HyperScience Delivers an Automated Solution to Efficiently and Accurately Transform Documents into Data

TL; DR: HyperScience is on a mission to help organizations process documents more efficiently by swapping manual data entry processes for advanced machine learning solutions. By reducing both errors and data-entry costs, the company enables its users to focus on enhancing customer service and driving new business opportunities. With more than $50 million in funding and a strategic investment strategy, HyperScience aims to bring the power of automation to an even broader customer base.

Before moving on to greener pastures, I spent the early portion of my career putting together printed membership directories. It seems archaic now, but back then, clients would sometimes mail me handwritten membership information that I would manually convert to text.

Aside from being time-consuming, the process introduced the risk of human error — a frightening prospect in the world of print.

These days, I’d likely turn to a solution such as HyperScience, a machine-learning tool capable of capturing digital data from handwritten, cursive, and printed text on forms, invoices, checks, invoices, PDFs, and even low-resolution images. Using the power of automation, the technology effectively eliminates manual processing, increasing productivity.

“Classifying and processing documents is still a very manual, painful, and expensive process for today’s organizations,” said Peter Brodsky, HyperScience CEO and Co-Founder. “Businesses spend $60 billion each year on data entry, and that figure is only getting larger. HyperScience solves this by using the latest in machine learning to unlock and lift data from diverse documents.”

The headache-eliminating tool is easy to set up, implement, and maintain, with access to an optional API for easy integration into existing workflows. Over time, HyperScience’s built-in quality assurance mechanisms ensure that the highly accurate system becomes even more so via advanced machine learning models.

The technology also reduces human error and data-entry costs, empowering users to focus on what’s most important: driving new business opportunities. Moving forward, HyperScience will leverage its $50 million in funding to pursue strategic investments, bringing the power of intelligent document processing to an expanded user base.

Peter Brodsky, Krasimir Marinov, and Vladimir Tzankov founded HyperScience, based out of New York City, in 2014. Prior to that, the founders had spent nearly a decade working on machine-learning projects involving complex Extract, Transform, Load (ETL) data processes.

These jobs weren’t exactly satisfying. In an article on the HyperScience site, Peter referred to ETL as “mind-numbing, soul-crushing, awful, horrible, terrible work,” that “requires high levels of domain expertise and delivers excruciatingly negative levels of job satisfaction.”

So, upon founding HyperScience, the team set out to automate their old jobs using first-hand knowledge to build a more intelligent solution. They also took into account the document processing challenges that exist in the real world, such as handwriting and skewed or stretched scans of paper documents.

“At the time, no robust, reliable automation platform existed,” Peter said. “Instead, companies relied on outdated data-capture technology and teams of data keyers,” Peter said. “HyperScience took a fundamentally different approach, building a proprietary machine-learning solution that delivers high rates of accuracy and automation out of the box — and continues to get better over time.”

Since its founding, the company has expanded significantly, with a team of more than 100 employees and offices in New York, London, and Bulgaria.

Today, HyperScience’s machine learning platform helps organizations spanning the globe and across industries — from finance and insurance to healthcare and government — reduce the costs and errors associated with manual data entry.

Peter told us that organizations that implement HyperScience typically enjoy a range of benefits, from time savings and higher productivity rates to the ability to operate with agility and boost ROI.

“The HyperScience platform helps decrease costs and errors associated with data entry while freeing up users to focus on activities that drive the business forward,” he said. “Companies that choose HyperScience may see increases in capacity of up to 10 times, as well as up to six-hour reductions in service-level agreements (SLAs).”

This means more reliable processing and faster response times for customers of all types — whether they’re business partners, internal customers, or simply individuals looking to open a brokerage account. Benefits like these are the product of HyperScience’s precision, with accuracy rates of more than 98% on the first day and continued improvements over time.

“Documents are messy, so we’ve built a solution that classifies and extracts data across diverse inputs and even low resolution, distorted images,” Peter said. “For example, we know that a Social Security number is only valuable if every digit is correct, and we read documents accordingly — with context — so we can deliver higher accuracy.”

When it comes to features, one of Peter’s personal favorites is the HyperScience supervision platform, which provides guidance on how to handle flagged data known as exceptions. The lightweight, intuitive technology is as easy to use as it is functional.

“HyperScience is exceptionally good at identifying when it’s likely to be right as well as when it needs help. It sends edge/exception cases to an organization’s data entry teams to review and resolve them, which in turn fine-tunes the underlying model,” he said. “The way we do it, however, speaks to our easy-to-use product ethos.”

Significant advancements have been made in deep learning — a subset of machine learning involving artificial neural networks — and researchers are actively working to push forward the frontiers of knowledge.

Peter told us that HyperScience prioritizes investments in product and engineering to provide the team tools for testing new ideas and keeping up with emerging trends.

“By working at the leading edge of the field, we are able to experiment with many different things, some of which have turned into performance breakthroughs,” he said. “At the same time, data remains the key to Deep Learning, and we’ve been able to amass a proprietary dataset that is representative of the world and specifically tailored to some of the breakthroughs we’ve made on the model architecture side.”

Staying one step ahead of the competition is also a matter of keeping customers close to identify and solve their pain points. To that end, Peter said HyperScience is customer-obsessed. By working alongside clients, the company has been able to collect first-hand feedback, which is crucial to its product roadmap.

For example, it recently introduced a refreshed user interface, improved organizational tools, and French language support to better serve customers’ needs based on feedback from users.

“It’s not surprising, but ease of use continues to be a huge differentiator,” he said. “Personal and consumer tech has very much shaped enterprise expectations, and we work tirelessly to create a sleek, intuitive platform that is designed for nontechnical business users.”

As for what the future holds, Peter told us HyperScience is committed to helping organizations transform their workflows through the power of automation.

He said 2019 represented an exceptional year for the company, which in many ways is just getting off the ground.

“Since closing a $30 million Series B round in January, we’ve passed the 100 employee milestone, opened our second European office in London, and consistently achieved double-digit growth month over month,” he said. “We also look forward to attending (and hosting) more events and sharing our industry insights and expertise.”

In the meantime, HyperScience will continue to invest in the research and development needed to move the organization forward.

“We’ve taken significant steps toward our ultimate vision of making our platform input-agnostic — capable of extracting data from every document type (i.e., any structure and language) and flexible enough to adapt to any processing workflow,” Peter said.