Intelligent Document Processing Definition Glossary

OK. What is intelligent document processing? RPA, OCR? ML, DL, DMS, ERP, HITL? WTF?!! Understanding the technology is hard enough. Introducing a new language is plain cruel. This is your go-to guide for every single intelligent document processing definition I can think of. Don’t give me TL;DR. Bookmark it. Pin it to your wall. It’ll help you understand the language of IDP…

This blog post is long enough, so I won’t faff around. Let’s jump into the list of IDP terms…

Intelligent document processing definition glossary

2-way matching

2-way matching ensures all the data on a purchase order and invoice match.

3-way matching

3-way matching ensures the data on a purchase order, invoice, and sales receipt match.

AI OCR

Unlike traditional OCR, AI OCR software doesn’t need specific rules or template setups. OCR powered by AI uses computer vision and machine learning technologies to recognize characters in images and documents with greater accuracy than traditional OCR. AI OCR software relies on training to be able to function as required.

Annotations

Document annotation uses machine learning or AI to identify fields and values in a document. Extracting information based on predefined rules. It means that information can be found quickly and removes the need for a human to trawl through the entire document. 

AI document processing relies on annotation to understand, process, and interact with the human language. Making it easy to structure data and present it to other users.

Application programming interface (API)

An API is a set of rules or protocols that lets software talk with each other to exchange data, features, and functionality. Examples of APIs include mobile payment processes, a rideshare app, switching your lights on from your phone.

Artificial intelligence (AI)

Artificial intelligence is a technology that enables computers and digital devices to emulate the cognitive functions of humans. To learn, read, write, talk, see, create, play, recognize, categorize, analyze, and make logical decisions.

Audit trail

A date and time-stamped record of the activity in a document workflow. Defining who, what, when, and how actions were taken. To analyze and improve workflow and participant performance. Often a regulatory requirement for compliance.

Business process management (BPM)

I’m not going to compete with Gartner… 

“Business process management (BPM) is a discipline that uses various methods to discover, model, analyze, measure, improve and optimize business processes.”

Cloud computing

The delivery of various services through the Internet. Include tools and applications like data storage, physical/virtual servers, databases, networking, and software. 

Rather than keeping files on a proprietary hard drive or local storage device, cloud-based storage makes it possible to save them to a remote database (in the cloud or a virtual space). Files stored in the cloud are accessible via the Internet, from anywhere in the world. 

Cloud computing is popular because of cost savings, increased productivity, speed and efficiency, performance, and security.

Cloud-native IDP technologies are designed, built, and operate in the cloud. They’re all about speed and agility. They embrace rapid change, scalability, and resilience. Cloud-native company examples include Netflix, Uber, Airbnb, and Rossum.

Hey! Have you heard the one about how cloud computing stores your data in the sky, and a storm can be catastrophic?!! For more intelligent document processing myths, take a look at this post. 

Cognitive document processing

A synonym for intelligent document processing.

Complex line items

Complex line items are detailed components in documents such as invoices, that contain complicated calculations or structures.

Complex tables

A complex table is a table of records that contains intricate structures – multiple rows and columns, merged cells, or nested structures – in a searchable format.

Computer vision

A field of artificial intelligence that trains computers to interpret and understand images, videos, and other visual inputs.

Concatenate

String concatenation is the process in which multiple character strings are joined or combined end to end into a single string.. For instance…

  • IBAN 1234 5678 9000 would become IBAN123456789000
  • PO: 5214 256 would become PO:5214256

Confidence score

A confidence score shows the extent to which the AI engine is confident that it got the text and the location of the field correct. To turn individual confidence scores into actionable data, you need to apply a threshold that determines if data is probably correct or incorrect. 

A threshold ranges between 0 and 1. Humans come into play when the confidence score is too low and human annotation is needed with the AI learning with each annotation.

Content management system (CMS)

Software used to manage the creation and modification of digital content. Multiple contributors can create, edit, and publish blog posts, webpages, etc. An example of a CMS is WordPress.

Customer relationship management (CRM)

A technology, strategy, or process, CRM is the management of customer data and interactions. Revenue and profitability is optimized along with driving customer satisfaction, trust, and loyalty.

Data capture

Data capture is the process of extracting structured and/or unstructured information from documents – paper or digital – and converting it into a machine-readable format.

Data extraction

Data extraction tools pull data from various sources – paper or digital – and move it to a new destination for processing or storage.

Data entry

Data entry software digitizes data by entering information into a computer system. A task that can be done manually by humans typing on a keyboard, or through automation by scanning a document.

Data leaks

A data leak is when information is revealed to unauthorized people because of internal errors. Often caused by poor data security, outdated systems, or a deficiency of employee training. A data leak can lead to identity theft, data breaches, or a ransomware attack.

A data leak is caused by an internal source exposing information. A data breach is when an external source breaches the system in a cyberattack.

Data mining

Also known as knowledge discovery in data (KDD). A process that identifies patterns in large data sets, and extracts useful information.

Deep learning (DL)

Deep learning is a subset of machine learning that powers most AI. It uses multi-layered neural networks – deep neural networks (DNN) – to replicate the decision-making ability of the human brain. 

DNNs are trained on large volumes of data to identify and categorize events, recognize patterns, and make predictions and decisions. DNNs provide greater accuracy than single-layer neural networks.

Digital archiving

Digital archiving preserves and stores digitized data, including documents, contracts, financial records (invoices, purchase orders), and customer data. Easy and wide access is available, along with the long-term preservation of digital data for use in the future.

Digital process automation (DPA)

Digital process automation combines people, devices, applications, and information across a business. Creating an agile and digital organization. Often called the next-generation of business process management (BPM), its goal is to improve and digitize business processes. 

Where DPA differs from BPM is that its aim is to provide automation across the entire business. Everyone in the company, not just IT, can benefit. Differentiators include a higher use of low-code development, consumer-grade user experiences, and AI-enabled innovation.

Digital worker

A digital worker used to refer to a human being with digital skills. Now, it’s defined as a virtual employee. An automated team member. Trained to perform specific tasks or processes, faster and error free, in collaboration with humans.

Document management system (DMS)

A document management system is how a business captures, stores, shares, tracks, and manages files or documents in digital format. With all documents stored in one place, it means users have access to the same set of information. A single source of truth.

Document matching

A process in which a document’s data is matched to one or more supporting documents. Matching is used to reconcile…

  • Purchase orders – what has been ordered
  • Goods received note (GRN) – what has been received
  • Purchase invoices – what’s been invoiced
  • Sales orders – what’s been ordered
  • Sales invoices – what’s been invoiced.

Document processing automation

Automated document processing uses technology to streamline and automate the data processing of business documents – invoices, purchase orders, bills of lading, etc. 

Tackling tasks such as data extraction, classification, validation, workflow management, and reporting. Reducing manual effort. Eliminating manual data entry errors, and increasing productivity.

Document splitting

Document splitting is an AI feature that checks to see if there are multiple documents within an original document, applying splits automatically when identified. For instance, a PDF containing multiple invoices that needs to be split so all relevant fields can be identified.

E-invoice

An electronic invoice is an invoice that’s issued, transmitted, and received in a structured electronic format that means it can be automatically processed.

The impact of manual invoice processing in numbers. $2.03 per invoice. 105 keystrokes per invoice. 111 seconds per invoice. 12.5% invoices needed rework. 3,840 invoices per month per FTE. 5.5 minutes to rework invoice.

You’ll find this invoice automation software post useful, if you want to automate your invoice process.

E-procurement

Electronic procurement, also known as supplier exchange, is the centralized management of a business’ procurement and supply chain through an online platform or digital system. A digital transaction process that automates the procurement lifecycle.

Electronic document management system (EDMS)

An EDMS is software that stores, organizes, and tracks digital documents and information in a single repository. Enabling easy document retrieval, collaboration, and workflow management.

End-to-end automation

The full and seamless automation of a process, covering every stage. From initiation to completion without the need for manual intervention. Rossum orchestrates the transactional process end to end, from document reading and understanding to workflows and answering emails.

Enterprise resource planning system (ERP)

Let’s go to the industry leader. SAP‘s definition of ERP… 

“Enterprise resource planning (ERP) is a software system that helps you run your entire business, supporting automation and processes in finance, human resources, manufacturing, supply chain, services, procurement, and more.”

Few-shot learning

A machine learning framework in which an AI model is trained to perform tasks with only a small number of examples, enabling it to generalize and make accurate predictions with minimal training data.

Rossum Aurora - our advanced AI - requires 10x fewer training examples to reach must-have accuracy levels, compared to previous AI generation.

Rossum Aurora – our advanced AI – requires 10x fewer training examples to reach must-have accuracy levels, compared to previous AI generation.

Generative AI

AI technology that can create various types of content. Examples are Dall-E for image generation and large language models (LLM) like ChatGPT for creating content, music, video, and code. While it’s touted as new, it was originally introduced in the 60s in chatbots.

Who’d have thought?!!

Get your chops around this article, if you want to learn more about the role Gen-AI plays in invoice fraud, Generative AI is Fueling a Surge in Invoice Fraud.

Hallucinations

These come about when AI – usually an LLM – produces something that’s coherent and grammatically correct, but is factually wrong or doesn’t make sense. Not intentionally lying. Just wrong. Hallucinations occur if training data is limited, biases in the model, or complexity of language.

Example from ChatGPT showing incorrect information. Asked how many letter S in the word elephant, ChatGPT says two. Corrects itself to one. Finally to zero.

Addressing the elephant in the room.

ChatGPT explaining why it made a mistake - hallucination. Machine learning model relying on patterns and data. Not infallible and errors can occur.

Now I feel like such a bully.

Human-in-the-loop (HITL)

Human-in-the-loop is when there’s a need for human interaction, intervention, and judgment to control or change the outcome of a process. 

Humans and machines working together to perform a task. Machine learning and generative AI have brought human – AI collaboration to the forefront.

Hyper automation

Hyper automation is when everything in an organization that can be, is automated. Using AI, RPA, and other technologies to streamline processes so as to eliminate human intervention.

Instant learning

At Rossum, Aurora for Instant Learning rapidly increases the accuracy of new document formats and custom fields. No technical expertise is needed to train the AI, as it learns from user feedback, document to document, through an ergonomic UI for fast human – AI collaboration. And if confidence is low, it’ll be flagged.

Halfway through my intelligent document processing definition list. How you holding up? I did warn you it was a long post. If you want to go grab a coffee, I’ll meet you back here.

Intelligent automation (IA)

Sometimes called cognitive automation, IA is the use of automation technologies – AI, BPM, RPA – to reduce costs and streamline repetitive and routine tasks. Learning and adapting with increased data, efficiency improves over time.

Intelligent character recognition (ICR)

ICR is an advanced OCR that uses AI and pattern recognition to interpret and convert printed or handwritten text into machine-readable content.

Intelligent document processing (IDP)

IDP is document workflow automation technology. It uses AI, machine learning, RPA, computer vision, deep learning, and NLP to capture, extract, analyze, and process data from multiple types of documents – paper documents, PDFs, Word docs, spreadsheets, etc. 

The primary goal is to extract valuable information from large sets of data without manual intervention.

Intelligent process automation (IPA)

IPA, also called hyper automation, is the use of AI, OCR, process mining, and RPA to streamline and optimize business processes, enabling automation with a higher degree of decision making and adaptability.

Invoice compliance

A set of standards used by businesses to ensure their invoices are accurate and compliant with regulations. Standards can vary depending on the industry, but typically they include requirements for documentation, reporting, and payment.

Invoice ingestion

Invoice ingestion is when 100% of an invoice is captured. Touchless invoice ingestion is when the data is extracted with automation and it’s added to your ERP system without human intervention.

Large language model (LLM)

Large language models are a specialized type of AI that’s been trained on large volumes of data so it’s able to recognize, summarize, translate, predict, and generate content.

Low-code

Low code is a software development approach that means people without coding knowledge can build applications and processes.

Machine learning (ML)

A field of artificial intelligence which uses data and algorithms to imitate the way humans learn. Improving accuracy over time. For example, trained on images of your logo, the algorithm would be able to recognize and find all instances of it on social media.

Manual data entry

Manual data entry is the process of a human inputting information – customer details, invoice data, etc. – into a spreadsheet, table, computer system, or database, without the use of automated tools or processes.

Metadata

Metadata means data about data. It makes searching and working with data easier. Allowing users to sort or locate specific documents. Examples of metadata include source, type, owner, creation date, date modified, file size.

Natural language processing (NLP)

AI designed to understand and interpret human language, using a mix of computational linguistics – rule-based modeling of the human language – and statistical and machine learning models. The models are trained to break down a piece of language – written or spoken – into machine-readable data. NLP can recognize, understand, and generate text and speech.

Nested values

Nested values refer to data structures in which a set of values is embedded within another. Often seen in lists. Data is organized hierarchically.

Neural network

A neural network, which has its roots in AI, is a series of algorithms that processes data in a similar way to the human brain. Using interconnected nodes or neurons in a layered structure that looks like a human brain. Computers can use this adaptive system to learn from mistakes and continuously improve.

On-premises software

On-premises software is installed and operated from a customer’s in-house server and computing infrastructure, or a third-party data center. It typically requires a software license for each server/end user, with the customer responsible for security, availability, and management. 

On-premise software is generally more expensive than cloud or on-demand software, and needs support and maintenance from your IT team. The risk of data loss is high, compared with a cloud-based IDP solution that keeps data backed up. 

Scalability can also be an issue. As a business grows, more storage space will be needed which requires new hardware installation and resources to build a new system.

Open-source software

Computer software that’s developed and maintained via open collaboration. Publicly accessible for anyone to use, modify, and distribute. Typically, it’s free. Open-source software examples include Mozilla Firefox, VLC Media Player, Linux OS, Google Android, WordPress.

Optical character recognition (OCR)

Sometimes referred to as text recognition, OCR converts images of printed, typed, or hand-written text into editable and searchable data. The software scans an image-based file – physical document, PDF, jpg, png, etc. – and translates it into a text-based file – Excel, Word, etc.

While it can eliminate the need for manual data entry, there are several headaches. A lower accuracy rate, limited language and font support, formatting errors, resource intensiveness, dependent on image quality, lack of contextual and semantic understanding.

More information in What is OCR Technology?

Process mining

Every time a process is completed, data is created. For example, when a purchase order is raised with, data will include the time it was received, who managed the order, who approved, the time it took to complete, etc. Process mining looks at this data to understand trends – time taken, handler, etc., to find areas that can be improved.

Prompt injection attacks

Prompt injection is a vulnerability that impacts AI and machine learning models centered on prompt-based learning. An attack happens when a user’s input attempts to override the prompt instructions for a large language model. The attacker hijacks the prompt to make it do what they want.

Rossum’s Transactional Large Language Model (T-LLM) integrates the power of LLM tech while prioritizing enterprise-grade safety. This ensures there are no hallucinations, data leaks, or prompt injection attacks.

Procure-to-pay (P2P)

Also called purchase-to-pay, the P2P process covers the end-to-end workflow from the need to purchase goods or services, to choosing a supplier, and finally, payment. The P2P process is usually performed by a procurement team and accounts payable team.

Thinking of automating your P2P process? P2P Automation Drives More Value From Every Transaction.

Return on automation (ROA)

ROA quantifies the benefits and efficiency improvements from implementing automation technologies in business processes. Identifying the increase in productivity, cost savings, and accelerated performance.

Robotic process automation (RPA)

RPA is a process in which a software bot uses a combination of automation, computer vision, and machine learning to automate repetitive, high-volume, and rule-based tasks. Eliminating the need for human intervention.

You might find this a fun article… RPA vs IDP | When Intelligent Document Processing Falls Short

Rossum Aurora

Rossum Aurora, our next-generation AI for transactional documents is powered by our proprietary Transactional Large Language Model (T-LLM).

It quickly learns from user feedback to adapt to new document types and custom fields. Achieved through an ergonomic user interface for fast Human – AI collaboration. To reach higher accuracy levels in fewer documents than before. No matter the complexity.

Our advanced AI is built on the most extensive data set of annotated transactional documents and uses 50x more parameters than the platform’s previous generation AI. 

Rossum Aurora – our next-generation AI for transactional documents.

Semi-structured data

Semi-structured data doesn’t follow a tabular structure, but it includes tags and metadata. It can be in the form of XML, JSON, TCP/IP packets, zipped files, web pages, or CSV files.

Service level agreement (SLA)

An SLA is a contract that sets the expectations between the service provider and the customer. Outlining the products or services to be delivered, the point of contact for end-user issues, and the metrics that’ll be used to monitor and approve the effectiveness of the process.

Straight-through processing (STP)

When a document is completely processed by an intelligent document processing software, without the need for manual intervention or validation.

Structured data

Structured data is organized in a predefined format. Such as a database table. Examples include phone numbers, banking/transaction information, product databases, CRMs, invoicing systems, etc. It’s way easier to process and can be queried using structured query language (SQL).

Supply chain management (SCM)

The handling of the entire production flow of goods or services. From raw material sourcing to production, logistics and delivery.

Structured query language (SQL)

Programming language for organizing, managing, and retrieving archived data from a computer database.

Task automation

The use of automation technology to perform repetitive, time-consuming, or manual tasks that are normally tackled by humans.

Task mining

Task mining is a process mapping tool that captures user clicks and keystrokes. So employees’ performance can be analyzed and optimized.

Template free

Template-free document processing uses advanced technologies – deep learning, machine learning, natural language processing – to extract data from documents without needing to create new templates for each new document structure.

Time to value (TTV)

Time to value realizes the effectiveness of an investment. When we’re discussing intelligent document processing, that would be the time it takes to reach an optimal level of accuracy when extracting data. A slow time to value will decrease the return on investment (ROI).

Transaction documents

A legal record that contains the details of a financial exchange between parties. Examples of transaction documents include invoices, purchase orders, receipts, bank statements, bills of lading, insurance policies, payment confirmations, account updates.

Transactional document automation

Rossum’s transactional document automation platform is powered by next-generation AI, and is the first of its kind. It automates the entire transactional document process end to end. From reading and answering emails to document approvals.

It’s cloud-native, trained on the market’s largest and ever-expanding dataset, ensuring our AI only gets smarter. Specifically designed for transactional documents, our proprietary LLM technology guarantees unparalleled accuracy and speed.

Transactional large language model (T-LLM)

Rossum’s tailor-made LLM that powers Rossum Aurora AI. Our T-LLM understands transactional documents with enterprise-grade safety embedded at its core. While it provides all the benefits of LLMs, there’s zero hallucinations, data leaks, or prompt injection attacks. And, it achieves human-level accuracy super quick.

Unstructured data

Unstructured data doesn’t have a predefined format. Not held in a structured database format. It doesn’t follow conventional data models, and is usually text-heavy with dates, numbers, etc. Examples of unstructured data include invoices, emails, ticker data, surveillance data, weather data.

User interface (UI)

A user interface is how we interact with computers, smartphones, and other devices. It can include mouse, keyboard, screen. Websites also have a UI. The design of which should be user-friendly. 

The user interface works hand in hand with user experience (UX). The result should be easy to navigate, intuitive, and visually pleasing.

Workflow automation

Workflow automation optimizes processes by replacing all the manual tasks involved with IDP software. Work can be standardized, human errors eliminated, and efficiency increased.

Workflow management

Workflow management is how to organize and oversee each task that needs to be completed to produce a specific outcome for a business. Analyzing each step in a process will highlight areas for improvement.

Any intelligent document processing definition I’ve missed?

Did I miss any IDP terms? Happy to add forgotten ones or new ones as they land.

Now that you have this intelligent document processing definition glossary at hand, I’m hoping you can chat with IDP vendors and not be overwhelmed with the language. Rather, you can discuss all the ways AI document processing will benefit your business, team, and customers.

Like to see intelligent document processing in action?

Rossum's transactional document automation platform can automate your document workflows end to end. Take a look!