X5 Retail Group is a leading Russian food retailer which operates the Pyaterochka, Perekrestok, and Karusel chains.
By integrating Yandex SpeechKit into their call center operations, the company successfully decongested their call center, with voice bots now handling up to 50% of customer requests.
Why get bots involved?
The X5 call center receives hundreds of calls every day, with the lion’s share of them quite similar, e.g. a customer wants to activate a loyalty card or complain about service quality. The large number of repetitive calls meant that call center operators spent a lot of time processing them, with more urgent and important issues ignored or delayed.
Automating inbound and outbound calls lightened the load on their call center employees, but they still had to deal with repetitive tasks on a daily basis. The solution was to add voice assistants who could communicate with customers autonomously, without human invovlement.
As a result, requests processed by call center representatives decreased significantly. Operators can now focus on more important issues needing human consideration, without wasting time on more trivial questions.
How voice assistants work
When a customer contacts the call center, they access the Interactive Voice Menu (IVR). The voice assistant names the most popular questions or lists possible problems, and the person presses the relevant button. IVRs like this are used by all sorts of companies today, from banks to mobile phone operators.
The customer then follows a predefined path, with the bot identifying the reason for the call and maintaining the dialogue. During the call, the bot learns the caller’s name, date of birth, phone number and loyalty card number, and enters this information into the corporate CRM system.
In order for the answers to users' questions to be accurate, the voice assistant should be able to clearly recognize their speech. This is made possible by preparing answers, standard phrases, and scenarios for various situations often brought up in customer questions in advance, e.g. how to transfer points between cards or join a loyalty program.
Developing the X5 voice assistant
Voice assistant bots work best when there are typical scenarios for customer interaction, and X5's large volume of similar customer calls was an ideal fit.
The X5 big data team had developed a speech engine including models to analyze text, understand customers' intentions, and navigate users through the dialogue. They still needed a service that could convert user speech into text. X5 tested several companies' offerings and selected SpeechKit from Yandex Cloud.
X5 experts developed scripts for the voice assistants and prepared more than 300 phrases. Whenever additional scripts for new tasks were needed, they were able to to develop them within a day’s time.
Scenarios were immediately tested on real-life incoming calls. The conversion rate was quite small (less than 10%), but it improved with each release: we tried different voices and edited the conversation scenarios. Using advanced settings in Yandex SpeechKit, they were able to shrink the pauses in between phrases in the customer-bot conversation, considerably improving the conversion rate.
The system can, in fact, understand human speech on-the-fly before it’s completed. The bot immediately detects the user’s question and answers it right away, rendering its speech as human-like as possible.
When the company gathered customer feedback about the voice assistant, most believed they had spoken to a human rather than a bot.
What can the voice assistant now do?
Process inbound calls from customers. It can register a loyalty card, transfer bonus points between cards, tell the current card balance, or block it at the customer’s request.
Make outbound calls. Call customers, run NPS surveys. When the system receives a call request, the robot calls the customer, asks them questions about the quality of products, service quality and agility, and so on.
Voice assistants now handle up to 50% of all loyalty requests in the Pyaterochka retail chain. Conversion to targeted actions is more than 60%, and voice assistants cost 5-7 times less than human operators.
Now our main focus is on scaling up cases with voice bots in the Pyaterochka and Perekrestok retail chains and creating a user-friendly app. We will continue developing our speech platform to add order confirmation and delivery notification, to conduct screening interviews for job candidates, to run surveys and research, to increase customer engagement with our mobile apps and bonus programs, and to consult users on a broader range of issues, etc. This technology can help us enrich our standard call functionalities and make calls more effective.