Don’t Choose Between Automation and Accuracy

By now, every organization is deeply familiar with the value of data. When you lift up the hood of some of the world’s largest organizations, however, you may be surprised to learn that a meaningful percentage of data remains trapped in documents and other legacy systems. In fact, one of the biggest challenges facing businesses today is learning to mine these latent documents to access their rich troves of data – and to do it reliably, efficiently, and at scale.

Traditionally, organizations have relied on some combination of manual data entry and capture software to index and extract data from the billions of documents moving between businesses and customers. (For example, invoices will come into a central location and the A/P team will open the files, pull the relevant information and enter it into the system of record.) The problem with this approach, however, is that it’s slow, expensive and unreliable.

On one hand, the high costs associated with manual entry mean organizations can only afford to extract the bare minimum number of fields required, leaving valuable data and insights on the table. On the other hand, manual entry is error-prone, and most legacy extraction software requires pristine conditions to achieve accurate results, and also fails to produce a reliable measure of accuracy. This means that while a document with 100 fields might return an accuracy rate of 90%, in practice, users have no idea which 10 fields are incorrect. What’s more, most solutions measure accuracy at the character level, failing to account for the fact that a Social Security or bank account number is only useful if the entire field is correct.

When it comes to high-value transactional processing, organizations can’t afford to be wrong, so even with data capture or “automation” software in place, employees are still required to manually check transcription outputs or manually identify which fields are incorrect and must be fixed.

Whereas these other approaches force organizations to choose between accuracy and automation (high accuracy, low automation or low accuracy, high automation), with HyperScience, organizations don’t have to choose. By taking a fundamentally different approach to document processing, we’re able to classify and extract data from diverse document types with greater accuracy and automation than ever before. As an organization, we’ve developed and trained our own proprietary Machine Learning models to process documents as they exist in the real world (across structured forms and semi-structured documents like invoices).

Not only is the data extraction engine at the core of our system stronger, but our product’s built-in quality assurance mechanism is exceptionally good at knowing when it’s going to be right as well as when it’s likely to be wrong, sending exception/edge cases to an organization’s data entry teams to review and resolve. By combining human expertise with the latest in Machine Learning, the HyperScience solution allows organizations to unlock more data and higher quality data – the first step towards unlocking insights and staying competitive.


Copy: George Vallone

Back to Articles