Intelligent document recognition. I understand the words, but what do they really mean?
Let’s deconstruct the phrase. Intelligent. Ok, so this stuff is supposed to be smart enough to tell one document from another. Document. we’re talking about documents; check. Recognition. This stuff recognizes documents.
Yet I still have trouble really getting to the essence of what intelligent document recognition (IDR) really is and why anyone should care. I blame it on us: the software vendors that use words like these.
If you go to document capture solution websites, you’ll come away thinking that IDR is all about using document separators for scanning a batch of different types of documents. Other websites make IDR seem like magic by describing it as technology that “works like humans.” Through sophisticated techniques, the software doesn’t need to rely on strict rules and can be “trained” to understand one document from another. Wow, that DOES sound “intelligent”!
But is it really intelligent? In most cases, the answer is no.
IDR is really about enabling folks to scan or import a variety of documents into a system without requiring them to do any prior sorting of the documents. Got a stack of paper? No problem. Just stack it in a scanner, push the button, and magic: Individual pages are put into the proper order and combined into documents. It might seem intelligent, but there is still someone telling the system what rules to use in order to sort pages and create documents. These rules can be based upon the overall layout of the document (the look) or specific content expected to be in the document such as a title, page numbers, etc.
And when you read words like “training” or “learning”, most often this capability relies very significantly on user input. A user tells the system when it gets something wrong and makes the correction. The system just stores this new very specific input and uses it the next time. But this type of “training” creates a system that is itself very rigid – not like a human at all.
There are some higher-end, more complex solutions that do incorporate more sophisticated artificial intelligence to create systems that develop inferences based upon both supervised and unsupervised training. These systems tend to be more flexible and can accommodate a broader range of needs. But they are expensive and complex. For the software to be really intelligent, it needs to monitor user behavior, collect derived data, and make adjustments automatically to document recognition and validation workflows: without user intervention.
Parascript is working on several projects that provide more intelligence through automated adjustments but that can be implemented without high cost or complexity. Keep your eye on this space for more on these developments.
As for the current state of intelligence in IDR, where you really want to invest your time is to understand the specifics of the type of IDR within a candidate solution and to find out if it works to solve your problems. If your basic question is “does it work?”, the answer to that is an unqualified “yes”. The reality is that you don’t need “intelligent” software to automate the identification of pages and documents. You just want it to solve your problems and be easy to use. Now THAT is intelligent!
Learn more about IDR: