Promsvyazbank "employed" Naumen Erudite neurobot

On January 29, 2019 the Naumen company announced end of the first stage of an implementation project of the virtual assistant on the Naumen Erudite platform in Promsvyazbank. Within this stage the chat-bot for customer service without involvement of call center specialists is created. He is able to answer frequently asked questions and to render a number of banking services. It is planned that the neurobot will answer clients not only in text, but also in voice channels.

At the beginning of 2019 the virtual assistant on the Naumen Erudite platform will be available to limited number of clients from a segment of SME on the website of bank in the test mode. But until the end of the first quarter 2019 with its help everyone will be able already to register business and to select a rate for service taking into account specific types and fields of activity. The chat-bot will help with reservation of the checking account, will prompt what documents need to be provided in office of bank, or will create the request for departure to the client for end of necessary formalities. Further implementation of model of a virtual assistant to voice canals of service is planned.

Within pilot testing the chat-bot should process not less than 10% of the entering addresses arriving from clients from a segment of SME. In the long term the virtual assistant is capable to provide to 30% decrease in load of call center operators.

The virtual assistant works at the Naumen Erudite platform which cornerstone technologies of neural networks and machine learning are. Feature of Naumen Erudite — the thought-over user interfaces providing convenient management of all cycle of robotic service:

• for preparation of training data (the interface of an asessor);
• for training of the robot in subject scenarios of service (including tools for training of a tematizator, a script editor of conducting dialog, templates of answers, etc.);
• for training of the robot in reactions to non-standard remarks (social jargons, the counter and specifying questions, interjections, expressions of a greeting, discontent etc.);
• for online and offline - control of efficiency of the robot (dashborda, reports);
• for analytics and work with historical dialogs with the extraction purpose from them training data.

Using the listed tools, business users can train and retrain the robot without participation of programmers and linguists. The product allows to load big arrays of historical data for the subsequent sorting according to subjects. Via the special interface asessor select training data from archive: analyze dialogs, select those which can be used for training, investigate them on remarks; then data are loaded into a system for training of the robot.

As an additional solution component the functionality of the product Naumen Omni-Channel is involved that will allow to apply uniform scenarios of robotic service of the addresses arriving by phone through a web chat on the website and in mobile application of Promsvyazbank. Depending on the channel on which the address arrived the robot will independently authenticate clients: or on the basis of the available data (at the address through mobile application), or requesting optional data in the dialog mode (at the address via the website or by phone). Using Naumen Omni-Channel it is possible to switch any channels (both voice, and text), integrating them in a uniform ecosystem.