There is an old adage that “everything old is new again,” and document imaging industry veteran Greg Council thinks he may have already seen some of the new “innovations” sweeping the market these days. This includes AI, RPA, and even crowdsourcing. We recently caught up with Greg, who is the VP of marketing and product management for automated recognition ISV Parascript, and he shared his thoughts on how these innovations have matured and started to have an effect on our industry.
“I think what you are seeing is the introduction of a lot of new names for things,” Council told DIR. “But, it’s still important to get some clarity around all of this and see how it fits together.”
Council compared RPA to technology that was historically used to issue sales tickets or set up new employees. “This was customized software provided by firms like EDS that would automatically route sales tickets to the right people or if you hired a new employee, it would provision a PC and e-mail account and whatever else you might need,” he said. “Today’s RPA then really got its start in IT automation, in areas like provisioning databases and automatically configuring other things based on well-known rule sets.
“The new trick that RPA applications use is the ability to learn by example. They typically already have APIs in place to connect to multiple applications, and then they record the actions of subject matter experts and create bots for executing the same process every time.”
Council noted that one major difference between RPA and document capture, which is also designed to reduce manual labor and keystrokes, is that today’s capture still typically requires manual set up and it has to deal with exceptions. “If you look at the future vision for document capture, it will have better up-front learning by example, but that’s going to involve the incorporation of more AI,” said Council.
According to Council, AI has actually been employed by recognition ISVs for many years. “AI is really a bunch of different technologies, but in this market everyone is focused on machine learning and deep neural networks,” he said. “If you talk to our CEO, he’ll tell you the company has had deep learning and neural networks in its software for 20 years. In fact, all OCR and ICR developers have utilized neural networks to train their software in the recognition of characters. Of course, all that has mostly been in the lab—after the product comes out, the users don’t continue to train it.
“But, if you look at the capture market about 10 years ago, some of the bigger vendors started to introduce a learning process into their applications. They would basically take the data entered during an exception handling process and use it to create a new rules set. I can’t say how well it performed but this was a precursor to the use of AI.”
Last year, Parascript announced that it would be incorporating a neural network in its FormXtra software to help set up capture applications. The methodology involves training the neural net on a set of documents and a set of truth data that a user wants extracted from them. Due to the complexity of this type of operation, Parascript initially was offering to do the set up for its customers, but eventually the plan is to enable users to do their own set ups [see DIR 2/17/17].
And removing the complexity from capture through techniques like AI is how Council believes the technology will be able to fully leverage potential avenues for growth like the cloud and integration with RPA. “Capture has had a scalability problem,” he told DIR. “When people started out 20 years ago doing imaging and indexing for archiving, it was fairly simple. But, as capture has expanded to cover more needs, the curve of complexity has gone up. To the make the technology more accessible to a wider market, including organizations like mid-sized businesses, capture vendors have to make it simpler. Utilizing AI and machine learning to make the upfront configuration more automated is one way to achieve this.”
Council believes this simplicity must be baked in before capture vendors can fully reap the benefits of the cloud. “If you look at EFSS applications like Box, they didn’t start out by trying to put ECM in the cloud,” he said. “They made their software user friendly from the beginning and then started adding more complex capabilities once they were established.”
Council believes some of the crowdsourcing capture applications he’s seen have figured out the UI challenges. “They have had to create extremely nice front ends because they are distributing fields and tasks to a general audience,” he said. “That is one level. The second level is that capture has to be offered as a service that can be consumed by other applications. This includes RPA applications where you want to add more document-based workflows. But, it also includes line of business applications like CRM systems or accounting systems, from which a user might want to execute a document process, like capturing data from an invoice. For example, we are talking with a payroll provider that wants to incorporate our capture technology within its product roadmap to capture certain forms.”
Kofax has an OEM agreement with Microsoft through which is makes its cloud-based Kofax Invoice Capture Service available to Dynamics NAV customers [see DIR 6/20/2017]. Council explained that, in contrast, Parascript’s customers usually want the extra control provided by incorporating Parascript’s recognition tools into their own cloud. “Even though we have a cloud service, when we talk to solution providers and they find out we have an SDK, they typically want to go that way,” said Council.
Since he brought up crowdsourcing, we asked Council if he thought there was the possibility that crowdsourcing technology could become so advanced, it could marginalize the need for automated data capture. “Crowdsourcing is definitely a threat to traditional service bureaus,” he said. “It’s really offering the same service in a more accessible way. But, there are still cost savings to be gained through automated capture, especially when you start asking for rapid turnarounds like you are seeing more often in the market today. Even a crowdsourcing model can start to get cost prohibitive when you are looking to turn something around in less than 24 hours, not to mention an hour.”
For more information: https://www.parascript.com/blog-listings/