Paper processing is tedious as it takes a lot of time and resources to manually enter all of the data into the system, but it is unfortunately a necessary evil for many companies. This is exactly why more and more companies are using OCR (optical character recognition) software to support the process. With traditional OCR, up to 80% of document workflows can be handled. It is able to recognize almost any variant of machine-made and clearly printed text, based on font and symbol recognition, but as soon as the text is smeared or crooked, it becomes difficult. This is the 20% in which a human being has to intervene again, which inevitably means that we are faced with three major challenges:

1. Inaccuracy: Typos and Handling of Exceptions (we’re all just humans)
2. Resources: It is difficult to find employees who are willing, and most importantly, able to extract text from low-quality documents.
3. Safety: The transition from machine to the human and back to the machine poses safety risks. This plays a major role, especially in strictly regulated industries with sensitive data, such as financial services, authorities or even organizations in the healthcare sector. 

If we are being honest, we need a better solution for these 20%. But before we get into problem solving, we should clarify an important question first. 

What exactly are low-quality documents?

A low-quality document can be a fax or a scanned document of poor quality. Likewise, it can also be a delivery note, a timesheet, or a form for patient registration, which has been filled out by hand.

Traditional OCR is no longer sufficient to extract data from these document types. AI-powered OCR, on the other hand, uses more advanced technologies, such as a highly skilled machine learning models and advanced computer vision engines. By combining these two technologies, we get an OCR software that is able to replicate the way people are able to read low-quality documents. If the model is good enough, it may actually be able to extract (handwritten) texts better than humans. But each model is only as good as the data set on which it is trained.

How do I find the right provider for me among all the different vendors?

Terms like ‘artificial intelligence’ and ‘machine learning’ are used too often and not every vendor can prove the functionality of their technology. Look out for providers with transparent numbers and ask yourself the following questions in advance: 

• Is the vendor’s solution AI-powered or is it just a well-marketed mix of human data entry and machine learning? 
• Can vendors provide accuracy for each process performed, as well as each document read and extracted? 
• Why is the provider active in the OCR industry and how much experience do they have in this field? 
• Does it offer a cloud-based SaaS solution or do you need to host the solution on-premises? 
• How long does it take to use the product? How much training is needed? Are special expertise required?

Fellow Consulting AG / PolyDocs GmbH started collecting data by humans. As a result, we have the largest human-audited dataset (over one billion fields) in the industry. We offer high accuracy from day one of product use.

That was a lot of information and a lot of questions for now! Take your time, think about it, weigh all the pros and cons and decide correctly – for DOC² (Polydocs) 😉