RPA meets cognitive capture
Transactional capture has evolved into classifying, understanding, extracting and validating information from ever more complex document sets. Transactional capture software and services are now worth $2.3 billion, having grown by 11.4 percent between 2015 and 2016, according to our latest research. Including RPA, we are predicting it will grow to $3.6 billion by 2021—a CAGR of close to 10 percent. Originally transactional, capture was set up to eliminate key entry from standardized forms—applications, medical claims, surveys, remittance advices and (back in the early days) tax returns. In those days, a stack of the same forms (often single page) was prepared that could use predefined zonal capture, with exceptions out-sorted for manual processing. As simple form processes have become more digital in origin, more complex forms sets such as mortgage applications, property and casualty insurance claims, new account openings, invoices, shipping documents and contracts have become capture source material. Those types of transactional inputs require complex classification, which has started to leverage cognitive capture services and AI to understand what they are and what information is needed from them even if they have not been seen before. Advanced capture solutions (those we call Capture 2.0 systems) are designed to be modular and deal at a transactional level rather than a batch.
Next big step
But capture evolved from a batch process—convert the information on a bunch of forms to usable data. One challenge in the capture process has been setup time associated with “training” the system—skilled IT resources have had to scan documents and then set up rules and validation for the differing fields. A new form type required new setup. Although current capture solutions OCR a complete form and “read” field types, they need knowledge about which fields to capture and what validation is required, which is not very efficient. So, vendors such as OpenText are starting to track what a user does in setting up new data types in a similar manner to RPA. Cognitive understanding of the inputs being integrated further reduces the need for humans.
RPA started from a different premise—automate a white-collar clerical worker’s boring job of capturing data from one media to another by mimicking what the clerk does.
RPA has started to move from basic key entry and rules processing to needing to classify and understand the inputs more. Cognitive RPA is the next big step. But before that, companies need to recognize they must improve their OCR to deal with image files and paper. Most of the vendors I spoke with recently were using open source Tesseract, for their OCR solution, which is fine to start with, but to improve RPA effectiveness, a better OCR tool is needed. ABBYY has recognized that and is partnering with UIPath to provide not only OCR and capture, but also NLP-based comprehension. Magic Lamp, an IBM reseller, uses Datacap OCR and rules sets to validate the OCR. But none of the RPA vendors seem to have solved the issues of handwriting recognition or OMR capabilities.
Invoice processing has become a multimillion dollar business for capture vendors. It started with simple scanning and manual posting but has now evolved using OCR in conjunction with procure-to-pay systems with integration to the ERP systems to extract line items and validate and cross-check the billed amounts.