In order to process information, computers need it in a structured form. All pieces of data must have strict meaning and format. On the other hand, humans can infer meaning from context and circumstances. We humans often use this capacity to simplify communication. Due to this gap, humans often need to “spell out” everything to computers.
For example, invoices. They come in many layouts. Usually a human needs to look at the invoice and type in numbers into an accounting software. This is pure transcription that rarely adds real value.
10 000 000€ salary savings for nurses alone
Many of us perform data entry as part of our daily lives. Did you know that medical researchers need to regularly copy data from field reports to databases? Nurses spend roughly 20% of their time writing down notes, or in other words turning them into structured, computer readable form.
We don’t need to worry about the SkyNet or whatever sentient AI. Computers are already dictating how we work; We are forced to serve them.
Recent advances in office automation, robotic process automation are running into this bottle neck: Automating many processes involves documents. More mature optical character recognition OCR, is just the beginning of the process.
The problem is acute and many players ranging between startups and cloud giants are building solutions for the problem. Many of them are relying on datasets of millions of documents and teams of data scientists working to solve one document type. This does not scale. This cannot quickly respond to changes in the operation environment.
Intelligent document processing with small data is efficient
We believe the best solution is to keep the data to a minimum and to keep the structure of the information at the center. The structure helps the system to find the essential information and improve the accuracy. The structure helps Digisalix to use the least amount of data, which in turn enables us to serve all use cases, not just the data-rich ones. Ultimately, clear structure simplifies automation work – results are in the language understood by computers.