Financial statements of VTB Factoring counterparties process artificial intelligence

2022: Processing counterparty financial statements with artificial intelligence

• On IT GreenData March 9, 2022, the company announced that together VTB Factoring with it it has implemented functionality that simplifies recognition the accounting reports provided in a scanned form by replacing the manual entry of reporting data into the system with automatic recognition of PDFfiles graphic format files. One page is recognized in seconds, and the entire document is recognized in a few minutes, depending on the number of pages.

The solution greatly accelerates the process of converting documents into the required format for further processing in the analytical system. The neural network is trained to recognize reporting with an accuracy of 80% in accordance with the RAS (Russian Accounting Standards) standard. To develop functionality, GreenData used libraries for processing graphic information using recurring neural networks. Downloaded and recognized data is stored on VTB Factoring internal servers, which ensures their protection against unauthorized access, security and confidentiality. As a result, a customized product was created that takes into account all the requirements of VTB Factoring for a high degree of data protection.

"In the context of VTB Factoring's growing business and the increasing number of graphic PDF files provided with counterparty reporting, we faced the task of improving the quality of service. To optimize the processing and digitization of financial data, we decided to create an auxiliary automated tool with the necessary quality of document recognition and high speed. At the end of 2021, we began testing, and at the beginning of 2022 we integrated it for work in VTB Factoring. Automatic recognition of counterparty financial statements had a positive effect on the speed of processing and monitoring financial data, and, as a result, on the speed of making credit decisions, " commented by Igor Klyuev, Director of Risk at VTB Factoring

"Colleagues from VTB Factoring turned to us with the task of automating reporting from customers. Previously, customers provided them with a large amount of data in graphic PDF files. We showed them a test version of the functionality and, after its approval, began to implement the development of the tool with compliance with all business requirements of the customer. The key requirement was a recognition quality of at least 80%. To train the neural network, free-access data libraries were used - signs, letters, numbers and keys. The data quality is determined automatically when loading, and the user checks only the totals to make sure that the data is correct. Now we are continuing to work on optimizing the system so that it works even faster and more efficiently, " noted Maria Kovalenko, project manager from the GreenData

For the convenience of users, the quality of cell recognition is rendered using color. Cells with a recognition quality of less than 80% are highlighted, and then the checking employee of the company can make the necessary adjustments to the data. After decryption, the information is automatically structured in the system and loaded into the database for further work.

GreenData successfully cooperates with VTB Factoring in the automation of accounts payable and receivable financing processes since 2017.

As part of the joint work, a project was implemented to automate the calculation of risk models. It is the administrator's workplace, in which calculation cards with basic key figures and a python script are entered to determine the scoring result. The script processes calculated values of variables that are created in a separate directory using the algorithm designer to determine the value of each measure, and returns the result. The API for automatic calculation of the scoring result based on the data transmitted by the integrated internal service of the bank is also implemented.

Other significant projects include the creation of a system of individual monitoring of counterparties. As part of the project, a separate module with a monitoring card was implemented, which contains a set of indicators - covenants, financial and non-financial. This set of measures is unique for each counterparty.

Based on the data of the card, the indicators are calculated and monitored with a given frequency up to the daily check. In addition to the above, the module implements a notification system that automatically sends notifications and reminders to responsible employees, for example, if data is received and individual calculations are required or the deadlines for providing data for monitoring are violated. Also, the project implemented a report that can be generated on any date for the entire list of counterparties under monitoring.

This decision has greatly affected the individual monitoring process by accelerating and optimizing it.