Tinkoff introduces Оleg, the world’s first voice assistant created in Russia for finance and lifestyle-related tasks. His mission is to help users navigate within the Tinkoff ecosystem, make customers’ lives simpler and become a friend who is always nearby. Oleg is the first voice assistant built by a financial institution as a proprietary solution.
Who is Oleg
Oleg is man between 25 and 40 years of age. He is polite, but also firm when needed. He is funny, but he won’t teach anyone how to live life, unless he’s specifically asked. Oleg has an inkling that he’s not human, but he’s not overly concerned about that. He knows what’s important, which is that a day spent without helping Tinkoff customers is a day wasted.
What Oleg can do
Oleg lives in the Tinkoff mobile app. The voice assistant is able to recognise and interpret different user commands, ask follow-up questions, solve problems and speak on a variety of topics. Oleg will help customers use their voices to manage all the key financial and lifestyle services in the Tinkoff ecosystem.
In its first version, Oleg will be able to:
In the future, Oleg will become even more capable and integrated into all the elements of the Tinkoff ecosystem, including travel, mobile, investments, insurance, entertainment, business and Tinkoff Junior.
Oleg will also be able to identify a user’s voice using biometric data and carry out commands that currently require authorisation within the Tinkoff app. This will enable users to carry out such tasks as transferring money while driving a car.
How to talk to Oleg
To start a conversation in the mobile app, one has to say “Hi Oleg” or “Listen Oleg” in Russian. You can ask Oleg a question, like “What’s on tonight” or ask him to perform a task, such as transfer funds.
Oleg will be able to converse about various things, including himself. For example, he can answer such questions as “What's your name?”, “How are you?”, “What music do you listen to?” “What’s new at the bank?” or “Is Oleg cute?”
You can ask Oleg for advice and he will give his recommendations about money or lifestyle, based on information from the Тinkoff Journal. You can also ask Oleg to “Show me the offers on running shoes” or play a game with him. He can test your trivia knowledge by asking what products one can’t buy in Australia or what could Lev Tostoy afford to buy with royalties for his Anna Karenina masterpiece.
Oleg is a very young voice assistant, so he has a lot to learn. Users can give ‘likes’ and ‘dislikes’ to Oleg’s responses, which will be analysed by the Tinkoff team. Customers can also say “Enough” or “Speak with a representative” if they feel Oleg isn’t resolving their problem successfully.
Oleg is set to solve the majority of customer issues automatically. He will also be able to use some help from call centre staff to resolve the issues he can’t solve alone.
As Oleg spends more time interacting with customers, he will learn to change his tone from friendly to formal in real time, as appropriate for the nature of the conversation and the customer’s mood.
Oleg and voice technology at Tinkoff
Oleg is training and learning using Tinkoff’s supercomputer (the Kolmogorov cluster) capacities, one of Russia’s most powerful supercomputers. With the help of the voice assistant the Tinkoff team will be able to increase the speed with which it trains the neural networks for speech recognition and synthesis, natural language processing and open-ended conversation.
Deep neural network models and voice technologies, which Tinkoff has been working with since 2014 within its AI First strategy, were used to create Oleg. In 2016, Tinkoff has begun work on its own voice recognition technology. This technology works equally well with noisy speech received via phone channels and clear speech from high-quality sources. Currently, this voice recognition technology allows Tinkoff to correctly identify up to 95% of spoken words and uses terabytes of data and tens of thousands of hours of human speech to continue to train and improve this technology. In 2017, Tinkoff launched its own biometric system, which makes the right decision in 99.99% of cases. In 2018, Tinkoff became the first vendor bank of the Unified biometric system in voice biometrics. That same year Tinkoff began developing its own voice synthesis technology, based on neural models such as WaveNet, Tacotron and Deep Voice. This allowed synthesised speech to come as close as possible to human speech.
Voice technology isn’t just for the voice assistant. It also enhances Tinkoff automate customer service processes. Voice recognition technology helps field around one million calls, while a biometric system, trained on customer voice data, helps the call centre combat fraud.