Document and Input Management: Facts and Status Quo
In a fully digital world, every form of communication between clients and corporations would take place in a digital space that would allow direct and immediate communication.
However, the bittersweet reality is a bit different and companies still need to deal with thousands of written documents via snail mail, live chat, or email.
Workflows haven’t completely adapted to digital transformation yet, and it will take a long time before every process becomes standardized and fully digital. There still are quite a few constraints and hurdles that make such a world still a far-fetched vision.
Especially when it comes to insurance or banking, paper documents, emails, and PDFs (and other forms of unstructured communication) still represent the norm.
Legal and operational constraints make it difficult for insurers or financial institutions to completely switch to a paperless ecosystem.
Therefore, despite an ‘anytime, anywhere’ mentality being the focus of digital transformation, there are still areas in which open and unstructured information is exchanged between companies and other entities they need to interact with.
Insurance companies in Germany currently need to process approximately 1.4 billion unstructured documents a year, that cover standard cases that go from contract amendments, authorizations, terminations, claims, and medical reports.
Let that figure sink in. 1.4bn documents a year!
It’s an enormous amount of information that needs to be received, dispatched to the right department, processed, and archived.
Customers expect a quick reaction to their requests, but, right now, it takes several steps for companies to deal with every piece of written communication and the process often takes days. And that’s not compatible with the real focus of digital disruption which is offering a frictionless and instantaneous customer experience.
Additionally, input management processes quickly become extremely costly as these involve several touch points and different people who have to interpret and convert documents into action.
German insurers report that processing unstructured information can end up costing up to €23 per document.
Needless to say, companies worldwide are trying to deal with this inefficiency and there are a few technologies that can help businesses cope with the situation by automating different steps in the process.
A relatively new solution to the problems connected to input management is a software category defined as Cognitive Process Automation (CPA).
Let’s try to understand what Cognitive Process Automation is and how it eliminates inefficiencies in input management.
What is Cognitive Process Automation (CPA)?
Cognitive Process Automation is an input management technology that transforms any kind of unstructured information (emails, written documents, pictures of documents) into structured and actionable data using natural language processing to interpret the information contained in texts.
Cognitive process automation software uses artificial intelligence and machine learning to train modules that can classify, extract, and interpret information contained in specific documents such as invoices, contracts, claims, or medical reports to convert any form of text or form into ready-to-use categorized data that can be immediately processed.
A Cognitive Automation Platform like ExB can integrate with any other existing input management, Workflow Management, or Robotic Process Automation (RPA) systems to automate the entire input management process and drive maximum efficiency while drastically reducing time to response and costs.
Basically, a Cognitive Automation Platform bridges the gap between structured (but limited in scope and place) and unstructured (open and flexible) communication allowing firms to immediately convert any sort of incoming text into pre-digested information for process automation without any coding and without forcing companies to change or adjust existing processes.
Let’s dig a little deeper and see how Cognitive Process Automation is different from other document management solutions. Let’s first have a look at other technologies and how they work to clarify how a Cognitive Process Automation Platform combines different elements to provide an end-to-end input management solution that can be implemented in any processes.
Document Management Solutions
OCR Solutions: Digitizing information
The first problem companies often face is digitizing documents they receive.
To avoid manual data entry operations, companies often introduce OCR software to convert non-digital documents into machine-encoded text which can be then processed electronically with standard text editors often embedded in the OCR software itself.
OCR stands for ‘optical character recognition’ and describes the process of converting scanned images of text into machine readable characters.
OCR technology varies in terms of accuracy and level of integration, but its scope is limited to transforming scanned documents into digitally processable files.
This technology is useful to convert invoices, orders, paper documents, or PDFs into editable formats that can be processed either manually or by other systems.
There are different sorts of OCR solutions which are usually implemented in a document management system.
Standard OCR solutions process and digitize text of an entire text, while Zonal OCR is implemented to translate specific areas in a document with a higher level of specialization and accuracy
Intelligent character recognition (ICR) also recognizes handwritten portions of text and can convert letters and numbers into digital characters.
Optical Mark Recognition (OMR) also recognizes human-marked data in documents such as tick boxes, signatures, or other optical marks in surveys, forms, or questionnaires.
OCR solutions have improved a lot over the past few years and are becoming extremely wide-spread in several areas but they only allow companies to convert scanned documents into digital data.
OCR solutions don’t understand the information they convert and don’t allow any form of automation unless integrated into other systems.
Visual Recognition Systems
Visual recognition systems are trained to recognize and classify images.
Such systems can be trained to interpret information which doesn’t fall into the category of text or forms.
Such systems can be highly specialized on a particular category of images and process for example faces for ID recognition or street signs for car navigation systems.
Visual recognition is sometimes defined as computer vision and when processing documentation it is a very important element that allows companies to recognize and categorize, for instance, stamps or images contained in scanned documents.
Text-mining is the process of extracting analytical insights from text.
Text-mining solutions use Natural Language Processing (NLP), Computational Linguistics, and Data Science to convert specific texts into analytics which can then be visualized in different forms according to the output format and additional system integrations.
Text-mining solutions can be used to classify and interpret large amounts of data collected from digitized documents and are useful for categorization (topic modelling), and for statistical or sentiment analysis.
Text-mining solutions are usually trained on particular tasks (based on objectives and input data) and can be implemented for information extraction, data mining, and Knowledge Discovery in Databases (KDD) to carry out text analytics operations.
Single-Process-Focused Text Capturing APIs
The challenge of automatically digitizing and interpreting documents spawned a series of highly specialized solutions that understand and extract information for input management automation.
Text capturing solutions may be rule-based systems or use machine learning and Natural Language Processing (NLP) to categorize and classify data.
There are plenty of solutions in the market that focus on extracting information for example from either orders, or invoices, contracts and other forms of unstructured information, or are trained for fraud detection and data validation.
Such solutions can be simple rule-based systems (quite inflexible and not very innovative) or pre-trained AI models that can be implemented in specific processes to feed input management or RPA systems. These tools can run on-premises or process data in the cloud and vary in terms of depth of comprehension, accuracy, and scope of implementation.
End to End Input Management with Cognitive Process Automation
Cognitive Process Automation (CPA) combines different document management elements to provide an end-to-end solution for input management automation.
A cognitive process automation platform can take any form of text input, from emails to forms, PDFs, scanned or photographed documents, understand the content, and convert it into viable and actionable information for automation.
The process starts with an OCR and visual recognition stage which allows the system not only to capture and process text but also hand-made optical marks, images, tables, stamps, and signatures.
The system then uses Natural Language Processing (NLP) to truly understand context and taxonomy for classification, entity recognition, and validation.
Basically, the platform processes documents exactly as a human being and manages to understand the meaning of different elements based on the context (for example differentiating the issuing date of a medical report from the date of the diagnosis or the beginning date of a treatment).
A CPA offers the highest degree of flexibility in terms of scope and application as the system uses Machine Learning in order to train specific models to capture and process any kind of document for input management automation.
The AI stage is followed by validation and verification stages which benchmark the data internally and with external databases to guarantee the maximum level of accuracy.
The quality of data for AI extractors is measured using an F-score (a quality indicator that goes from 0 to 1). But for a CPA platform, the output quality is measured in terms of Relevant Field Recognition Rate (RFRR) after the post-processing stage.
In this case, it’s important to emphasize that when comparing different solutions, the ML unit of a CPA solution alone can’t directly be benchmarked against specialized single-process-focused input management solutions or rule-based systems with a simple F-score.
A CPA system, though, will always win in terms of accuracy and efficiency if taken into account as a whole processes from pre-processing, to NLP, and post-processing.
Comparison table: document/ information management solutions
ExB offers a very powerful and flexible Cognitive Process Automation platform which hinges on the company’s Cognitive Workbench, a training platform developed by top neurolinguists and ML experts to generate highly efficient NLP models for input management automation.
ExB’s Cognitive Workbench uses state of the art deep learning and NLP technology to create “extraction thinkers”, specialized modules that understand human language and convert documents into structured data not only using semantic classification and taxonomy but also using positional and contextual cues to interpret data providing therefore a much higher degree of accuracy.
The Cognitive Workbench can train any sort of model independently from industry, language, and complexity but it is now being used heavily to train CPA models for input management in insurance (from policy management, to invoice processing and validation, medical reports, and more).
A CPA module can initially be implemented for a single use case either on-premises or as AIaaS, but the very same system can then expand into other areas to potentially cover all document management processes.
The system can, in fact, provide a fully-fledged end-to-end input management platform so that companies don’t need to implement several different solutions for different types of documents.
For every document a company needs to process, ExB can provide a specific module that can immediately be integrated in the centralized CPA system.
One company, one solution for the entire input management process!
ExB’s CPA technology fully understands context, document structure, meaning, and cause and effect relationship providing clear data and insights which can then be used for process automation.
Benefits of a Cognitive Process Automation Platform
The advantages of Cognitive Process Automation with an end-to-end input management solution are plenty. Compared to rule-based legacy systems, the platform is extremely adaptive and can easily cope with variability and unpredictability.
Additionally, post-processing and review activities allow operators to provide feedback to the system whose accuracy exponentially increases over time.
Companies can order specific ready-to-use modules but also directly gain access to the Cognitive Workbench. They can then independently create and train their own models uniquely tailored for specific business requirements, jargon, and business processes without any coding and with very minimal user training.
The platform guarantees the highest level of compliance in terms of output, as it can adapt to specific regulations that dictate data format and content.
A CPA platform allows companies to increase productivity, reduce time to response when customers submit changes or requests while also dramatically reducing costs for document management.
Therefore, the return on investment is evident in terms of both direct savings and immediate scalability (in terms of volume) and use case expansion (one single centralized solution for every kind of document that needs to be processed).
ExB is the pioneering technology provider in this field and defines the category of Cognitive Process Automation with its unique approach to input management.
To learn more about this category, our product, and how your company can benefit from implementing AI and NLP into your input management process, please reach out and talk to our solution architects.