On June 30, an online conference Finbot 2020 was held – a case conference on chatbots, robots in voice channels and virtual assistants for banks, at which Igor Luzhansky (Co-founder of Chatbots.Studio), together with Vladimir Yastrebkov (Head of the Directorate for Product and Technology Development at VTB Group) talked about the joint case of creating a chatbot on Facebook Messenger.
In addition to the presentation of the Chatbots.Studio case, the conference featured other significant in our opinion case sessions about chatbots and virtual assistants from representatives of Sberbank, VTB Bank Group, Alfa Bank, Accenture, Raiffeisenbank, Rosbank, Tinkoff and other banks.
In this article, we would like to touch upon the particularly interesting achievements made at the conference.
- 1 VTB: how to deal with the collapse of quality assessment during a pandemic
- 2 Sberbank: launching a chatbot for legal entities on the market
- 3 Yandex Money: a brief history and results of a chatbot for a payment service
- 4 VTB Azerbaijan: neobank in messenger, Chatbots.Studio case
- 5 Alfa-bank: chatbot performance evaluation metrics
- 6 Omilia: a human-like user experience through voice assistants
- 7 Raiffeisenbank: Dynamics of growth in popularity of calls to chat channels in 2017-2019
- 8 Summing up
VTB: how to deal with the collapse of quality assessment during a pandemic
In December 2018, the ability to access the functions of the client-bank in the chatbot became available to the bank’s clients. It took the team about 5 months to put the bot into mass production (Dec 2018 – April 2019).
Alexander Raihin, Head of the Department at VTB Bank, shared his story of solving the problems associated with the pandemic.
During the pandemic, the volume of traffic to both the VTB Bank chatbot and the contact center increased sharply. It was a difficult period and we had to act in a stop-fire mode.
Problems encountered during the pandemic:
- Due to the growth of customers on the hotline, all chat operators were transferred to the contact center, leaving only the bot in the chatbot
- Frequently asked topics have changed and changed within a short time
- The chatbot stopped responding to current user questions and the bot’s performance rating decreased
- I had to forcibly reduce the accuracy of the bot’s response from 40% to 20%
To assess the quality of a chatbot in a team, the CSI metric is used, measured on a 5-point scale. The metric was introduced in December 2018, after the creation of a pilot version of the chatbot.
- We expanded customer service channels and launched bots in the Internet bank and Viber
- In a short time, we launched a bot to receive applications for credit vacations
- Analysis of popular topics focus on 5-10 to-topics and daily releases of new bot scenarios, or even 2-3 times a day
- Increase in the number of repeated requests
- Focus on negative customer feedback and troubleshooting
- Implemented the function of notification about unread responses for the client in the form of push notifications
Thus, eliminating problems in the chatbot, the team managed to cope with user requests for the period from 16.03 to 8.06, stop the drop in the bot’s performance rating and even reach a higher indicator than it was before the pandemic – from 4.42 to 03.16-29.03 to reach the rating 4.52 to 8.06. At the moment, the bank has a chatbot for WhatsApp and an omnichannel bot with a voice assistant in development.
Sberbank: launching a chatbot for legal entities on the market
Sberbank was represented at the conference by Sofya Makushova, the Owner of the Chatbot for YL product.
Among legal entities, Sberbank provides services for 2.4 million clients, which are served by nearly 4,000 employees. The bank launched a chatbot for legal entities in 2019, and by the end of the year, online chat served 9% of applications among other channels of communication with a client. Already at that time, the company noted a 2-fold decrease in the cost of customer service relative to calls with an operator. To process FAQ and dialog scripts, the bank uses an internal NLP platform that serves chats for retail and corporate clients.
As of March 2020, the bank through chat service channels (in chats on the website, WhatsApp, Web SBBOL and SBBOL mobile application) closes 172,000 calls per month, which are serviced. In addition, in 2020, online chat is already serving 17% of customer requests with a CSI of 85% (approx. CSI – customer satisfaction index).
The ability to glue multiple user messages into one intent was introduced, without running scripts for each message separately – which significantly increased FCR.
FCR (First-Call Resolution) – a metric of the Customer Support departments, measured in the number of resolved customer issues on one line within 1 hour. International standard – at least 80%.
Now the bot is testing the functions of proactive communication with the client, which allows not only serving current applications, but also offering banking products. And also, data is being collected to target this function if possible
Yandex Money: a brief history and results of a chatbot for a payment service
Among popular services for electronic payments in Russia, Yandex.Money comes in second place. At the conference, Dmitry Ofitserov, Head of the Yandex.Money client service development department, shared his success story and information on the volume of the chatbot’s work.
At the time of the search for the optimal chat solution for the payment service, 20% of all requests were already served in the chat of the mobile application and in the chat on the Yandex.Money website. Previously, the company already had a successful case of automating Yandex.Mail processes using chat solutions. With the introduction of the virtual assistant, it was planned to automate the processing of a third of calls, without operator intervention. The chat assistant was originally implemented only in the mobile application and under the name Moneybot.
Moneybot serves clients’ requests for opening wallets, ways to replenish them, limits, commissions, cashback, and also advises on the issuance of virtual cards, working with payments and transfers, contactless payment methods and withdrawal options.
After 10 months of operation of the Moneybot, the volume of requests processed in the chatbot was increased to 41%, and the accuracy of the answer was recorded by 85% within 220 topics and 272 scenarios. In the future, the chatbot plans to cover up to 60% of requests and introduce personalization through integration with internal systems. In addition, Dmitry shared information that the chatbot has already paid off and only two people are engaged in its maintenance.
VTB Azerbaijan: neobank in messenger, Chatbots.Studio case
VTB was represented at the conference by Vladimir Yastrebkov. After 10 years of operation of VTB Bank Azerbaijan and reaching a retail client base of 10,000 clients, in 2019 the revision of strategic priorities was towards the digitalization of the bank’s processes and a non-trivial task was set: to figure out how to quickly increase the volume of business without any serious investments from side of the parent company.
The choice fell on the chatbot in Facebook Messenger as an inexpensive solution compared to a mobile application and a quick solution to market.
How the bot works: most often a new client learns from an advertisement received directly in the messenger, or from a message from friends who already use banking services in the chatbot. When switching to a contact, the client can get himself a virtual prepaid DIRECT card.
So far, DIRECT serves only Azerbaijani customers, but the technology has the potential to be applied in other divisions of VTB Group. Technically, the chat solution allows you to deploy another digital bank in the messenger in just 2 months. At the moment, the product is used by several thousand customers, and in the feedback, customers note the speed and convenience of working in the messenger.
You can find out more about it on the case page of the chatbot for VTB or in the bank’s article on habr.
Alfa-bank: chatbot performance evaluation metrics
Alfa-Bank was represented by Nikita Komarov, Product Analyst, and Natalya Balyberdina, Head of Digital Online Channels Development.
The percentage of full automation of service in the chat of Alfa-Bank has grown from 8% to 16% since April 2019. The chatbot’s penetration into dialogs increased from 18% to 26%, and the chatbot analytics pipeline was also tuned.
Nikita Komarov talked what groups of metrics to measure the work of chat solutions, in the context of three levels – model, scenario and user.
The model can be measured by 3 parameters:
- Distribution of confidence in answers
- Confusion matrix
Among scenario metrics
- Successful closure of sessions
- Number of repetitions of intent
- Number of operator calls
- Negativeness and user ratings
User metrics measure accuracy, automation, or any other time-based metric.
Omilia: a human-like user experience through voice assistants
Platon Begun, Director of Business Development in Russia at Omilia, presented on a live demo the ability to recognize user intents in a dialogue with a voice assistant. Due to the flexible context memory of the platform, the technology handles unstructured user cues, ambiguous cues and mentions in the context of the current conversation.
Among the metrics of the accuracy of the company’s voice solution:
- 21 languages for deep speech recognition
- Serving 15 countries
- Semantic accuracy 96%
- 90% of dialogs completed successfully
The speaker also noted that for the introduction of the company’s voice technology into the omnichannel customer service, it is most convenient to implement through the client’s chat platform. Among the cases are Piraeus Bank, Alpha Bank, Eurobank and Alfa Bank.
Raiffeisenbank: Dynamics of growth in popularity of calls to chat channels in 2017-2019
The chat channel was launched at the end of 2017 and then it served about 2% of requests. By December, the share of calls to chat was 18%.
In 2018, to clarify the relevance of the bot’s intents, the chat solution team conducted several customer surveys on a sample of 1000+ respondents. From the survey, it turned out that the most popular user requests – questions about ATMs and branches, online banking, followed by questions about debit cards and loans – were the focus for optimizing chatbots.
In 2019, Raiffeisenbank’s clients began to visit branches 30% less frequently than a year earlier, preferring to resolve their issues through technological communication channels. For example, 80% of active customers use a mobile application, and the number of transactions in the Internet and mobile banking increased by 60%.
“In 2019, the number of client calls through chats has more than quadrupled – from 38 thousand to 165 thousand, and the trend for the migration of clients in them has significantly increased in 2019. A year ago, only every 12th call to the contact center came via chat, now every third one. “
And already at the end of 2019, Raiffeisenbank issued a press release, in which it was reported that the bank was launching a virtual assistant based on machine learning (ML) through the channels: WhatsApp, Telegram, Viber, in the Internet and mobile banking, on the bank’s website for individuals and small and micro-businesses. According to Ilya Shchirov: “The popularity of remote channels of communication between clients and the bank is growing, and today we already receive more than 35% of calls through chat, we expect that this share will already be 50% by the end of 2021. The virtual assistant, according to our calculations, will increase the efficiency of non-voice communication channels by 40% by the end of 2021″
Many interesting cases of banks were presented at the conference. The speakers showed what bots have already created value in practice: reducing the cost of servicing requests, reducing the time it takes to process customer requests and improving customer service.
In addition, the direction of chatbots for banks has prospects in omnichannel solutions in messengers that provide a seamless user experience.
Also, the experience of banks shows that robotization or bottization of a bank’s contact center is a non-trivial task. Despite the promises of vendors to implement simple and fast solutions, the integration of chat solutions into the bank’s infrastructure requires attention, effort and project management from the bank itself.
We managed to divide the use of chat solutions by banks into two groups:
- customer service
- digitalization of banking products in messengers
In the first group, customer support to reduce the cost of servicing frequently asked questions by agents and to shift the focus of agents to more complex tasks. The second group includes chatbots as a marketing tool for business development and entering new markets.
Among the innovations in the field of omnichannel interaction with the bank’s clients are voice assistants in instant messengers. At the moment, voice solutions have been implemented at Tinkoff Bank, Home Credit Bank and Alfa Bank. In addition, the Yandex.Dialog website contains information about the virtual assistant from Sberbank, integrated into the voice assistant Alice, but judging by the average user rating (2.7 stars based on 117 ratings), the solution requires improvement.
On behalf of Chatbots.Studio, we would like to express our gratitude to the organizers of the Finbot 2020 Conference Conglomerate for the quality content and the opportunity to exchange experience with the leading banks in Russia. At the conference, we once again saw that the market is becoming more mature and solutions for banks are becoming more advanced.
The Chatbots.Studio team develops bot platforms for banks in instant messengers – Apple Business Chat, WhatsApp, Telegram, FB Messenger and Viber. Among the clients for whom we have already implemented chatbots are PrivatBank, VTB Azerbaijan, CenterCredit, Sberbank Kazakhstan and Concord Bank.
For any questions about building RBS in messengers, developing chatbots in WhatsApp, Telegram, FB Messenger and Viber – fill out the contact form or write to us at firstname.lastname@example.org.