The use of automation is on the rise in financial services because it saves employees from the burden of repetitive and tedious tasks. In addition, automation can be integrated with other learning models to solve more complex tasks.
- Automated ID identifier and interpreter
This robot can identify the type of a Romanian identification document, reads it using OCR, scrapes the useful data, validates it with known rules and outputs the data in a structured file. Then, each file is processed and moved to a specific folder.
If the robot fails to identify the document type or if the data is not cleared, the file will be moved to a special folder, where you can correct it manually.
- Automation of correspondence between company and customer
Correspondence addresses are split and automatically formatted into a table, which can then be used to set up correspondence with customers. Basically, each element from the address is identified (street type <street, bld, alley, etc>, street name, number, etc.) and formatted in a table with each element in its column.
This automation was made possible by an NLP – Natural Language Processing algorithm, more precisely with NER – Named Entity Recognition algorithms.
Moreover, the process is divided into 3 parts:
- Address running and reading, then splitting by elements,
- Manual validation – where the user can correct addresses that are not trusted enough,
- Collection of corrected data to re-train the address detection algorithm to improve its score with more examples.
- Contract processing
It is a process of reading with OCR – Optical Character Recognition of contracts in PDF format and dividing them into templates, because each template must be processed differently.
This automation also performs the following tasks:
- Identifying the ID and extracting information from it.
- Extraction of key data from the contract (number, date, person, email, etc.) and generation of a centralized table with the extracted information.
- Processing of court decisions
After scanning the document, key data (court, type of decision, etc.) is identified, the most important to extract being the status of the decision (suspended, refused, admitted, admitted in part).
Often, the status is not written in a clear format, but must be extracted from the context. So, the robot will process all the text and it will assign a known status to the decision.
- Payment’s verification and comparing the extracted data
As part of the document processing, the robot checks whether the payments in the Excel file received from customers match those in the Excel file generated from the company’s internal database. Then, it generates a pivot table for each piece of information extracted, then it compares the pivot tables.
Using our automation, companies have been able to:
- Reduce operational costs
- Reduce human error rates
These benefits enable companies to:
- Grow Faster,
- Focus on more important activities, that bring profit to the company.