Modern Architecture Key to Haystac’s Potential

0
Anthony Macciola has been in the document capture market for more than 25 years. As you can imagine, he’s seen quite a few automatic document classification and extraction offerings. So, why is he excited to have recently joined Haystac as Chairman of the Board?

 

“On the surface, Haystac’s technology looks similar to what you and I are familiar with relative to KTA and some other capture products,” Macciola told DIR. “However, when you look at the way Haystac is implementing its technology, it’s much more contemporary than most of what is on the market already. Haystac is leveraging deep learning, and its technology is much closer to being born on the cloud than traditional classification and extraction.

 

“A lot of legacy capture products have been in the market for several years, and they all started out with rich client/server architectures. For the most part, they haven’t gone through a technology refresh. Haystac’s architecture is newer and more contemporary. And to me, that represents opportunity.”

 

According to Macciola, Haystac’s deep learning capabilities make its software easier to deploy, train, and consume. “We all know this type of technology has had challenges related to training and configuration,” said Macciola. “And once its implemented, there is maintenance that needs to be put in place to ensure it maintains its accuracy.

 

“The advantage of using deep learning boils down to increasing the level of graciousness with which a classification and extraction engine can run. It’s about reducing the number of samples, along with the rigidity and reliability of the results and how tolerant they are. From the user’s standpoint, deep learning makes the technology more flexible and forgiving and easier to learn.”

 

Haystac is based in Newton, MA, just west of Boston. It is a spin-off of Solaris Development, a custom solutions shop focused on the healthcare industry. Solaris was founded in 2003 and run by Barak Tsivkin, who is now CEO of Haystac. In 2016, Solaris was sold and the proceeds were invested in launching Haystac. “At Solaris we did a lot of work for hospitals in the Boston area,” said Tsvikin. “In 2007, we started working on the machine learning and AI technology we have leveraged at Haystac.”

 

According to Tsivkin, Haystac utilizes a cascading approach to capture. “We have several technologies embedded in our platform to address different stages of processing,” he said. “For example, we’ll look at the layout of an image and see if we can classify it that way before applying OCR. Transforming documents is not always a good thing, as it creates more margin for errors. At the bottom of our technology, we are utilizing computer vision and feeding information to our machine learning engine.”

 

Tsivkin added that this hybrid approach to recognition makes the technology especially strong when applied to poor quality images. “We get less false positives than anybody else,” he said.

 

Macciola offered Haystac’s browser-based design tool as an example of its modern architecture. “It’s very intuitive and easier to use than most of the older thick-client capture design tools on the market,” he said. “For this type of technology to gain greater adoption, it has to be simpler to deploy and train. If anyone doubts the impact of simplicity on today’s market, they just need to look at the changes mobile devices have brought to the business and IT worlds. One of the reasons that robotic process automation (RPA) is taking off is because it’s designed to be easy to implement with no training. You should be able to give RPA tools to business analysts and in two days they should be able to start building robots.

 

“Haystac’s architecture and underlying technology seem to touch all the check boxes toward making that type of thing happen in the capture market. There should be a bunch of players in the market taking this type of modern approach, but I’ve only seen a few. A lot of vendors have aging architectures and technology they aren’t willing to upgrade.”

 

Opening new markets
Macciola noted that Haystac is looking at applications where traditional capture has yet to be applied. “Haystac is looking at areas like risk management and fraud detection,” said Macciola. “Capture has been used a lot in the mortgage industry in areas like customer onboarding, but what Haystac showed me was a way to enable organizations that purchase loans to understand exactly what they might be getting in a packet. This involves clustering documents and extracting data that can provide insight into the associated risks of the mortgages.”
Macciola left Kofax this summer and recently accepted a position as the Chief Innovation Officer at ABBYY [see DIR 11/3/17]. “Barak’s and my paths had crossed when we were considering M&A opportunities at Kofax,” Macciola told DIR. “When he found out I had left Kofax, Barak reached out, and we had a discussion not knowing where the relationship might go. We decided that through a position on the board I would be able to provide some non-executive feedback, guidance, and direction.”

 

Haystac is currently in the process of trying to build a referenceable customer base. “Barak is focused heavily on building credibility and awareness for Haystac,” said Macciola. “From what I’ve seen, you need to establish about a half dozen strong direct accounts to give you credibility and allow you to move forward with other channels.”
Macciola said that eventually he could see Haystac’s technology being white labeled and/or sold as an OEM offering. “It definitely seems like something that could be embedded as part of a total solution,” Macciola said. “It has some enabling capabilities. First though, we need to get some large referenceable sites. Barak has an impressive pipeline to help make that happen. After that, the dominoes should start to fall.”
For more information: http://bit.ly/HaystacMacciola

You might also like More from author

Leave A Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.