What document AI can and cannot do (yet)
Artificial intelligence supports the recognition, classification and data extraction from documents as well as the analysis of the information they contain. But: What does that mean exactly? What works and what doesn’t (yet)?
A proven rule of thumb is that an AI model can do document processing that could otherwise be done by a working student - i.e., the classification or extraction of information according to clear specifications.
The limit of feasibility for AI document processing lies in content evaluation - since the AI model has no content expertise, it cannot currently be used for complete subject evaluation, for example of doctor’s letters or essays.
What is already possible with document AI now
With Natural Language Processing based on Deep Learning, it is possible to process documents automatically. Document AI works at about the level of a working student. The text comprehension is there. Specifically, this means:
- All important basic data can be captured regardless of the position of the data in the document
- Information is correctly recognized in context - regardless of the wording
- The evaluation of information according to given criteria is possible
Thus, reliable extraction of basic data from invoices, purchase orders or delivery bills is possible - and even more complex information, such as contract clauses, can be extracted with the help of document AI.
The (current) limitations of document AI
AI technology can be used to support the review and evaluation of very complex issues - but it cannot replace a technical expert. It is not yet possible to evaluate content in a fully comprehensive, subject-specific manner.
This means, however, that when it comes to evaluating subject-specific content, a “human in the loop” is (still) important. This is the case, for example, in the final legal examination or medical appraisal.
Therefore, kinisto develops solutions that allow AI technology to be used not only for full automation, but also to support the process.
Examples from practice for achievable automation rates
- 98% - Recognition of change of name, change of address, change of court, liquidation and merger in commercial register announcements
- 96% - Recognition of company, address, contact person and contact data in freely formulated text notes from call center employees
- 95% - Recognition of contract type, contracting parties, contract duration and termination modalities in freely formulated contracts
Conclusion
Overall, with a mature solution for most processes, a degree of automation of between 90% and 98% can already be assumed - i.e. a direct saving of at least 90% of the effort compared to manual processing.
Outlook
With the help of artificial intelligence, many document-related tasks can be almost completely automated. The possibilities of document AI are constantly evolving and more and more processes are being automated.
In the case of complex content-related issues, we assume that document AI will primarily act as a “co-pilot” in the next 10-15 years.
The AI system for intelligent document processing
Completely integrated into your work process, the AI software kinisto can support employees - previously manual work steps on the document are automatically done by the AI. As a specialist for AI technology in practical use, we will be happy to advise you on the topic and on your specific case!
