from theory to practice

Initially, chatbots resembled regular answering machines. Businesses considered them only to replace routine operations of redirecting the user from one operator to another.

First breakthrough happened with the spread of the “bot push-button interface” in Telegram, which became the “de facto” standard in the field of text bots.

The second breakthrough The spread of speech-to-text systems can be considered, thanks to which it became possible to interact with systems using voice commands.

Third breakthrough emerged in the field of artificial intelligence (AI), bots became more “human” and interaction with them moved from the category of “robot” to the category of “assistant”.

The first examples of chatbots were developed mainly for B2C interaction, i.e. as an “autoresponder” for a potential client.

This type of interaction showed low efficiency in the conditions of the struggle for a potential customer of the service and was practically forgotten before the invention of GPT. The result of the development of this direction was a wide coverage of the B2B segment with an emphasis on the internal use of chatbots.

Instead of serving external interests, text bots began to process internal requests, such as sending news or providing quick access to popular information.

Implementation of chatbots in HR processes of companies

• sending notifications

• preparation of template documents

• exchange of information

• conducting surveys among employees

The main task of all text chatbots of the first type is to standardize the methods of information delivery. In this mode, they work 24/7.
Instead of a dozen channels for transmitting documents such as “vacation application,” only one is selected – a bot.

In it, you can quickly obtain an up-to-date document template and immediately submit a signed sample, request a 2-NDFL certificate, receive a salary statement or request a copy of a work book and much more.

The tool proved particularly useful during the Covid-19 pandemic, when it was necessary to send and receive thousands of documents per day to ensure that employees could access their workplaces in retail facilities.
In addition, automated collection of feedback from employees who have left the company, as well as in the formation of a list of company values, has proven itself to be quite effective.

The purpose of the second-level chatbots was to train or motivate staff.
There are several different products under the general designation of “educational and motivational complex”.

Such complexes contain a user part in the form of a text chat bot, an operator part in the form of a WEB shell with reports, and an administrator part in which what users see is constructed.

Experience in implementing such products shows high user involvement in the process. This result is achieved thanks to the gamification system and the introduction of game currency, which can be exchanged for real prizes for achieving the best results, if necessary.

Such systems are somewhat more difficult to manage, but the user experience of working with such complexes is much more positive.

The success of text chatbots and the development of related technologies have contributed to the further evolution of the direction:

• Neural networks make it possible to recognize voices

• The concept of a “smart home” emerged, which required voice control

• Large investors from the banking sector were found who needed ready-made technology

The initial demand was simple – to recognize the speech of bank employees to assess the quality of their work. Such systems became the ancestors of voice assistants, since it was on the basis of their DataSets that smart devices were subsequently trained.

An organic development of this direction has become bots that call people “according to a script”.
On this basis, a competitive struggle between “smart devices” began, which allowed businesses to form a new vision regarding the possibilities of using voice chatbots.

Implementation of audio bots together with transcription systems in the B2C segment as a replacement for a “live” operator.
The task of all audio bots was to save money on call center operators, as well as to reduce the waiting time for a response by the service consumer. An additional advantage was the ability to transcribe the operator's voice to identify deviations from the agreed template.

Practice has shown that such robots successfully save money and check templates, but the response waiting time has not decreased, since many users demanded to call an “operator”. Thus, the negative from waiting for a response was replaced by the negative from “communicating with a bot”.

Next came the niche of spam calls, in which a voice robot acts as the “first line of calls,” but without serious analysis of the responses, such services caused nothing but irritation to the end recipients of the service.
As a result, a market for services providing transcription services has emerged.

The development of chatbots is confidently moving towards “smart assistant” systems.
However, many experts estimate the return on such investments as low, considering AI technologies a “bubble”.

It is likely that the market is now looking for a niche in which such a service can become profitable at least in the medium term.
At the moment, only very large companies that can afford huge investments in training their own models can boast of their solutions in this area.

The rest are content with open integrations with all the ensuing risks in the form of instability of services and dependence on the pricing policy of the service provider.
In particular OpenAI began discussing raising the cost of using its services amid rising interest in ChatGPT.

The author is not aware of any commercially viable launches in this area.
However, attempts to launch services based on AI technologies are observed steadily.
Thus, in China, by 2024, ~80,000 startups related to neural networks closed.

The features of the technology determine its further development and the range of tasks it will work with. It is already clear that voice, video and image analysis is impossible without ML technologies.

AI-based bots are aiming to conquer the user interface segment, i.e. to become a universal remote control for interaction with a person. Instead of dozens of interfaces, only one remains – a voice command.
And instead of a dozen unverified sources – one trusted one.

At this stage, the technology is in the R&D stage, when a person is looking for how he can interact with it.
This observation is confirmed by observations of children who, due to the lack of other means of access to information, actively interact with smart devices.

Thus, the B2C sector can expect a surge of interest and monetization opportunities in this area in 10-15 years, when children who are already mastering the nuances of voice interfaces become adults and solvent.

Conclusions

Today, it's not the choice of a cutting-edge bot that matters, but how the technology is structured and used within the company.

Unlike text bots and transcription systems, launching AI-powered bots is a complex and labor-intensive task that will require significant investment and human resources.

Based on the experience of implementing technologies in different companies and general trends in the development of systems, several conclusions can be made:

• Machine learning-based systems are in their infancy.
  They are not commercially profitable, but they can perfectly solve standard tasks of generating texts and images without copyright.

• The implementation of voice-to-text translation systems allows to simplify the task of interaction with the user to the level of “text chatbot”.
  This service will also be useful for mass checking of dialogues for the presence of target words.

• Most of the internal processes of any companyas well as the first line of communication with clients can close a regular text chatbot based on standard algorithms.

All this suggests that it is too early to write off the technology and in the near future it may show significant growth due to the discovery of new methods of interaction with users.

Maxim Fabrin

Head of Software Development Department, Lavka Tekhnologiy