Chatbots in Russia: market features, notable projects, near-term prospects. TAdviser Overview

In the early 2010s, the first text robots joined the sites of mobile operators and banks, as well as popular instant messengers. Then, as speech recognition and synthesis technologies developed, virtual operators appeared in contact centers.

Virtual assistants to the largest Russian banks, − notes Anna Vlasova, head of the linguistics department at Nanosemantics, are already intellectual bots.

Companies can use chatbots to solve several problems: increase customer engagement; increased volumes of lidogeneration; Optimize customer service costs flexible expansion of the customer service during busy hours, simplification of customer customization processes.

Depending on the tasks, such automated dialog systems are conditionally divided into chat bots of technical support services, chat bots for marketing and customer engagement needs, chat bots for banks and insurance companies, medical chat bots, as well as chat bots for communications in instant messengers.

Modern chat bots can perform much more tasks than just answering typical customer questions. They can create FAQs for customers or employees, notify customers of promotions and special offers, or record them for a variety of services. The range of tasks solved by the bot is almost unlimited, and it depends only on the goals of the business, - says Denis Afanasyev, director of the B2B products division of Salut SberDevices.

For businesses, especially those related to customer service from the mass segment, automated online consultants on the company's website, in its contact center, on pages in social networks or in popular instant messengers, are very profitable. Depending on the development company, they can remove from 40% to 90% of the load from employees advising clients in "digital channels," and at the same time work for free in 24 by 7 mode without breaks for lunch and sleep, do not take sick leave and do not need vacation.

Already in the mid-10s, the advantages of robotic agents were appreciated by medium and. small business The launch of chatbots was then announced one after another - Internet services, online stores retail networks, taxi services.

Over time, companies that stood at the origins of the chatbot market and invested a lot of financial and intellectual resources in its development, such as the CST (formed in 1993) and Nanosemantic (created in 2005), evolved from written speech recognition and synthesis technologies to the development of systems based on artificial intelligence and machine learning, learned to accurately understand, process and analyze natural language speech (NLP).

We look at this issue more broadly. After all, a chat bot and components of artificial intelligence are only part of the technological components of robotization (as the ability to read, write and think...), − says Pavel Borchenko, CEO of Robin, the developer of the first Russian RPA platform. − Adding to them RPA, process control, processing voice, documents, etc. (as an opportunity to act, watch, talk...) - We get full-fledged digital employees who can execute complex complex processes and bring maximum business benefits.

At the same time, digital agents serving customers have yet to win their loyalty. As a study by analysts of the markswebb company of dialog interfaces implemented in banking applications showed, customers are ready to interact with chat bots on simple issues, and complex ones prefer to find out in communication with a live operator.

To begin with, it is worth analyzing the available data and conducting research, what benefits it will bring, what processes it automates, where it will help to save and improve service metrics, − the expert recommends and advises not to try to automate all processes at the same time. - It is better to choose for the beginning frequency and the least labor-intensive processes in implementation, − the expert concludes.

First of all, it should be convenient to use a chat bot, there should be no dead ends in dialogues, and the user should always solve the issue for which he turned to the bot, − he notes.

In addition, the developer of dialog interfaces needs to have a full set of technologies and tools. Such, in his opinion, are speech recognition and synthesis (TTS, text-to-speech) and (Speech-to-Text), a dialog engine, natural language processing tools - NLP. At the same time, according to the expert, it is important that all of them are of high quality.

You cannot launch a robot, get the primary result and hope that everything will remain at the same level over time, − he stated.

In order to quickly adjust the behavior of the robot without resorting to long-term training, as well as to ensure the transparency of its logic, the expert recommends combining both the rule-based approach and the AI-based approach when developing it.

Customers today expect a personal approach and a pleasant tone of voice, − notes Sergei Popov, − which we, as a vendor, can achieve with deep integration with the customer's systems to access customer data), studying client experience and improving dialog experience (adding sajest buttons to simplify communication, working on high-quality sound and interrupting capabilities) and the development of dialogue opportunities, for example, flexibly switch from topic to topic and take into account the context and history of dialogue at any point.

Alexander Krushinsky, Director of the Department of Voice Digital Technologies, BSS, calls another important condition for success - the toolkit.

The convenience of the tools is one of the main factors affecting the time-to-market, the expert notes. - In addition, as our practice shows, it is very important that the customer can build the bot skills himself, without the need to order improvements from the vendor or hire expensive specialists with a narrow specialization. And for this you need a convenient no-code toolkit. Both for configuring bot scripts and for training recognition models.

Svetlana Zakharova, head of work with partners and clients of Just AI, cited several components of success in the development of chat bots. The first of these is natural language processing (NLP) technology, which includes algorithmachine learning, neural networks and semantic analysis and involves training the bot to recognize and respond to user intentions accordingly.

Write in detail the user's path and conclude what information he needs, what goals he pursues. This will make it easier to write useful replicas. Dialog control is the process of controlling the flow of conversation between a chatbot and a user. Anticipate user requests and responses: Think through the common questions and answers they may have and prepare the bot for effective interaction, − she advises. − Think through the bot's personality and tone of voice. Plan performance indicators: Based on your goals and objectives, identify key performance indicators for your chatbot. These can be KPIs for number of calls, response rate, user satisfaction, and so on.

The expert calls the positive experience of the client the third component of a successful chatbot and recommends designing such an interface of an automated dialog system and the course of the conversation taking into account the user's needs.

The fourth term of success, according to Svetlana Zakharova, − careful testing of the chatbot before release and adjusting its design and functionality based on user reviews.

And finally, the fifth term is the one that Sergei Popov called, − the integration of a robotic assistant with the customer's internal systems, as well as with third-party services.

The secret of success really is, − admits Anna Vlasova ("Nanosemantic"). - It consists in the fact that the chatbot should appear at the right point of communication with the company's customers and solve business problems, and not just demonstrate ownership of a set of technologies. For example, a bot with the simplest possible technology - to show the form and offer to fill it out - will be super effective if the form is simple, does not raise questions from the interlocutor, and, in general, saves him time. But if the technology of filling out the form when communicating with a bot is used in the field of knowledge where the interlocutor-person is not so sure, then the same technology will cause only irritation, and people will throw a dialogue with the bot.

Since a bot can solve various problems, from technical support to "just chat," the expert notes, then indeed, the more technologies are used to create it, the more opportunities to make a text or voice assistant useful, in demand and, most importantly, bringing income to the business.

In the dialogue, the bot has to classify the replicas of the person-interlocutor. It is in this way that the bot "understands" the phrases addressed to it - the built-in neural network compares them with the groups or classes of replicas known to it, − she explains.

Neural networks allow the bot to highlight objects in the phrase: personalities, addresses, phones, mail and much more, the expert notes. Named Entity Recognition (NER) technology. It allows you to train the bot to immediately perceive all the information necessary for dialogue, if it sounded in one person's replica, and immediately use it, and not ask boring questions on a long list: say your last name, and now tell the phone, and now something else.

If the bot is intended not only for instant messengers (text communication), but for communication by phone or through any physical device (smart home devices, smart speakers, and so on), then it will have to understand the sounding speech and voice its answers. These are ASR (automatic speech recognition) and TTS (text-to-speech synthesis) technologies - also neural network technologies.

To fine-tune some bot and human communications, regular applications, rules, heuristics and other methods can be used to more tightly control bot responses.

Also, the bot can have additional modules connected to determine the communication language (Russian, English, Chinese, and so on), to check spelling and grammar, stylize bot responses, and many others.

In general, no technology is mandatory - although the bot should still have some "brains" - but again, the more technologies are involved in the bot, the more different communication problems it can solve, − concludes Anna Vlasova.

The feasibility of using AI in a chatbot depends on the tasks that it solves, − explains Sergey Ivanov, (AMT Group). The simplest dialog systems do not require artificial intelligence. For example, to collect readings of electrical appliances or confirm an appointment with a doctor, it is enough to use the final set of client responses or programmed buttons. But, for example, it is advisable to use a "smart" robot for the task of classifying cases. Especially if the company has many products and topics of appeals. In this case, AI will help increase the accuracy of topic recognition, increase customer service metrics.

We have all encountered "push-button bots" or "menu bots" in Telegram and other instant messengers. They may well solve simple problems of helping to choose a product/service, sell a product/service, sign up and unsubscribe, and so on, − she notes.

According to the expert, smart bots are needed by commercial companies in cases where they are required to give information to clients on a variety of issues or advise in a complex subject area. It is impossible to navigate the menu in such a situation, because the menu will be too long and complex. Thus, the bot has exactly the same tasks as the person: to understand the request in the most diverse form, to collect information throughout the conversation, and not just from one phrase, to remember previous conversations with the same interlocutor and use information from previous conversations, and a whole range of intellectual tasks.

And there are other tasks, up to the maximum imitation of human communication behavior, i.e. masking a bot as a "living" person. Here, without intellectual technologies, the bot will be exposed very quickly, − concludes Anna Vlasova.

Companies interested in maintaining competitiveness should focus exclusively on bots with artificial intelligence as a smarter option for interacting with users in a "convenient" language for them, − she stressed, adding that ordinary chatbots in modern realities have long been outdated.

In addition, according to the expert, the prices for effective robots with AI and NLU are currently the same as for simple automated dialog systems.

Over the past five years, chatbots have proven their benefits for business both globally and in Russia, which is confirmed by the high growth rates of these markets. Thus, the global chatbot market was estimated at 17.17 billion US dollars in 2020, and, according to forecasts, by 2026 its volume should reach 102.29 billion US dollars, that is, the CAGR will be 34.75%.

The Russian market for chat bots, or conversational AI in 2020, amounted, according to Just AI estimates, to $76 million, or 5.5 billion rubles, taking into account government orders.

In 2020, largely due to the pandemic and a certain maturity that speech technologies have achieved, there has been an explosive increase in the demand for speech technologies, − says Alexander Krushinsky, Director of the Voice Digital Technologies Department, BSS. − We definitely record this and see by the interest of companies and organizations in different areas. By the number of projects, increasing every year. In terms of the scale of these projects. On the interest of customers to develop and improve the used speech solutions. I cannot speak in numbers for the entire market, but our revenue growth indicators from speech projects unequivocally demonstrate positive dynamics.

The growth rate in the first five-year plan of the 20s was predicted by it in the range from 38% to 81%, which was supposed to provide a market size of $561 million by 2025.

How do its participants assess the state of the chat bots market in the Russian Federation at the end of 2022? Arseny Kondratyev, senior manager of technology consulting practice Axenix (formerly Accenture), notes that at the moment the chatbot market in Russia is in a turbulent state and that the demand for bots to optimize costs is growing.

According to customer service research conducted by Naumen, in 2022, chatbots launched in chats on the website, mobile application, instant messengers or social networks were one in three retailers from the top 120, an increase of 55% compared to 2021. Chatbots in unauthorized channels were held by 23% of banks from the top 100, which is 21% more compared to 2021, and 15% of insurance from the top 75. In insurance, the increase in the share of companies with chat bots amounted to 83%, which is due to the "low base" effect: the level of penetration of chat bots into this segment is relatively small.

As for the volume of the market, Naumen estimates it slightly lower than Just AI experts. This, in particular, can be associated with different market boundaries.

If we talk about the total volume in monetary terms, robots in voice and text today make up the market of 2.8 billion rubles, − says Sergei Popov, director of the department of contact centers and robotic systems at Naumen.

By the beginning of the second decade of the 21st century, more than a hundred players were present on the Russian chatbot market. Among them are companies that initially developed speech technologies: speech synthesis and recognition, and then supplemented them with technologies for natural language processing (NLP - Natural Language Processing), natural language understanding (NLU - Natural Language Understanding), as well as Dialog Management dialogue management.

As mentioned above, the pioneers in this market were the CST and Nanosemantics companies, which have developed and are developing their own technologies. Thus, Nanosemantics have a professional dialogue system of the corporate level DialogOS, designed to create intelligent voice and text robots; native language of linguistic programming Dialog Language; as well as speech recognition technologies NLab Speech ASR and speech synthesis NLab Speech TTS (included in the Unified Register of Russian Software).

Our JAICP platform for the development of bots of various complexity includes an in-house developed NLU core that allows chat bots and voice assistants to understand natural language, − comments Svetlana Zakharova, Just AI.

Hyperscalers who have formed their ecosystems, in which there was a place for voice assistant developers, stay a little apart in the chatbot market. These are SberDevices, Yandex Cloud, VK, offering their speech-based products as cloud services. For example, the Yandex SpeechKit tool is actively used to create and run voice robots on Yandex.Clouds virtual servers by developers and integrators that are part of its ecosystem. The SaluteBot platform, which is being developed by SberDevices, allows you to create chatbots that will respond to customers in any communication channel of the company and can take on up to 85% of calls. The developer also has the SaluteSpeech platform, which underlies all speech services and many SberDevices products and devices.

Arseniy Kondratyev notes that the Russian market historically has stronger recognition and synthesis of the Russian language, but less mature platforms for developing dialogue logic, including tools for processing natural speech - NLP. He calls the creation of his own platforms by large customer companies, especially in the banking sector, another distinctive feature of the Russian market.

In the west, companies are betting on platform solutions. And this is not surprising, because a number of platforms: botfire, Chatfuel, pandorobots, INF.AI allow you to quickly create a chatbot with basic functionality, and then work out its correct response to user requests on the database, connect them to instant messengers. In general, in the list of leading players in the global chatbot market, all leading technology companies: IBM, Google, Nuance Communications, AWS, which have recently been present on the Russian market, but have not shown themselves in this capacity.

One of the factors that had a noticeable impact on the market was the departure of a number of large foreign players associated with the sanctions imposed against our country, as well as the blocking of sanctions applications in the AppStore and Google Play caused by them.

According to Svetlana Zakharova, head of work with partners and clients of Just AI, the latter factor influenced traffic and even the fate of some chat bots that were at that time under implementation.

We feared a drop in business activity in the market, but have not yet seen dramatic changes. On the one hand, projects in some representative offices of international companies were suspended or canceled. On the other hand, vendors from unfriendly countries have left the market, and this, of course, opens up opportunities for import substitution for us, − recalls Alexander Krushinsky, director of the voice digital technology department, BSS.

According to Arseniy Kondratyev, the sanctions brought to life two main trends in the market for the introduction of chat bots, which turned into significant additional costs for companies. Firstly, a reactive replacement of solutions of vendors began, which ceased to support either all Russian customers, or those of them who fell under the sanctions, and secondly, a proactive replacement of the introduced foreign solutions with Russian analogues or the development of their own products.

Russian solutions for chat bots were the leaders in our market even before the imposition of sanctions, − says Sergey Ivanov, head of the AMT Group contact centers department, − and now they have strengthened their positions even more. The first shock from the sanctions passed, and the companies continued to develop automation of customer service. It can be argued that we have successfully overcome the crisis. This applies to both chatbot manufacturers and companies that serve customers with bots.

In some cases, the launch of helper bots can be time-consuming or delayed − he notes.

One of our banking clients in 2022 focused on launching a voice assistant as soon as possible to replace the outdated IVR menu and cope with a flurry of incoming calls by product after advertising campaigns. IVR improvements were very limited due to the fact that IVR as part of the contact center was deployed on a foreign platform, and the bot assistant helped unload incoming line operators, route calls correctly and maintain a high level of service, − said Sergei Popov.

Despite the changes in the IT landscape that occurred in our country in 2022, the leading players in the automated dialogue systems market were able to add several rather large projects to their assets.

Since 2022, we have strengthened the direction of automation of outgoing communications, − says Sergei Popov. − The team helps companies launch voice bots for outgoing calls according to the service model, for example, sales, surveys, work with orders and delivery and takes all launch and setup work on itself. The service provides three conditions that are important for outgoing call tasks: quick launch, conversion orientation, as well as a transactional payment model only for effective calls. "

We think that internal automation, which improves the experience of employees, is one of the areas beyond which the future lies, − explains S. Popov.

Today, its implementation and debugging has ended for most of the clients, − comments Stanislav Ashmanov, its general director.

According to him, this platform uses deep neural networks and technologies of computer linguistics and allows you to reduce the load on the call center by 90%.

The central project of Naumen in 2022 − the comprehensive development of a contact center in Дом.РФ Bank: migration to Russian software with a move to the Naumen Contact Center platform and the creation of a voice assistant for processing mortgage lending requests on the incoming line. The work of the bot assistant made it possible to cope with an increase in the customer base by 30%, while it was possible to reduce the service time of 68% of them. As a result, the voice robot releases more than 80 working hours per month for each operator. Automation of the incoming line also provided a high level of service - 80% of calls are received within 20 seconds.

Another project to modernize the contact center of the credit institution was implemented in UBRD. Intelligent robots, text and voice, on the Naumen Erudite platform helped the bank cope with the growth of client requests and reduce the burden on operators: collectively, a chatbot on the site and a voice assistant on the incoming line took on about 27% of calls and messages coming to the contact center. By combining robotization with such Naumen Contact Center capabilities as a single operator working window and transparent work control, the bank's contact center managed to get a 25% increase in efficiency.

For Magnitogorsk Iron and Steel Works, we have implemented virtual assistants on the enterprise's website, in the Telegram messenger and the My MMK application for round-the-clock support of employees and suppliers. The chatbot is available 24/7, maintains 19 consultation scenarios and works with 26 topics, providing support on various procurement issues. An intelligent robot, as an element of digital transformation, helps to save time for specialists and optimize the company's business processes, − said Sergei Popov.

These are realistic (or stylized) 3d characters with lipsink (lip synchronization of the speaker with the text-approx pronounced by him. ed.) And animations in which we combined all our many years of experience in speech recognition and synthesis, as well as computer vision for user recognition, − said Stanislav Ashmanov. − Such digital characters can be integrated into websites, mobile applications, AI stands and atypical devices. We see great potential in such interactive systems, for example, for the HoReCa segment, shopping centers, airports, etc., and have already launched several pilot projects. "

Denis Afanasyev, director of the B2B products division of Salut SberDevices, cited an example of a bot placed on a Sberbank technology showcase.

In this case, the bot carries not only service functions and answers typical customer questions about products, but also informs about discounts, collects contact details of customers who applied after hours and does much more, "Afanasyev said.

According to him, the use of the bot allowed not only not to increase the support staff with a multiple growth of the portal audience, but also to create additional leads for the purchase of products. A positive impact on the CSI (Customer Satisfaction Index) and NPS (Consumer Loyalty Index) service metrics was recorded due to the fact that customers quickly receive an answer to their question, but can always contact the operator if their question is complex and requires human participation.

As the digitalization of all spheres of life continues, the need for a variety of businesses in voice and text assistants will grow. And, as a result, the tools for their development will become more accessible and easier.

Technologies are becoming more practical and simplified - now you can design chatbots in many tools without special training. They are added to many procedures in organizations. Therefore, the unification of employees, RPA robots, chat bots into effective processes is one of the key areas that determines the development of integrations of chat bots into complex solutions and platforms for robotization, − notes Pavel Borchenko (Robin).

Svetlana Zakharova (Just AI) believes that in the future this role will use platforms that integrate with the technologies and products of other vendors that small businesses need - CRM systems, telephony, etc.

In the future, in such interfaces without code, it will be possible to create more and more intelligent bots with advanced dialog skills. This will make robotization available to a larger range of companies, since the no-code approach saves the costs of a team that will develop and support the already launched AI assistant, − predicts Sergei Popov.

The main purpose of digital assistants will remain a client service, which means that they will become more and more human and empathic. For this, according to the expert, the answers of voice robots will be personalized, the accuracy of understanding and interpretation of client requests will be increased, variable formulations for the same answer will be used, which make the dialogue alive.

Moreover, − says Anna Vlasova ("Nanosemantic), − that the voice bot will be required to be as similar as possible to a person both in sound and in the manner of conducting a dialogue. You can predict the appearance of bots with specific voices and favorite "words" of famous personalities. In other words, Dmitry Nagiyev, who is already firmly associated with the MTS brand, may be paired with a chat bot Nagiyev, which completely imitates the style of communication, voice and - why not - three years later, the appearance of the actor. And any client, when calling the support service, will communicate with Nagiyev-bot. By the way, this is a completely legal practice, provided that the company receives the rights to use the voice, appearance and everything that will be indicated in the contract.

In particular, we are developing the so-called domain solutions - script templates and recognition models focused on a specific subject area (finance, medicine, power, etc.), − he notes.

According to Svetlana Zakharova (Just AI), chat bots will also be used in internal communications. Today, a lot of funds are invested in, industry where a lot of people work and, accordingly, there are a large number of internal processes that management wants to automate.

ChatGPT is the latest generation of neural networks trained on the latest algorithms on a large amount of data. ChatGPT is capable of communication that is so reminiscent of the natural that it practically relieves the discomfort usually experienced by a person when communicating with a robot, − she says. - At the same time, this technology allows you not to prepare all possible responses of the chatbot in advance, but to entrust artificial intelligence to generate a suitable response.

And although the neural network − not a panacea or a complete replacement for a person in communication, it can be a good addition to the intellectual technologies "under the hood" of a modern chatbot, − says Anna Vlasova.

Chatbots, once created to process text messages, are actively evolving. In a little more than a decade of active use, automated dialog interfaces have learned to "understand" the voice and have gained the ability to pronounce words.

At the next stage, they are likely to gain the ability to sympathize with their interlocutors, and then, possibly, appearance. Turn from text bots to full-fledged digital agents. And they will form a completely different client experience.

• Speech technology
• Chatbots (Chat-bot)
• Chatbots in medicine
• Chatbots in banks
• Virtual Assistants (Global Market)
• Artificial intelligence