RSS
Логотип
Баннер в шапке 1
Баннер в шапке 2
Project

100% of hotline calls Raiffeisen Bank processed by Ray's voice assistant

Customers: Raiffeisen Bank

Moscow; Financial Services, Investments and Auditing

Contractors: Raiffeisen Bank
Product: Yandex SpeechKit Cloud

Project date: 2020/03  - 2020/08

2021: Voice Assistant processing 100% of calls

In June 2021, the share of calls processed by the assistant reached 100%. This was reported on June 28, 2021 in Raiffeisen Bank. Thanks to automation, customers can quickly solve questions or switch to the right specialist, and communication remains as comfortable as when talking to a person.

Raiffeisen Bank launched voice assistant Ray on the hotline in September 2020 and has since gradually increased the share of automatically processed calls. During the scaling process, the parameters of the assistant were correlated in order to achieve the highest efficiency of its work.

File:Aquote1.png
"Ray's assistant is completely our development. We closely monitored feedback from users and relied on it in key updates. It is important for us that the assistant really facilitate the life of our customers without causing inconvenience, and in June the level of user satisfaction with communicating with the bank in remote channels reaches 90%, "commented Ilya Shchirov, head of the development of the Raiffeisen Bank chat and chat bots channel.
File:Aquote2.png

The assistant knows how to help users without involving a person - the share of automatically resolved requests is 30%. The ensemble of machine learning models is responsible for the high accuracy of understanding queries and finding the necessary answers. Throughout the development of the voice assistant, the team studied and introduced special mechanics that take into account the features of live speech: the ability to interrupt the script - for the user, the mechanics of refining the request - for the assistant, and others.

File:Aquote1.png
"Since the introduction of Ray, we have worked a lot with user experience. For example, the mechanics of interrupting the assistant was not very popular among users. But the mechanics of waiting are in demand: often users, in response to Ray's request to name some data, answer "Now" or "Wait." We taught the assistant to understand such phrases correctly, without perceiving them as a new request. These are non-obvious things that open only in the process of analyzing dialogues, but it is from such trifles that the quality of communication consists, "Ilya Shchirov explained.
File:Aquote2.png

In addition, integration with various bank systems allows the voice assistant to work with personalized queries. Answers to them come exclusively to the client's phone as push- or SMS-messages, which ensures the security of information transfer.

The assistant replaced the standard voice menu on the hotline. For ease of navigation, several branches of the voice menu have been saved, but the vast majority of questions are processed in the format of speech analysis. Also, at any time of the conversation, the client can switch to a conversation with a specialist.

File:Aquote1.png
"We do not stop - ahead of us is an increase in the topics with which the assistant works, the introduction of new speech processing mechanics and the launch of outgoing calls. We set ourselves an ambitious goal - to implement a voice bot, calls from which will not cause irritation, but will become concise and unobtrusive, "concluded Ilya Shchirov.
File:Aquote2.png

The voice assistant works at a basis own chats-boats-platforms of bank with use oblachnykhrechevy technologiyyandex SpeechKit which are responsible for transformation of the speech ringing the text, and the text of the answer of the assistant - in the sounding speech. The bank's bot platform, which also serves 11 bank chat channels: an online bank and applications for entrepreneurs and individuals, an official website, as well as social networks and instant messengers, is responsible for the logic of the operator, processing requests and searching for answers.

2020: Start Voice Assistant in Contact Center

On October 8, 2020, it became known that Raiffeisen Bank launched a voice assistant at the contact center. The voice assistant works at a basis of own chatbot-platform of bank with use oblachnykhrechevy technologiyyandex SpeechKit. Yandex SpeechKit is responsible for converting the caller's speech into text, and then the answer text into sound. The bank's bot platform is responsible for operator logic, processing requests and finding answers. The process is similar to the work of a bot in bank chat channels, the voice assistant uses the best practices in working with text appeals, as well as unique mechanics - for example, automatic sending of information in SMS or push notifications.

The pilot was held in September 2020 for 27 days. During the pilot, the voice assistant received incoming calls, processed questions and offered an answer, or switched the caller to a specialist. Switching to a specialist occurred in the following cases:

  • if the bot's confidence in the response was below the threshold;
  • if this was implied by the bank's internal policies (e.g. requests for bank products);
  • if the caller himself asked to connect with the operator.

File:Aquote1.png
We use automation technologies to achieve the best user experience in remote support channels. Solutions at the base of the voice assistant allow customers to solve many issues immediately at the time of treatment, communicating with the bot naturally, as with a person. At the same time, for the comfort of the user, we leave the opportunity to contact a specialist at any moment of conversation,

- stressed Ilya Shchirov, head of the development of the Raiffeisen Bank chat channel
File:Aquote2.png

At the moment, the assistant can work with more than 100 topics of appeals. He showed the greatest effectiveness in the topics "ATMs and branches" and "Exchange rates": the share of automatically processed requests in these areas was 53% and 55%, respectively.

According to the results of the pilot, a decision was made to scale the technology. It is planned that by the end of 2020, the voice assistant will participate in 50% of incoming calls, in 2021 he will be the first contact for all calls to the hotline. Thus, customers will be able to receive service support instantly even during peak load hours on the contact center.