Tech, ExplainedEnterprise SaaS and software
Product

Intelligent document processing (IDP)

The supplier invoice, read into fields.

BEFORE ITAFTER ITIntelligent documentprocessing (IDP)NanonetsInfrrdInstabaseABBYYCOVERSField extractionConfidence scoresAny layoutHuman reviewOffice printers and copiers↗ WORKPLACE DEVICESsendsscanned pagesRobotic processautomation (RPA)collectsinvoicesfrom emailposts invoicefieldsEnterprise resourceplanning (ERP)fillscase fieldsBusiness process management(BPM) and case managementfilesthe originalEnterprise contentmanagement (ECM)BEFORE ITAFTER ITIntelligent documentprocessing (IDP)NanonetsInfrrdInstabaseABBYYCOVERSFieldextractionConfidencescoresAny layoutHuman reviewOffice printersand copiers↗ WORKPLACE DEVICESsendsscanned pagesRobotic processautomation (RPA)collectsinvoicesfrom emailposts invoicefieldsEnterpriseresourceplanning (ERP)fillscase fieldsBusiness processmanagement(BPM) andcase managementfilesthe originalEnterprisecontentmanagement (ECM)BEFORE ITOffice printers and copiers↗ WORKPLACE DEVICES↓ sends scanned pagesRobotic process automation (RPA)↓ collects invoices from emailIntelligent documentprocessing (IDP)NanonetsInfrrdInstabaseABBYYField extractionConfidence scoresAny layoutHuman reviewAFTER IT↓ posts invoice fieldsEnterprise resource planning (ERP)↓ fills case fieldsBusiness process management (BPM) andcase management↓ files the originalEnterprise content management (ECM)
Intelligent document processing receives scanned pages from office copiers and invoice PDFs collected by RPA, reads each field with a confidence score, and sends the fields to the ERP and case management and the original to the content store.
opens a page inside this one a page in the same scene a page in another scene a concept page a label only, with no page

Intelligent document processing (IDP) is software that reads a scanned page or a PDF, such as a supplier invoice or an insurance claim form, and turns it into fields that another system can use. It grew out of optical character recognition (OCR), the older technology that turns the image of printed letters into text. IDP goes further. It understands which number on the page is the invoice total, and which is the tax.

Take, for example, the purchase team of an auto-components maker in Aurangabad. It receives about 3,000 supplier invoices a month, each laid out differently. One arrives as a PDF from a steel supplier. The IDP software first turns the page into text. Then an AI model finds the fields: the supplier's GSTIN, the invoice number, the invoice date, the purchase order number, the HSN code of each item, the taxable value of ₹84,000, GST at 18% of ₹15,120, and the total of ₹99,120. For each field, it gives a confidence score, a measure of how sure it is that it has read the field correctly. It also checks the arithmetic: ₹84,000 plus ₹15,120 is ₹99,120.

On this invoice, every score is above 95%. So the fields go straight to the ERP. There, the invoice is matched against the purchase order and the goods received note, the record that the steel arrived at the factory. This check of three documents against each other is called a three-way match. On another invoice, the date is smudged, and the score for the date is only 60%. That invoice goes to an employee, who checks the date on screen and corrects it. Her correction is used to improve the model.

Older OCR products needed a template for each supplier's layout, such as "the total is in the last box at the bottom of the page". A new supplier meant a new template. Today's IDP uses a large language model (LLM) that reads a new layout much as a person does. Because of this, reading documents is becoming a feature of general AI models. So IDP companies are moving on to running the process that the document starts, such as paying the invoice through accounts payable (AP) automation, or deciding an insurance claim.

IDP rarely works alone. A software robot may collect invoices from an inbox and pass them to it. A case in case management software may wait for its fields. The original document is kept in enterprise content management (ECM). Pages that start on paper are scanned on office printers and copiers.

Tech, Explained is a part of ingrid.fyi. Every page explains one piece of technology in plain words.