Business Impact

  • 84%

    solution accuracy

  • 90%

    accuracy in the Purchase Order (PO) number

  • 92%

    accuracy in the invoice id (invoice number) field and 94% accuracy in the the total amount field

Customer Key Facts

  • Country : United States
  • Industry : Manufacturing

Problem Context

Sanmina is a Fortune 500 company and a leading provider of integrated manufacturing solutions for the global Electronics Manufacturing Services (EMS) market. With 15,000 varying templates in different languages (English, German, Hungarian, Swedish), Sanmina receives approximately 300,000 pages of invoices every month (or about 3 million per year), and their current invoice processing platform is a batch-based template-driven system.

To extract these invoices from the Google Workspace mailbox and confirm the PO number for invoices, the client requested an automated workflow solution based on machine learning (ML) and a collaborative AP Workbench where AP users and buyers could see a single pane of truth regarding invoice state, scanned document, and its associated PO information.


  • Inaccuracies in data extracted by the existing Optical Character Recognition (OCR) software
  • 17% of the total invoices being placed on hold
  • Labor-intensive manual coordination process involving accounts payable, the buyer, the receiver, and the supplier

Technologies Used

Cloud Run
Cloud SQL
Cloud Scheduler
Cloud Build
Cloud Storage
Document AI - OCR and Invoice Parser
Secret Manager
Container Registry

Developed an end-to-end accounts payable application


Quantiphi developed an end-to-end accounts payable application that consisted of an invoice parser solution, an account payable workbench, and invoice matching, validating extracted information with PO data from ERP.

This engagement included building a custom UI with the following features:
– UI to have multiple user access with restricted information access to different users
– Summary list of successfully matched invoices and unmatched invoices
– Invoice view with the corresponding list of extracted entities with a layer of Human-in-the-Loop (HITL) to review and edit the results
– PO data for validating extracted invoice information


  • High individual field accuracy as compared to the previous process (overall accuracy recorded is 84%)
  • Less manual intervention as compared to the previous solution with 90% accuracy in PO number

About Sanmina Corporation – a Fortune 500 company, is a leading integrated manufacturing solutions provider serving the fastest growing segments of the global Electronics Manufacturing Services (EMS) market. Recognized as a technology leader, Sanmina provides end-to-end manufacturing solutions, delivering superior quality and support to Original Equipment Manufacturers (OEMs) primarily in the industrial, medical, defense, automotive, communications networks and cloud infrastructure markets. Sanmina has facilities strategically located in key regions throughout the world. More information about the Company is available at www.sanmina.com

Looking for similar project?

Let's Talk

Get your digital transformation started

Let's Talk