Overview of Intelligent Document Processing (IDP) and its Benefits
October 12, 2021
|
12 min
IDP
DATA-EXTRACTION
DATA-ENTRY
OCR
ARTIFICIAL-INTELLIGENCE
RPA

Intelligent Document Processing automates data capture from multiple documents and data sources and organizes it for further processing. The technology enables businesses to seamlessly integrate with core processes, eliminate manual labour, address challenges faced in reading different complex document layouts, and meeting legal & compliance requirements. Accurate data is the foundation of every organization, and IDP assists businesses in dealing with the complexity of processing huge volumes of documents, helping them automate manual data entry processes, and move away from traditional semi-automated OCR workflows. 

So, what exactly is intelligent document processing, and what are its different use-cases in different industries - we’ll find out in this blog.

Let’s get right into it:-

Table of Contents

What is Intelligence Document Processing (IDP)

Intelligent Document Processing is the automation of data extraction from complex semi-structured/unstructured documents and converting it into structured usable data. It is also referred as Cognitive Data Processing or Intelligent Data Capture. IDP takes advantage of Artificial Intelligence (AI), Machine Learning (ML), Optical Character Recognition (OCR), Computer Vision, and Intelligent Character Recognition (ICR) technologies to classify, categorize, extract relevant data, and validate the extracted data for improved accuracy.

Often IDP is used interchangeably with OCR which is wrong because IDP is the next-generation data extraction technology developed only to overcome the limitations of traditional OCR in extracting data from more complex and non-standard documents.

Data extraction from documents can be done in 3 ways:-

  1. Manual Data Extraction
  2. Template based Data Extraction
  3. Intelligent Document Processing

To make it simpler, OCR is a subset of IDP but the reverse is not true. That means, IDP uses traditional OCR to extract data at some level but IDP goes beyond it. With the help of Named-entity recognition and classification, supervised/unsupervised learning, and NLP context analysis, IDP has a lot more to offer for improved accuracy in document processing and analysis. 

IDP Workflow

To start with, scanning hardware devices capture information from paper-based documents, convert them into electronic formats, and provide the digitized versions of documents as input to IDP solutions. Computer vision algorithms in IDP solutions are able to recognize different document layouts from scanned images, PDF files, and a variety of file types, both in digital and paper-based forms.

Natural Language Processing (NLP) technology used with IDP workflows is able to recognize characters, symbols, letters, and numbers from paragraphs, tables, or unstructured text in documents. It synthesizes them using OCR, and by using techniques such as named entity recognition, sentiment analysis, and feature-based tagging, it successfully reads information from documents and enters into content management systems with a 99%+ accuracy.

Following are the key steps in the IDP workflow:-

IDP workflow steps


Step 1 - Document preprocessing 

Where there is data extraction, there is OCR. As a document is ingested into a document processing solution, it goes through the first phase of document pre-processing in IDP workflow. The overall accuracy of OCR depends on how accurately it can identify/distinguish a character/word from the background. Some of the basic techniques employed in this phase are:-

a) Binarization - In simple terms, binarization is the technique to convert a colored image into a black and white pixels. Now, the image consists of only 2 kinds of pixels - black pixel value = 0 & white pixel value = 256. The aim to create a binary and distinction between the characters to be read in a text file(black pixels) and background(white pixels)

b) Deskewing - While scanning a document, the scanned image might be slightly aligned horizontally, which is not ideal for OCR. Techniques such as Projection profile method, Hough transformation method, and Topline method are used for skew correction. 

c) Noise removal - The aim of this step is to get rid of any unwanted small dots/patches so that OCR doesn’t confuse these dots with characters.

Step 2 - Document classification

Document classification happens in 3 steps:-

a) Identify the format - Figure out whether the file is a pdf document, JPG, PNG, TIFF, or any other file format.

b) Identify the structure - The OCR solution tries to differentiate amongst structured, semi-structured, and unstructured documents. Structured documents have a fixed template and layout, whereas semi-structured documents have some form of structure in it that means they may contain similar information at different locations in the document. An invoice is a great example of semi-structured documents - vendor’s address in different invoices can be at different places. To make sense of these values, the document processing solution should have some kind of contextual understanding of data and the document.
     Unstructured documents have hardly any structure yet organizations need to extract data from them for various purposes. In an unstructured document, sometimes certain values may not have any key assigned to them - such as dates or email addresses may be there on documents but without any key identifier such as “Date” or “Email”.  A contract is a good example of an unstructured document.

c) Identify the document type - The third step of document classification is trying to figure out the document type, that is, to find out whether the ingested document is an invoice, bank statement, t12 statement, shipping label, or any other document. The ability to identify a document type successfully and queue it for data extraction depends on the data already fed into the IDP solution. 

Step 3 - Data extraction

There are mostly two parts of data extraction:-

i) Key-value pair extraction - Extracting the values assigned to unique key identifiers in a documents

ii) Table extraction - Extracting line items arranged in a table form

There are certain ways to do it:- 

a) OCR - OCR is the first step of data extraction. As essential as this step is, there are certain errors that can happen during OCR:-

  1. Error in word detection – Failing to detect a text block in the image, an error commonly caused because of the poor image quality. 
  2. Error in word segmentation – Interpreting a word incorrectly, due to wrong interword space detection, different text alignments, and spacing.
  3. Error in character segmentation – Unable to detect single characters in a segmented word. This is frequent for cursive or connected alphabets. 
  4. Error in character recognition – Failing to identify the right character in a bounded character image.

These errors could be rectified by dictionary look-up, k-mer, and n-gram language models. 

b) Rule based extraction - Rule based models work well for structured and semi-structured documents. These models can identify key-value pairs/line items by taking a position reference in a document. Named-entity recognition approach and n-gram model come handy in identifying a value assigned to a key identifier. For example, no matter the position of invoice number in an invoice, a set of strings next to “Invoice Number” or “Invoice No” is the value the model is looking for.

c) Learning based approach - Deep learning and ML-based OCR-hybrid data extraction techniques need supervised/unsupervised learning to train their models on. The efficiency of these models are determined by the accuracy rate and confidence score. With the increasing number of documents processed and the amount of training and feedback provided, the model grows in accuracy. Docsumo takes the similar approach of data extraction where an ML-based model sits on the top of  template based OCR. At Docsumo, Simple OCR correction approach along with context based NLP is used to improve the accuracy and the quality of data.

Step 4 - Data validation 

This step is crucial in detecting the inaccuracies of the extracted data. Certain data validation rules are applied within the document so that any inaccuracy could be detected and flagged for correction. For example, the ‘total amount payable’ in an invoice should be a sum of ‘subtotal’ and ‘tax payable’. If there’s any discrepancy between two, the invoice gets flagged and held for review. 

Step 5 - Human review 

Although we’d like it to be, no data extraction model is 100% accurate, hence a layer of human intervention is there in the IDP workflow. Any document flagged red is reviewed by a human-in-the-loop. This is especially helpful in the supervised learning of the model and improving the accuracy of the model. The more documents are processed and reviewed, the more improves the accuracy of the data extraction model.

Once the data is extracted and cleaned up, the software can push to the database or export it in multiple formats. IDP workflows let users convert documents into different formats such as JSON, XML, PDF, etc. 

Intelligent Document Processing Use-Cases

IDP solutions have following capabilities to offer:-

  1. Identify difficult content  - IDP can decipher content from substandard quality documents, which cannot be read by traditional OCR. AI-driven IDP solutions can also decipher the relevance of a character/word based on the context and defined rules which is not possible with traditional OCR solutions. 
  2. Reading barcodes/QR codes - IDP solutions are capable of processing barcodes/QR codes.
  3. Document auto-classification - Multiple document types coming from different streams and sources in different forms - IDP solutions are capable of auto-classifying documents in different classes for further processing.
  4. Extract information you need - IDP models can be trained to extract specific information in the document.
  5. Validate data - Processed data can be validated against the rules set to determine the accuracy of the system and to assign the document for manual review.
  6. Organize data and reports - IDP solutions make it easier to classify different documents, extract data from multiple sources, and assemble it at one place for further analysis.

Based on these capabilities, IDP solutions find different use-cases in different industries:-

Industry specific IDP use-cases
1. Lending

Whether it is commercial loans, personal, real estate, or small business loans, lenders use IDP solutions to process loan applications to run a careful credit risk analysis of their borrowers.  IDP can eliminate manual data entry tasks involved in processing loan applications and ensure faster turnaround times leaving lenders with more time for analysis.

For mortgage loans, IDP makes it easier to validate and verify customer data, credit reports, personal identification documents, income verification documents, and various other document types that support loan and mortgage applications.

2. Insurance

IDP is used by the insurance industry to manage huge volumes of customer data and do credit profile analysis. IDP solutions help determine the risk appetites based on customer data supplied, and give applicants the best possible premiums with other attractive benefits.

3. Logistics

The logistics industry has data exchanging thousands of hands all the way from shipping, transportation, warehousing, and doorstep customer delivery. This information has to be validated, verified, cross-checked, and even re-entered as requirements for manual processing by third parties. On a supply chain level, companies use IDP to deliver invoices, labels, and agreements to contractors, vendors, and transportation teams.

IDP solves the problem regarding variability of documents, and helps in reading unstructured data from different sources, thus eliminating the need for manual processing and saving countless hours of time in the process. When businesses expand and scale up to accommodate larger client user bases, IDP keeps up with them thanks to intelligent automation of various document processing elements in logistics workflows.

4. Commercial real estate

Intelligent document processing finds its different use-cases in the commercial real estate industry in the form of rent roll processing, lease agreements, offering memorandums, operating statements, T12 statements, and for comparing real estate market rates for figuring out the most lucrative investments. 

Commercial real estate property owners can pull details from multiple data sources using IDP workflows and decide whether renting/leasing/buying new properties give substantial returns on investments.  When buying new properties, they can determine if they’re getting the deals based on market rates by doing cash flow analysis and market comparisons using insights derived from IDP.

5. Accounts payable

IDP in accounts payable automation enables accounting professionals to offer clients a seamless user experience. Invoices with different layouts and structure can be processed through an automated accounts payable solution, and can be matched against purchase orders in real-time. 

Benefits of Intelligence Document Processing

The technology is used to eliminate manual repetitive tasks and converts unstructured data into readable forms which can be used in various applications and systems.

Intelligent Document Processing offers users the following benefits:- 

Faster document processing Speed up data extraction up to 10 times with AI-native IDP solutions.
Improved accuracy Achievable data extraction accuracy up to 99.9% for different documents types with 95%+ straight through processing .
Improved productivity With improved processing time and straight through processing, built-in IDP solutions need little or no human intervention. That means your employees don’t need to extract data from unstructured text and manual data entry, thus improved productivity.
Electronic document storage IDP solutions ensure paperless document processing and data sharing amongst peers enabling digital transformation at your organization.
Cost efficient No more manual data entry, no more human errors, and no more manual review - combine all of this with fast and accurate document processing which results in cost savings up to 70%.
Business level automation IDP solutions are easy to integrate with your existing system combining which with other automation solutions, you can achieve a fully integrated RPA system.

Types of IDP vendors

To make the complete sense of the industry, we’ve divided IDP vendors into 4 categories:-

1. Innovative IDP vendors 

This is the set of most recent IDP vendors who have built AI-native platforms to automate document processing. Because of this, these vendors are able to process complex and varying documents with great precision. With an AI-centric approach, these vendors are able to offer an end-to-end document processing solution which requires little or no human intervention and leaves a greater impact on the business. Here are some examples of such vendors:-

  • Docsumo
  • Hyperscience
  • Rossum
  • Infrrd
2. Legacy IDP vendors

Instead of taking an AI-native approach, these vendors build a working IDP model based on their legacy OCR/RPA solutions. These solutions are useful in processing documents in bulk that can be ‘templatized’, have simple layouts, and don’t offer too many variations. Often these vendors have a broader portfolio of automation products to offer, and that’s why IDP takes a back seat. Here are some examples of such vendors:-

  • Abbyy
  • Kofax
  • AntWorks
  • Automation Anywhere
3. Niche IDP vendors

This set of vendors could be a subset of the two categories mentioned above but what differentiates them is that they are focused on solving a narrow set of problems often catering to a particular industry. Since they’re focused on a specific problem, they are able to provide quick, reliable, and efficient solutions within the industry. Here are some examples of such vendors:-

  • EvolutionAI  
  • Instabase
  • Ocrolus
  • ClickAI
4. IDP components technology providers 

Instead of providing a complete IDP solution, these vendors focus on providing different technology components such as OCR and computer vision. They provide general purpose technology components that could be used by different businesses to build a solution that is specific to their use-case and requirements. As a business, you need to ensure that you have a team of IT professionals and data scientists who can design a use-case specific solution for your business, if you're opting for these vendors. Here are some examples:-

  • Google cloud vision
  • Amazon textract
  • Microsoft azure computer vision

How Docsumo is revolutionalizing document processing for businesses

Docsumo integrates seamlessly with various document workflows and business processes. Docsumo is able to help businesses with:-

  • Invoices and bank statements processing
  • Simplifying data extraction from income/identity verification documents
  • Automated data extraction for IRS Forms
  • Non-standard lease agreements/sales comps/offering memorandum data processing
  • Processing bill of lading/shipping labels/receipts
  • And more!

The biggest advantage of using Docsumo is the use of trained APIs. Docsumo comes with pre-trained APIs for some of the common document types such as bank statements, acord forms, invoices, IRS forms, driver’s license, etc. That means you don’t need to invest much time into training the model from scratch. 

Docsumo APIs flag missing values, fields, and duplicate data entries, thus eliminating data redundancy and error-rates. Once the APIs extract data accurately, users simply have to review and approve the final changes on the platform. Later, users can upload documents in bulk and process them for further use.

Is document processing becoming a hindrance to your business growth?
Join Docsumo for recent Doc AI trends and automation tips. Docsumo is the Document AI partner to the leading lenders and insurers in the US.
Written by
Pankaj Tripathi
Share this Blog:
  • I agree and understand that Docsumo may send me marketing communication via email. I may opt out at any time.

Overview of Intelligent Document Processing (IDP) and its Benefits
IDP
|
September 2, 2021
|
12 min
Share this article

Blog

Explore more