Russian Post implements voice assistant based on Yandex technology

Russian Post has launched its own voice assistant based on Yandex technology SpeechKit (one of the services of Yandex cloud platform) in order to handle requests from its customers hotline.

Implementation of the voice assistant will enable to increase productivity of the hotline on postal services issues by taking pressure off the call center agents. Currently, 30% of the calls are handled automatically. The voice assistant quickly finds necessary information in the databases, and the average time of each call has reduced twice: from 3 minutes when communicating with an operator to 1.5 minutes with voice assistant. It is especially effective while coping with peak loads, for example during increasing demand at New Year`s Eve or self-isolation.

The voice assistant already services the requests on 9 topics. It successfully assists in finding a specific post office throughout Russia, giving answers about work time of post offices, statuses of parcels and letters. Moreover, it helps to find out how to send a parcel promptly and to receive an item on the doorstep. Nevertheless, the voice assistant actively learns new skills.

“The dialogue of our voice assistant is approached to human interaction; we update its artificial intelligence every week. With Yandex SpeechKit Russian Post reached a new level of service quality in contact center service and became faster and more convenient for our customers” – notes Mikhail Volkov, Deputy Director General for Operations and Retail of JSC Russian Post.

The project on robotization of the contact center was realized by ATS Aero using Yandex SpeechKit technology of the cloud-based platform Yandex.Cloud. The voice assistant is able to hold a dialogue, which is approached to human interaction with the customer, and recognizes  92-95% of speech thanks to the specific language models and NLU (Natural Language Understanding) tools. For understanding of Russian-language interpretation of track-number digits and Latin letters a specific module for checking multiple hypotheses of voice messages was created, which utilizes several recognition models.

Source: JSC Russian Post

    footer picture