate according to general rules plus supplier-specific requirements, so each supplier’s layout and terminology is handled on its own terms rather than forced into one template.
• A Dataverse database storing the extracted values together with the original document, giving every certified number a traceable link back to the PDF it came from.
• An automatic comparison of the extracted values against the internal spec sheet, flagging any certified number that falls outside specification before it reaches the operator’s approval.
• A Power App built on Dataverse that shows the extracted content next to the original PDF, so the admin operator validates what the machine read against the source in one screen - human on the loop, not out of it.
• A push of validated certificates into the ERP system, replacing the Excel template as the route into production data.

Key outcomes

• The retyping is gone - values move from PDF to ERP without passing through a spreadsheet, removing the step where most errors were introduced.
• Every certificate is checked against spec - the comparison runs on all of them automatically, instead of depending on someone remembering to look.
• Validation instead of data entry - the operator’s job becomes confirming what was read, with the source document on screen, rather than transcribing it.
• Traceability by design - each stored value keeps its link to the original certificate, so any number in the ERP can be traced back to the document that certified it.
• Room for new suppliers - supplier-specific extraction rules sit alongside the general ones, so onboarding a new supplier’s certificate format is a configuration change, not a rebuild.

From retyping PDFs into Excel to an end-to-end automated flow, with the operator on the loop instead of in it.