How Yandex SpeechKit helped Raiffeisenbank create a voice assistant

Raiffeisenbank developed its voice assistant to improve customer service and reduce wait times. Now the assistant answers over 80% of incoming calls and resolves customers' issues independently in 35% of cases.

The Raiffeisen voice assistant uses Yandex SpeechKit for accurate speech recognition and natural speech synthesis — in other words, it serves as the ears and voice of the bot. Bringing in a voice assistant reduced the call center’s workload, allowing operators to devote their time to handling more complex questions. As a result, customers were able to get better service faster, while the bank reduced the cost of processing such calls by 65%.

Raiffeisenbank first developed its bot platform, including chatbots in its app, site, and messengers, that processed over 50% of customer queries. But some customers preferred to ask questions verbally rather than chatting with a bot via text.

As the bank’s customer base grew, the call center’s ability to respond quickly deteriorated, and customers had to wait several minutes to reach an operator. In most cases, however, the customers' questions were standard, e.g. asking for the location of the nearest branch or the current dollar exchange rate.

The bank decided to launch its voice assistant to relieve the call center, redirect human operators' efforts to more complex tasks, and speed up call response times.

Who will teach the bot to speak?Who will teach the bot to speak?

Flexibility was a top priority as Raiffeisen developed its voice assistant. The bank wanted it to be able to respond to different scenarios and interact with users in a variety of ways. The team also wanted to be able to test new hypotheses and implement changes quickly.

This was impossible to achieve with pre-recorded phonograms, since each new action would require a new recording — a long and laborious process. The team also rejected the idea of combining synthesized and pre-recorded speech, as the resulting speech would not sound sufficiently natural, and the conversation scenarios would require additional preparation.

After weighing the options, they chose Yandex SpeechKit, a cloud service for speech synthesis and recognition. This decision helped Raiffeisenbank save money by paying only for the cloud resources they used, not on infrastructure and development. It also made it possible to launch and develop the flexible voice assistant quickly — the bank’s team launched their pilot project in just three months, including planning and development.

Another benefit is that Yandex Cloud (which Yandex SpeechKit is part of) ensures total data security. The platform complies with Russian data protection laws, GDPR, ISO, and PSI DSS industry standards.

Raiffeisenbank’s voice assistant was ready in just three months. They trained it to respond to over 100 topics, including pricing, ATM locations, branch hours, and currency exchange rates. The voice assistant automatically processes requests related to these topics, recognizes a question, and answers it immediately. In most cases, customers are able to receive an answer to their question without speaking to a human operator.

In the pilot project, however, the assistant could only respond to standard requests, since additional integration with various bank systems was required in order to answer personalized questions, e.g. requesting an account balance. Specialists at Raiffeisenbank have since trained the assistant to identify and respond to more personalized requests. The customer receives responses via push notifications or SMS messages.

Today, a voice assistant receives all incoming calls to the bank’s hotline. It processes about 155,000 calls a month and completes 54,000 of them automatically, without involving a human operator.

Raiffeisenbank’s specialists continue to train the bot on new topics and integrating it with new systems so that the assistant can quickly respond to personalized customer requests. One example is a new script allowing customers to interrupt, meaning they don’t need to listen to the bot’s entire message — they can interrupt it by speaking up.

Automation technologies help the bank respond more flexibly to increasing customer requests and process them more quickly and efficiently, without increasing employee workloads.