The Company

Hyperscience was founded in 2014 in Sofia, Bulgaria, and is now headquartered in New York City while maintaining a substantial presence in Sofia. Currently led by Interim CEO Charlie Newark-French, the company employs approximately 300 people and has revenues of around $40 million. It has raised $300 million to date and is at Series E funding.

The Technology

Hyperscience is best known for developing and launching a modern approach to optical character recognition (OCR). However, it does much more than provide OCR functionality. Hyperscience offers an advanced platform to extract and process unstructured information, and additionally offers a specific application for intelligent document processing (IDP).

The Hyperscience platform consists of a series of pre-defined, yet customizable, modules or “Code Blocks.” Each block contains specialized machine learning (ML) components that can be linked to create a customized end-to-end document processing solution. Some of the blocks available are standard; for example, blocks to input, classify, identify, transcribe, route, etc. Others are highly specific, such as a “Box Folder Listener” that monitors and, when needed, retrieves files from Box folders. Others simply connect to systems such as Salesforce, Pega, or Bizagi. From this still-growing library of Code Blocks, a user initially works within a module called Flow Studio. As the name suggests, this is a visual interface to build simple workflows consisting of a series of Code Blocks (see Figure 1). To give Hyperscience credit, this is an intuitive and easy-to-use design environment.

In summary, this is a core platform that typically runs on VMware (either in the cloud or on-premises) and incorporates pre-trained AI components including extensive deep learning elements alongside basic file storage. Hyperscience can integrate with a wide range of data sources via pre-built connectors and APIs. It supports a range of different languages including Korean, Arabic, and common European languages, and comes with extensive and detailed real-time reporting capabilities. Layered atop and leveraging the core platform are a range of pre-defined building blocks that can be accessed and sequenced in the Flow Studio environment. In our analysis, this is a simple and very effective approach, as traditional IDP systems can be highly complex to configure, code, and manage.

Here’s the thing, though. All IDP systems do the same thing, or at least have the same goal, which is to extract, understand, and process unstructured documents. As a result, most are, under the covers, similar both in approach and underlying technology. Hyperscience, though, differs noticeably from the pack in its conceptual approach to IDP as well as its extensive use of AI. What is this conceptual approach? Almost all modern IDP systems (indeed, all back-office AI products) involve Human in the Loop (HITL) whereby if the AI is unsure of something, it will flag it for a human to correct or clarify. Once the human does so, the system learns and uses that knowledge for future similar exceptions. Hyperscience embraces the use of HITL far more extensively than others we have seen. Rather than simply utilizing HITL to correct errors, Hyperscience is built to ensure that humans and the machine work collaboratively.

The company uses the term “Human Centered Automation,” and this is more than a catchy marketing term. Hyperscience recognizes that though AI and machines in general can work at speed, at volume, and with a high degree of accuracy, there will always be break points and limitations, for example when a machine reads handwriting. Hyperscience focused much of its early R&D on training its system through deep learning to recognize handwriting, and it does as good a job as any we have seen. But common sense tells us that no machine is going to read handwriting correctly all the time. Likewise, in many IDP use cases, a variety of different forms of documentation and language may be processed in parallel; for example, machine-generated PDFs alongside handwritten notes and human-generated documents. The Hyperscience goal is to bridge these divides and to normalize all the information into a form that is equally readable by a human and a machine. For complex document processes in areas such as insurance, healthcare, or financing, that strategy makes Hyperscience a much more approachable and human-centric system than one solely focused on machine processing. In our estimation this is a good approach that makes any potential IDP deployment less risky and easier to execute and scale than more traditional methods.

In our analysis the two obvious differentiators for Hyperscience, in what is otherwise a crowded marketplace, are these: 1) Compared to many of its competitors it is very easy to use and deploy, and 2) it does a good job with reading/decoding handwriting, typically a weak area for IDP products. The biggest differentiator, though, is its human-centric approach to automation.