How chat bots give reason for Sberbank's IT projects
Financial services are already actively used by the generation, accustomed to chatting in chat mode. The customer experience of this audience is in messengers, and business has to go after it.
The direction of chat bots began in SberTech thanks to the Sberbank-Companion domestic social project, then it was further developed in several other projects. About pilot projects - under the cut ...
')
We have to admit that there are already quite a lot of developments in the field of chat bots and natural language recognition. However, most of the open libraries and large speech transformation projects focus on English. Russian from this point of view has its own nuances. And although in Russia there are also interested teams, in particular, in large technical universities or private companies, such as Yandex, Mail.ru and SberTech, there are not so many ready-made industrial solutions. And the tasks for the Russian-speaking business are no less ambitious than the English-speaking ones.
Today we want to talk about the success we have achieved in the field of chat bots and machine learning.
It must be said that commercial solutions that are sufficiently customized for the needs of a particular company do not yet exist - the industry itself is still too young. There are projects from private researchers, for example, for technical support, but this is still the initial version of the technology, which allows categorization of requests. At the same time, no solution is ready for such a scale of work as Sberbank has - to analyze such large volumes of information with a huge number of clients. In the world there are only tools for deep analysis. And in fact, from these basic tools, our company's specialists develop their own customized products for Sberbank's tasks.
Currently there are four significant projects for the company. Further - more about each of them.
The whole history of chat bots in the company began with the Sberbank Companion project. On duty, Sberbank employees are sometimes forced to travel to remote offices, sometimes using several types of public transport. At the same time, their colleagues drive approximately in the same direction by car, with several empty seats. It would be great to combine these streams of employees, suggesting a company to travel with one and the opportunity to get to work faster for others.
At the same time, in our country, people do not like to drive each other, worrying about their safety. But the internal Customer Development has shown that colleagues in a company with a strong security service rely more on each other, even if they don’t know each other personally. So the idea arose to create a kind of internal community, through which people could find each other for joint trips to and from work.
A chat bot platform was developed as a community engine.
For about a year, several offices have been running a pilot project. The resulting service has proven itself so well that now we are discussing the expansion of its functionality. In particular, third-party carriers, taxi services, are ready to join the project. It is also planned to replicate the project to all departments.
Moreover, the project is preparing to go beyond the internal kitchen of Sberbank. The Sberbank community, of course, will not open out, but partner organizations (carriers) may well replicate this successful experience to other companies or create their own communities.
Another early project that involved the developed chatbot platform was the payment bot. It was launched into pilot operation at the beginning of this year (that is, it was born about half a year earlier than a similar solution presented in the spring of Google at the Google I / O 2017 conference).
In principle, the choice of some services with the help of bots, the conversation has been going on for quite some time - such systems have already been spread around the world. But the main feature was the ability to use free wording (rather than the prescribed set of commands), as well as payment through the bot.
The pilot project was implemented in conjunction with a food delivery restaurant. Through the chat, you could choose food, while the bot suggested the best options for the user's request, remembered the tastes and the consumer basket and could prompt the dishes in the next order.
In fact, the pilot bot served as a food delivery service application. However, there are many services — there are many applications (for each provider you have to download and install your own). A chat bot is just one of the contacts in the phone, in fact it’s just a new channel for receiving the same service.
The chat bot for the corporate site was the first such project that went beyond the sandbox to the wide world. Here, of course, practices from the Companion and the payment bot are used, but more “adult” and Machine Learning algorithms for analyzing queries and finding suitable answers are implemented.
The task of the chat bot is to communicate with customers via chat on the site, answering questions about the company's products, as well as the accounts, credits and payment orders of the client (in the pilot version, the range of issues is limited to product data and opening an account). To do this, the chat bot analyzes the information received from the client, gives answers to simple questions formulated in natural language, or requests additional data if the initial request lacks them.
The bot is built on the technology of latent-semantic analysis, which provides the search for the answer to the client's question, taking into account the context. The technique allows for keywords and the context of the dialogue to find in the database of ready-made answers, which include a variety of product FAQs, the most relevant answer.
In detail, it looks like this. First, a client request in natural language is turned into a set of keywords. Each word in the model has a certain weight. Common words in most queries have less weight, and words unique to a given query have more. Thus, the request is transformed into a set of keywords with a certain weight, according to which the database is searched: a special algorithm is used to determine the probability that a document from the database of answers corresponds to the question posed. The client is given the most relevant response, and in his absence, the chat switches to the operator.
To evaluate the solution, feedback is collected from clients on the work of the chat bot - whether they liked the answer that the algorithm picked up. Together with this assessment, each chat allows you to specify the existing model and response base. If the chat switches to the operator, the answer chosen by the employee also goes to the bot's knowledge base.
The pilot project will be launched soon. While the “pilot” communication is accompanied by the operator, controlling the correctness of the answers selected by the bot. When a certain level of satisfaction is reached for customers who receive such automatic responses (and, consequently, quite high marks from the staff accompanying the pilot), a decision will be made to enter the chat bot into full operation. In the future, it is planned, on the one hand, to expand the range of issues with which the bot is engaged, and on the other, to completely replace the part of the operators' communication with the bot, at least on typical issues.
In a sense, this pilot project can be called the next evolutionary step of the entire chatbots platform. If the previous solutions somehow processed the requests, categorized them and searched for them to match the existing knowledge base, then the final task of the client request processing system is to draw its own conclusions, which may well affect Sberbank’s management strategy.
There are always some informational background around public companies, such as Sberbank. This can be either direct contact with the company via the feedback form or telephone, or evaluation of services or products, publicly expressed on the Internet - on the personal pages of clients on social networks, on sites with reviews. Moreover, opinions can contain both positive and negative assessment of the Bank's work. It is assumed that the system will process all calls coming through the available channels, classify them (separate complaints or, on the contrary, thanks, determine topics) and record trends at a higher level of abstraction — identify problems that most concern customers, identify products and services requiring close attention, prioritize. Since the development of services and products in Sberbank follows agile methodology, ideally we hope that the system will become an additional source of ideas backing from the outside in case the team developing this or that product has a look.
In a sense, it is a system for analyzing consumer opinion. But it not only collects data and conducts high-level cataloging by the same method as the chat bot, but also reveals the essence. As an additional “bonus”, we can use it to create a client profile that is interested in various services — we obtain a detailed cataloging of which segment of the audience and which products are interested. In addition, the data can be viewed in different sections. For example, analyzing all the channels in the complex, you can identify complaints about different products, the essence of which will be reduced to the fact that in a certain department they simply do not serve customers well. Those. An important factor may be the office, and not the parameters of any particular service.
The above are examples of analyzing the collected data arrays based on the assumption that we know what we are looking for. The important point is that the system learns itself. We build a model based on the customer’s customer history, complementing it with new data as it arrives, i.e. to a certain extent, every new conversion affects this model. As a result, the service allows you to solve a huge reservoir of problems associated with the allocation of topics about which we initially do not know anything. For example, the client may raise a topic that was not previously on the list of interest. The system allows you to add this topic to the monitoring list, so that future calls can be checked for compliance with the new subject as well.
The main consumer for this analysis of appeals is bank management, i.e. It can not be evaluated "outside." The system will allow management to keep closer to the real state of affairs: to better understand the general situation, the attitude of clients towards the bank and towards its services.
So far, like other projects, the system exists in a pilot version. At the moment, it is focused on the analysis of requests received through the feedback form. Now these are text messages, but in the future the voice will also be processed, since the main channel of communication with customers is voice calls, which will be translated into text and processed by the same tools.
An important aspect is the evaluation of the pilot's work. It will be performed by comparing the automatic cataloging of a certain reference to the standard (our knowledge of which topic the reference is). In the course of the development of the model, this “distance” from the standard will be minimized and minimized. It is assumed and expert assessment by bank employees.
In general, chat bots allow you to translate a lot of ideas that affect the topic of interaction with the client in a natural language. As the inherent algorithms develop, we have the opportunity to refine the previous projects. For example, if a business appears to be interested in a partnership within a payment bot, we can supplement it with more modern algorithms designed for a chat bot on the site and a system for analyzing customer requests. At the same time, we are sure that all this is only the first step in the direction of large-scale and interesting projects based on machine learning.
The material was prepared by employees of the Central Committee of SberTech billing technologies:
nill2 , Danil Kabanov, director of the Central Committee
aspera , Ruslan Halimov, leading engineer
Stanislav Kim, Development Manager
The direction of chat bots began in SberTech thanks to the Sberbank-Companion domestic social project, then it was further developed in several other projects. About pilot projects - under the cut ...
')
We have to admit that there are already quite a lot of developments in the field of chat bots and natural language recognition. However, most of the open libraries and large speech transformation projects focus on English. Russian from this point of view has its own nuances. And although in Russia there are also interested teams, in particular, in large technical universities or private companies, such as Yandex, Mail.ru and SberTech, there are not so many ready-made industrial solutions. And the tasks for the Russian-speaking business are no less ambitious than the English-speaking ones.
Today we want to talk about the success we have achieved in the field of chat bots and machine learning.
It must be said that commercial solutions that are sufficiently customized for the needs of a particular company do not yet exist - the industry itself is still too young. There are projects from private researchers, for example, for technical support, but this is still the initial version of the technology, which allows categorization of requests. At the same time, no solution is ready for such a scale of work as Sberbank has - to analyze such large volumes of information with a huge number of clients. In the world there are only tools for deep analysis. And in fact, from these basic tools, our company's specialists develop their own customized products for Sberbank's tasks.
Currently there are four significant projects for the company. Further - more about each of them.
Sberbank Hitcher
The whole history of chat bots in the company began with the Sberbank Companion project. On duty, Sberbank employees are sometimes forced to travel to remote offices, sometimes using several types of public transport. At the same time, their colleagues drive approximately in the same direction by car, with several empty seats. It would be great to combine these streams of employees, suggesting a company to travel with one and the opportunity to get to work faster for others.
At the same time, in our country, people do not like to drive each other, worrying about their safety. But the internal Customer Development has shown that colleagues in a company with a strong security service rely more on each other, even if they don’t know each other personally. So the idea arose to create a kind of internal community, through which people could find each other for joint trips to and from work.
A chat bot platform was developed as a community engine.
For about a year, several offices have been running a pilot project. The resulting service has proven itself so well that now we are discussing the expansion of its functionality. In particular, third-party carriers, taxi services, are ready to join the project. It is also planned to replicate the project to all departments.
Moreover, the project is preparing to go beyond the internal kitchen of Sberbank. The Sberbank community, of course, will not open out, but partner organizations (carriers) may well replicate this successful experience to other companies or create their own communities.
Payment bot
Another early project that involved the developed chatbot platform was the payment bot. It was launched into pilot operation at the beginning of this year (that is, it was born about half a year earlier than a similar solution presented in the spring of Google at the Google I / O 2017 conference).
In principle, the choice of some services with the help of bots, the conversation has been going on for quite some time - such systems have already been spread around the world. But the main feature was the ability to use free wording (rather than the prescribed set of commands), as well as payment through the bot.
The pilot project was implemented in conjunction with a food delivery restaurant. Through the chat, you could choose food, while the bot suggested the best options for the user's request, remembered the tastes and the consumer basket and could prompt the dishes in the next order.
In fact, the pilot bot served as a food delivery service application. However, there are many services — there are many applications (for each provider you have to download and install your own). A chat bot is just one of the contacts in the phone, in fact it’s just a new channel for receiving the same service.
Chat bot contact center
The chat bot for the corporate site was the first such project that went beyond the sandbox to the wide world. Here, of course, practices from the Companion and the payment bot are used, but more “adult” and Machine Learning algorithms for analyzing queries and finding suitable answers are implemented.
The task of the chat bot is to communicate with customers via chat on the site, answering questions about the company's products, as well as the accounts, credits and payment orders of the client (in the pilot version, the range of issues is limited to product data and opening an account). To do this, the chat bot analyzes the information received from the client, gives answers to simple questions formulated in natural language, or requests additional data if the initial request lacks them.
The bot is built on the technology of latent-semantic analysis, which provides the search for the answer to the client's question, taking into account the context. The technique allows for keywords and the context of the dialogue to find in the database of ready-made answers, which include a variety of product FAQs, the most relevant answer.
In detail, it looks like this. First, a client request in natural language is turned into a set of keywords. Each word in the model has a certain weight. Common words in most queries have less weight, and words unique to a given query have more. Thus, the request is transformed into a set of keywords with a certain weight, according to which the database is searched: a special algorithm is used to determine the probability that a document from the database of answers corresponds to the question posed. The client is given the most relevant response, and in his absence, the chat switches to the operator.
To evaluate the solution, feedback is collected from clients on the work of the chat bot - whether they liked the answer that the algorithm picked up. Together with this assessment, each chat allows you to specify the existing model and response base. If the chat switches to the operator, the answer chosen by the employee also goes to the bot's knowledge base.
The pilot project will be launched soon. While the “pilot” communication is accompanied by the operator, controlling the correctness of the answers selected by the bot. When a certain level of satisfaction is reached for customers who receive such automatic responses (and, consequently, quite high marks from the staff accompanying the pilot), a decision will be made to enter the chat bot into full operation. In the future, it is planned, on the one hand, to expand the range of issues with which the bot is engaged, and on the other, to completely replace the part of the operators' communication with the bot, at least on typical issues.
Customer Request Processing System
In a sense, this pilot project can be called the next evolutionary step of the entire chatbots platform. If the previous solutions somehow processed the requests, categorized them and searched for them to match the existing knowledge base, then the final task of the client request processing system is to draw its own conclusions, which may well affect Sberbank’s management strategy.
There are always some informational background around public companies, such as Sberbank. This can be either direct contact with the company via the feedback form or telephone, or evaluation of services or products, publicly expressed on the Internet - on the personal pages of clients on social networks, on sites with reviews. Moreover, opinions can contain both positive and negative assessment of the Bank's work. It is assumed that the system will process all calls coming through the available channels, classify them (separate complaints or, on the contrary, thanks, determine topics) and record trends at a higher level of abstraction — identify problems that most concern customers, identify products and services requiring close attention, prioritize. Since the development of services and products in Sberbank follows agile methodology, ideally we hope that the system will become an additional source of ideas backing from the outside in case the team developing this or that product has a look.
In a sense, it is a system for analyzing consumer opinion. But it not only collects data and conducts high-level cataloging by the same method as the chat bot, but also reveals the essence. As an additional “bonus”, we can use it to create a client profile that is interested in various services — we obtain a detailed cataloging of which segment of the audience and which products are interested. In addition, the data can be viewed in different sections. For example, analyzing all the channels in the complex, you can identify complaints about different products, the essence of which will be reduced to the fact that in a certain department they simply do not serve customers well. Those. An important factor may be the office, and not the parameters of any particular service.
The above are examples of analyzing the collected data arrays based on the assumption that we know what we are looking for. The important point is that the system learns itself. We build a model based on the customer’s customer history, complementing it with new data as it arrives, i.e. to a certain extent, every new conversion affects this model. As a result, the service allows you to solve a huge reservoir of problems associated with the allocation of topics about which we initially do not know anything. For example, the client may raise a topic that was not previously on the list of interest. The system allows you to add this topic to the monitoring list, so that future calls can be checked for compliance with the new subject as well.
The main consumer for this analysis of appeals is bank management, i.e. It can not be evaluated "outside." The system will allow management to keep closer to the real state of affairs: to better understand the general situation, the attitude of clients towards the bank and towards its services.
So far, like other projects, the system exists in a pilot version. At the moment, it is focused on the analysis of requests received through the feedback form. Now these are text messages, but in the future the voice will also be processed, since the main channel of communication with customers is voice calls, which will be translated into text and processed by the same tools.
An important aspect is the evaluation of the pilot's work. It will be performed by comparing the automatic cataloging of a certain reference to the standard (our knowledge of which topic the reference is). In the course of the development of the model, this “distance” from the standard will be minimized and minimized. It is assumed and expert assessment by bank employees.
In general, chat bots allow you to translate a lot of ideas that affect the topic of interaction with the client in a natural language. As the inherent algorithms develop, we have the opportunity to refine the previous projects. For example, if a business appears to be interested in a partnership within a payment bot, we can supplement it with more modern algorithms designed for a chat bot on the site and a system for analyzing customer requests. At the same time, we are sure that all this is only the first step in the direction of large-scale and interesting projects based on machine learning.
The material was prepared by employees of the Central Committee of SberTech billing technologies:
nill2 , Danil Kabanov, director of the Central Committee
aspera , Ruslan Halimov, leading engineer
Stanislav Kim, Development Manager
Source: https://habr.com/ru/post/337496/
All Articles