The Bank of Russia will introduce artificial intelligence to process appeals from citizens and legal entities

In February 2024, the Bank of Russia announced a request for quotations for the creation of a system for robotizing business processes with an initial price of about 55 million rubles^[1]. It follows from the published terms of reference that we are talking about the robotization of such processes as consideration of appeals from citizens and legal entities, as well as support for consumers of financial services using artificial intelligence (AI) and machine learning.

At the moment, the processing of the main flow of appeals with the Bank of Russia is carried out using the application of the appeal. Documents are received from the automated document management and document management system in an unstructured form, and various data types and file formats can be used. The handling of the files attached to it is processed by the hub operator, who enriches and typologizes them, checks information about the applicant. Based on the results of this work, routing for non-typical cases, a response template for a typical case, redirection of the case with a cover letter, etc. is determined.

In turn, the business process designed within the framework of the project involves the use of the AI module for preliminary processing of incoming requests before their distribution to operators. The results of the AI module are planned to be used to enrich and route calls, as well as to offer operators the most likely values ​ ​ during manual processing, the terms of reference explains.

𝓲

Фото: Freepik/Rawpixel

Based on the information on the access and the attached files, the AI module should return information on the proposed classification of the access, including on additional issues of deep typology, as well as the likelihood of assignment to each class. In addition, key and additional entities of the case extracted from the text and attached files, a list of attached documents and their attributes should be returned. And based on the data received, the final typology and routing of calls, as well as the processing mode, will be determined.

The AI module should provide processing of the flow of requests at the rate of at least 2 thousand requests per eight hours with an average processing time of one request of no more than 15 seconds.

As part of the project, you separately need to design a business process for maintaining, configuring and training the AI module, including metrics about the need for retraining and choosing the best model.

The goals of the project are to improve the quality of primary processing, typology and enrichment of appeals in the application of the appeal. Based on the results of the project, the Bank of Russia will consider the possibility of integrating the AI module into its Internet reception room.

All work under the project takes 15 months, follows from the procurement documentation.

Notes