How can your company
benefit from hyper-automation?
By incorporating hyperautomation into automated processes in tasks that previously required human intervention, we can achieve not only faster execution, because they are executed by a machine, more efficiently and working 24/7, but also with much fewer errors because, as we know, repetitive tasks introduce errors due to fatigue of the people who execute them.
Employees not only stop carrying out tiresome tasks but can now focus their efforts on developing skills in tasks of greater added value for the company.
From Data to Insights
Applying OCR is a fundamental task when it comes to processing documents that generally come from a digitization process and must be converted to text in order to extract their information.
This tool has become so important for our solutions that in addition to having our own OCR, we train models capable of recognizing and removing watermarks, and all those problems that are grouped under the concept of noise (for example, stains on the pages) that confuse the OCRs.
In addition, we have pre-trained models for table recognition. This allows us to understand those data that are in a structured format, achieving a better understanding when extracting entities.
Object detection and classification
Not all image processing solutions require OCR. We also find cases where it is necessary to detect an object in the image and here the problems are also very varied.
For this type of case, and in particular if we focus on document processing, Mototech has developed various solutions that allow, among other things, to detect if the document has been signed, if it has been signed in the place designated for such function or if, for example, it is crossed out.
But we are not only limited to the legal area. In a completely different business spectrum, we developed a model focused on images that, based on photographs, is capable of recognizing fractures and classifying them as natural and induced; a task that until then involved hours of work and that we were able to solve in seconds.
Another successful case in object detection and classification is the model that Mototech developed through the detection of objects in satellite images, such as: parking lots. This solution helped one of our clients in the United States, who performs asphalt repairs, to increase their CRM database with new opportunities.
NATURAL LANGUAGE PROCESSING
Natural language processing is a branch of artificial intelligence that uses machine learning to process and interpret text and data. NLP technology combines syntax, semantics, and machine learning to identify sections, specific entities, and understand complex structures.
Tryni, for example, is a solution developed by Mototech that integrates a large number of tools and pre-trained models for entity recognition in structured, semi-structured, and unstructured documents.
The inflationary situation in Argentina and the problems it entails is well known worldwide: updating information on supplier prices has become a headache for more than one organization. With Tryni we train a model capable of recognizing and extracting the entities from the suppliers’ price lists and automating the updating of the data.
We successfully train models that allow, for example, one of our clients in the banking sector in Chile, to extract the data of partners, company name and participations from the processing of the corporate bylaws prepared by notaries when registering the formation of a new company
There are cases in the government area in which, for example, complaint systems are developed in which citizens must select categories corresponding to the classification of the claim and then include a text describing said claim.
This produces a large number of miscategorized complaints given the complexity of the classification structure and its categories (in general, only 20% are correctly classified) and generates delays in handling them that exceed 7 days.
For these types of cases, Mototech trained a model that consists of a large decision tree to classify claims and that considers the description of the claim entered by the citizen as the only input.
This solution is currently integrated into production in a test stage, that is to say that manual classification is still present, but it has made it possible to increase the rate of correctly classified claims to 70% and drastically reduce processing times, taking them from days less than a second, also allowing claims to be addressed in a maximum of 24 hours.
If you have an idea, it’s time to turn it into a reality & find the estimated cost.