May 04, 2017
OCR (Optical Character Recognition) software has come a long way since its original role in converting images of text characters to digital text. Today, intelligent capture solutions can convert text image to digital characters and THEN parse through the information it has created to identify data that is actionable or relevant to a process. For example, intelligent OCR solutions can identify and extract purchase order numbers from dozens of different invoices that make their way into an AP processing center.
Intelligent capture solutions themselves have continued to evolve. Whereas earlier ‘zone OCR’ solutions were depending on documents with standardized formats or regions/zones where they could extract information, today’s systems are designed to be flexible enough to learn and adapt to an ever-wider range of document formats and types.
That evolution will ultimately lead to artificially intelligent systems that can read, understand and extract information and insight from any unstructured data source.
Here, briefly, is how these systems have evolved, and where they may be headed.
The 1950s: OCR’s Narrow Focus
As early as the 1950s, forward-thinking companies took advantage of OCR technology to convert type-written text into data that could be used to analyze sales trends or help process payments. At that time, OCR was limited to handling type-written text only, limited to documents with very specific formats and within strictly defined regions of those documents. While the process was wildly innovative for the time, it was still a very manual, labor-intensive process.
At the time, cutting edge companies relying on OCR included Reader’s Digest, who installed the first commercial OCR system to convert typewritten sales reports into data for the magazine’s subscription department. Standard Oil applied OCR technology to processing credit card bill payments.
By the 1960s, OCR systems were used to read coupon codes and airline tickets. While the applications for OCR continued to grow, however, the same technological limitations applied—text still had to be machine-printed, a predefined location on every document.
The 1970s: Font Flexibility
In the 1970s, one of today’s pioneering minds on artificial intelligence helped to usher in the next age of OCR innovation. Currently Google’s Chief Futurist, Ray Kurzweil, developed a software product that could recognize text images in any font. Originally combined with text-to-speech synthesis to enable computers to read printed material to the blind, Kurzweil’s software led to the possibility of converting printed pages of all kinds to computer text. This allowed for the extraction of raw character or ASCII data from a page.
While OCR was now liberated from restrictions related to specific types of text, it was still limited in terms of what could be done with specific text or information within a document.
The 1980s: Imposing Order with Templates
To overcome OCR technology’s ability to identify and extract information from unstructured documents, solution providers attempted to artificially impose structure on those documents. First-generation OCR-driven data capture programs relied heavily on templates, meaning that data elements had to be contained in specific locations to be read, saved, and associated with specific structured data fields like a customer name or invoice amount.
While templates overcame some of the limitations of earlier OCR technologies, the process remained very rigid and labor-intensive. A template had to be manually developed for every specific business form or document, which meant that OCR was either limited to internally-generated documents, or required customers, partners or vendors to use very specific document formats.
The 1990s: Rules-based Systems
To break free of the limitations of templates, software companies in the late 80s/1990s developed solutions that could analyze capture text against ‘anchor words’ like “invoice” or “amount” to identify relevant information. While this approach broke free of standardized documents and templated forms, it was heavily rule-dependent and resulted in high error rates that required human review and intervention to identify and resolve exceptions.
More sophisticated second-generation products combined anchor word methods with the ability to analyze the same form multiple times, allowing a program to practice and “learn” in a way that reduced errors and ultimately improved future recognition. But even with these “memory recognition” products, error rates remained high, efficiency remained low, while the cost and complexity kept these solutions out of the hands of most companies.
The New Millennium: Intelligent Capture Emerges
By the turn of the century, OCR had gained a level of intelligence that allowed the technology to be applied more broadly, with a level of accuracy good enough to tackle high-volume, repeatable processes like accounts payable and sales order processing.
By leveraging pattern-recognition techniques and mathematical models optimized for reading and interpreting text as well as hand-written characters and bar codes, modern intelligent capture solutions can handle virtually any type of document, learning and adapting to improve efficiency and accuracy, while relying on integration with structure data systems like ERP software to validate results.
Over a decade into the 21st century, these systems have become more affordable and more flexible, so that companies of virtually any size can take advantage of intelligent capture technology to handle scanned paper or native digital documents.
Within the last decade, these systems have evolved into ‘smart process platforms’ that go beyond extracting information from unstructured files to automated an entire business process from start to finish.
In the case of accounts payable, smart process platforms and applications can transform high volumes of invoices in multiple formats into transaction data. Smart process platforms can then handle how the documents are sorted, filed or routed for approval by applying business rules to the data extracted from those documents. They can also intelligently match documents like quotes/orders/invoices/receipts and manage a wider and wider range of exceptions.
The Future of Intelligent Capture
As business and technology news focused on new technologies like robotic process automation and artificial intelligence, smart process platforms stand to evolve into self-learning systems that can more easily adapt to handle any document and any process. While these systems threaten certain low-level, task-oriented jobs, in reality these systems will be designed to work in concert with their human counterparts, relieving them of routine, repetitive tasks, while relying on humans to handle exceptions and to define processes and procedures for these systems to follow.
At Artsyl Technologies, we have been privileged to be a part of the evolution of intelligent capture and the empowerment of workers to focus on roles and responsibilities that make the most of their knowledge, skills and expertise. We are particularly excited about the potential for smart process platforms to drive more value from technology investments that companies have made in enterprise resource planning and enterprise content management systems, and to allow companies to focus more on their strategic goals and professional passions than on routine tasks.
In that regard, the future truly looks bright - and we look forward to working with our partners and customers to write the next chapter.