50% of questions of clients of "Home Credit Bank" are answered by bots

During a pandemic of covid-19 "Home Credit Bank", as well as many other financial institutions, faced the unprecedented growth of requests from clients. It was promoted also the general nervousness against the background of feeling of complete uncertainty, and permanent change of information landscape. Technologies helped to cope with a flow of requests to bank: "Home-bot" in a chat and a voys-bot (a voice bot) Maria – on a telephone line. Natalia Bibetko, the head of department of automation of service maintenance of "Home Credit Bank", in July, 2020 told how it was.

"Home-bot"

The first bot in customer service we started in December, 2018. "Home-bot" — so we called it – answered clients in a chat on the website of bank and in mobile application. Already on start he could consult the client on products of bank, tell about debts or a payment method of the credit.

The chat-bot works at the Webim platform. The platform plays a role only of a visual channel of communication with clients (interface). "Golova" of a bot – artificial intelligence technology in a chat (AI) – is full development of our bank.

During a koronakrizis main "blow" from client requests assumed the chat-bot (a photo - the press service of "Home Credit Bank")

As modern technological bank, we actively move in the direction of microservice architecture. And already constructed in connection therewith the most part of systems. For the bot AI engine all service "binding" was also executed in the concept of microservices.

The central place is taken by service which is the buffer between the solution of a chat and the AI engine. In it there are already a lot of functions. It conducts the difficult scenario focused dialogs, caching answers of clients, performs requests on client accounts and transactions, a shablonirovaniye of answers with data.

Most services communicate with each other with the help of REST API, but where it allows business logic, we also use the data bus for interaction.

Botha we is trained independently. Self-training has always the back. It as with books: what you read, you also are enriched with that. Therefore ourselves teach a bot to the correct expressions and phrases which need to be understood, distinguished, and which he learns to answer with a certain method.

Voys-bot Maria

In 2018 we decided to try to use automation in a voice on outgoing calls. Seeing success of this campaign, we decided to expand this technology and to start a full-fledged voys-bot. Cloud solutions did not suit us therefore we selected experienced international vendor - Omilia company which was engaged in development of a bot in a circuit of bank.

Maria - so we called a bot - earned at the end of December, 2019. The help in generation of a PIN code to cards of our bank was its first function. At the same time Maria is able to identify the client regardless of what phone calls he makes – or others, mobile or city. At that time it was the unique case for the Russian market.

The vendor provided us the platform on speech recognition (ASR) and configured the scenarios developed by us. Separately we built the platform of collecting and data exchange in bank: when the client requests the answer in dialog with a bot, even the second delay can matter.

Voys-bot "talks" the announcer's voice, her voice was selected at all-bank tender. We considered speech synthesis options, including from the largest Russian IT companies, but as a result stopped by a "live" voice – it sounds more humanly. The technology of use of synthesis at us is also implemented and if we need to implement quickly the new scenario with new remarks which long to write, we will be able to make it without problems.

Bots and koronakrizis

During a koronakrizis main "blow" from client requests assumed the chat-bot. By that moment it was already trained in several hundred scenarios and successfully closed on average 45% of customer appeals.

Sharp change of information landscape became the main problem. The country leaders in March-April read several statements fundamentally important for the financial industry at once. First, in the country announced non-working days. Certainly, clients right there asked a question: and whether banks will work? Then two more important short stories were read: start of credit vacation and introduction of taxation of income on deposits.

In parallel because of crisis some clients faced difficulties in income acquisition. Therefore the possibility of a delay and/or change of a payment amount on the credit interested many. Top and three concerned questions during a quarantine these subjects:

• What to do at difficulties with payment of the credit?
• May I use credit vacation?
• How to change a payment date?

Home Credit is large retail bank with the huge customer base. Therefore to lower load of call center specialists and not to force clients "hang" in queue to the specialist, we quickly in two days "taught" to advise fully the chat-bot clients on credit vacation. Also we daily monitored news and quickly updated answers taking into account the changing situation.

As a result in peak weeks of "Home-bot" closed more than 50% of all questions of clients. And call center agents could be engaged in work with more difficult cases at this time.

What's next?

In July of the 2020th, after a koronakrizisny pica, load of call center and bots returns by "pre-crisis" times. However, during this time we many times increased number of services which clients can make completely remotely, without visit of office. In particular, at the end of June we started a new feature at a voys-bot of Maria. Now it is able to perform operations on partial ahead of schedule or complete early repayment of the credits at clients and already completely closes 25% of such addresses.

As for the chat-bot, in the next plans – to continue to increase number of the scenarios processed by it and the closed addresses. At the same time we work not only on quantity, but also on quality of consultations of a bot. So, based on recently conducted internal research, 91% of clients do not address to call center of bank within a day after consultation with a bot. It means that the answer received from "Home-bota" suited them and the bot managed to resolve an issue.