AI Powered Document Extraction Azure Form Recognizer Presented

  • Slides: 14
Download presentation
AI Powered Document Extraction Azure Form Recognizer Presented by: Drew Forward PCL Constructors Inc.

AI Powered Document Extraction Azure Form Recognizer Presented by: Drew Forward PCL Constructors Inc.

Outline • Who am I? • Our business problem • What is Form Recognizer?

Outline • Who am I? • Our business problem • What is Form Recognizer? • Pre-requisite steps • Feature Demo • Applicability to Accounts Payable • Questions?

Introduction • Drew Forward - Technical Solution Analyst • 8 Years experience with Dynamics

Introduction • Drew Forward - Technical Solution Analyst • 8 Years experience with Dynamics (2012 R 2, R 3, 365) • 3 Implementations – Supply Chain, Retail, HR, Payroll • Custom development

Business Problem • Unplanned, extra work occurs in the field and needs to be

Business Problem • Unplanned, extra work occurs in the field and needs to be recorded appropriately • Paper with signature used as a backup when quoting changes (owner/recovery of backcharge) • Work recorded per day, large volume of paper • Double entry in the trailer • No mobile app (Yet)

Azure Form Recognizer • AI based cloud service by Microsoft • Train a “model”

Azure Form Recognizer • AI based cloud service by Microsoft • Train a “model” based on the business form • Some pre-built models provided • Call API to upload document, and return values from that document • Nominal cost per form extracted

Pre-Requisite Steps • Get an Azure Storage Account • Get a Form Recognizer Account

Pre-Requisite Steps • Get an Azure Storage Account • Get a Form Recognizer Account • Determine format of the business form • Download Docker to locally connect to necessary accounts • See References for additional information

Train the Model • Tag keys on the document to extracted data • Data

Train the Model • Tag keys on the document to extracted data • Data extracted automatically • Upload minimum 6 documents to the storage account • Suggested 20+ to get variety of data & fields

Train the Model 2 1 • Upload minimum 5 documents • Suggested 20+ to

Train the Model 2 1 • Upload minimum 5 documents • Suggested 20+ to get a variety 3

1 2

1 2

Mass Upload • Wizard based process • Upload 1 -M documents to Form Recognizer

Mass Upload • Wizard based process • Upload 1 -M documents to Form Recognizer • Review & correction process • Confirmation to create business record

Demo

Demo

Further Applications • Vendor Invoice Processing • • Train 1 model per Vendor Invoice

Further Applications • Vendor Invoice Processing • • Train 1 model per Vendor Invoice format Call Form Recognizer Upload raw documents Create pending vendor invoices, lines & attachment, match against business data • Microsoft’s solution uses Data management framework but requires a different solution to call APIs • Several enhancements in October 2020 release plan • Microsoft is also demoing an accelerator using Power Apps & AI Builder

References • Azure Form Recognizer [Link] • Training the Model [Link] • Create a

References • Azure Form Recognizer [Link] • Training the Model [Link] • Create a storage account [Link] • Create a resource [Link] • Pricing [Link] • Documentation [Link]

Questions?

Questions?