The Company

Automation Hero was founded in 2017 by current CEO Stefan Groschupf, who also co-founded Datameer and Hadoop and built the predecessor of Elastic Search. That experience in the world of big data and search is reflected in Automation Hero’s products and approach to solving business problems.

Headquartered in San Francisco, the company is well funded, having privately raised a substantial amount from investors. Automation Hero makes some big claims, saying it is the most accurate AI automation platform and the best tool available to recognize handwriting.

The Technology

Automation Hero markets its platform as an end-to-end document process automation system. It includes, among other functionality, no-code automation, process mining, process orchestration, and intelligent document processing (IDP).
This report will focus on the platform’s IDP capabilities.
At heart, the company is an AI firm focused on making unstructured data actionable, particularly documents that play a part in large and complex processes (see Figure 1). Let’s look under the covers.

From the bottom up, we noted that Automation Hero, reflecting its big data heritage, has been architected to scale and is able to support 1,000-plus servers running in parallel. It has also been architected to provide a platform to design, deliver, and reuse microservices, and it can support 60 languages. It’s here that the big data roots show through, as Automation Hero leverages high dimensional data, a technique more commonly seen in areas such as gene analysis. It also leverages a modern transformer approach – well suited to advanced natural language processing (NLP) and natural language understanding (NLU) – rather than traditional recurring neural networks (RNNs).

At the document processing core, as always, lie optical character recognition (OCR) and NLP. Unusually, the OCR at Automation Hero is their own patent pending, contextual OCR, a combination of computer vision and predictive and contextual AI, with the outputs organized and linked via the use of a knowledge graph. We say this is unusual as almost all the competitors see OCR as a “solved” problem and leverage standard, commercially available third-party OCR tools. This approach alone is novel, and combined with Automation Hero’s emphasis on thorough pre- and post-processing of data, it makes for a compelling argument.

What this means in practical terms is that Automation Hero is doing something quite different from the legion of IDP vendors, long established companies and startups alike. It is applying lessons from the big data world to processing unstructured data at a huge scale. However, with such scale and complexity one expects it to be a beast to configure and manage, and there is no escaping the fact that a lot of technology is at play here. To its credit, Automation Hero has provided a no-code agile development environment and a range of easy-to-implement connectors, to considerably lessen that workload. In addition, it has built in a good deal of Human in the Loop (HITL) capabilities and largely avoided “black box” functionality, meaning that in practice, you can utilize a decision screen of sorts to see what is happening and why.

Without doubt, some very advanced and differentiated technology is at work here, but so what? Well, in practice, this kind of technology could be, and already is, readily deployed for traditional IDP use cases like contract review and claims processing. But it has the potential for more complex use cases such as financial or supply chain analysis beyond number crunching.