2022/08/03 15:23:28

Smart Process Processing

Intelligent technologies today have stepped far beyond individual operations, such as semantic analysis of a particular document. AI solutions began to become an integral part of very complex business processes. In what form is the "process intelligence" implemented today, and in what direction is its further development expected? The article is included in the TAdviser review "Artificial Intelligence Technologies"

Content

The "intelligence" of processes significantly depends on the field of activity
"Professional background" smart bot
AI in niche corporate processes
- AI in HR
- AI in marketing and sales
Other Review Materials
Other materials on the topic of AI

In August 2021, the analytical company Statista released a detailed report on the state of the AI market in 2021. According to analysts, automation of processes remains the main trend in the development of artificial intelligence, because it is the use of AI that improves the performance of business operations compared to other technologies in almost everyone industries and even makes a significant contribution to. GDP countries For example, for industrial production, AI could increase gross value added by nearly 4 trillion. (dollars about 372 trillion rubles) by 2035, for wholesale and growth retail will amount to 2.2 trillion dollars (USA about 205 trillion rubles), for industry information and communications - 1 trillion dollars. USA (about 93 trillion rubles).

The most popular areas of automation of business processes using AI

Source: Opinion poll conducted by IBM in 2021

Among the most popular areas of automation of business processes based on AI, Statista analysts include:

Personnel management automation
Automation of customer service (scoring, borrower models)
Automated analysis and prevention of cyber threats
Intelligent automation of business processes

On the Gartner Hype Cycle chart for the digital workplace, published in 2020, Smart Workspace technologies are at the top of expectations, and the productivity plateau is planned for 5-10 years.

Источник: «6 Trends on the Gartner Hype Cycle for the Digital Workplace», 2020

The intelligent workspace uses the increasing digitalization of physical objects to create new ways of working and improve workforce efficiency.

Examples of smart workspace technologies include: IoT, digital signage, integrated workplace management systems, virtual workspaces, motion sensors and facial recognition, etc.

And here, from the point of view of AI, we observe not only object recognition, but also planning, as well as solving such an applied problem as road safety (correctly assess the situation in the city, see a pedestrian in a timely manner, do not hurt anyone when rebuilding). And we see that from this point of view, the complexity of such a business process as managing an unmanned taxi is quite great, the expert emphasizes.

The introduction of recommended analytics and automated decision-making tools in the enterprise allows you to complement the "digital twin" of the enterprise with a model of business decision-making processes based on combining data from different sources and knowledge base and automate the processes of generating recommendations and instructions. This allows us to talk about the implementation of the concept of a "smart enterprise" in full. The development of the Triaflai platform and the tools built into it is built taking into account these approaches.

What specific features of process automation are important for the development of this area of activity?

The "intelligence" of processes significantly depends on the field of activity

So, I believe, a significant part of public services can be safely transferred to a digital employee, - says the expert.

A number of prerequisites contribute to this.

Firstly, all procedures are strictly regulated: obtaining certificates, tolerances, statements, accrual and payment, etc. taxes HOUSING AND PUBLIC UTILITIES That is, there is a knowledge base on operations, - explains Alexander Kazennov. - Secondly, all these processes are most often carried out sequentially, that is algorithm , working out looks quite simple, even for non-standard situations. Thirdly, in our to the country source data at the moment is quite well digitized, or present in the form of accumulated paper images to be recognized. Even water//electricity/heating meters are gas already installed with built-in modems for transferring data to the processing and charging center, the specialist notes. - Accordingly, we improve the quality of primary data - we improve the quality of processing. In this area, ideal digital employees have already been practically implemented.

"Professional background" smart bot

Modern software robots (RPA) look good from the point of view of the speed of routine operations interleaved with automatic recognition of texts, speech, etc. intelligent actions. True, it is still difficult to call a full-fledged "employee" of such a robot, because it does not have "professional knowledge" that makes up the information background of a real specialist (or the context of a software robot).

Anton Ermakov, head of Comindware's digital initiatives group, says that evolution is already underway in this direction: for example, modern chat bots hold the context of a dialogue and are able to talk in something like this:

You: how many mamziks are left in our warehouse?
Bot: Five.
You: order five more.
Bot: OK.

Bot understands that this is about mamziks, and that you are asking to order five mamziks, not five burchiks. Modern bots are also able to understand the intonation and mood of the interlocutor - you just ask if you are annoyed or about to rise up - and takes this into account in their answers, - says Anton Ermakov. - But a full-fledged employee, of course, is still far away - this is a task for "strong" artificial intelligence.

I think that the near future, like the present, is a combination of PRA/IPA and human labor, says the specialist. - Already today, robots call us and talk to us on the topic of banking services, offer goods and sell them to us in online stores. True, not everyone and not always it turns out organically.

The idea has matured, mastered the masses, and Gartner analysts have proposed a wrapper for it called hyperautomatization, which implies a combination of new digital technologies, which gives a synergistic effect, - notes Anton Ermakov.

Support for business process changes

Even the rules of the road are constantly changing and supplemented, and even more so are the business processes that take place in a company or in production. Therefore, if the system does not take into account this constantly changing landscape and works according to the old rules, and not according to the new ones, no one will need it, because in this case following the old rules will be an obvious mistake.

When using human intelligence, this problem is solved through formalization, that is, a detailed description of tasks and processes, which allows you to control something within the model, when the language has already been developed, "Dmitry Nikolaev reflects. - In the case of AI, difficulties no longer concern neural networks, but the need for a clear understanding of which model our business processes are proceeding, and which language should be chosen to better describe them.

For example, on the one hand, in the reconciliation report, clear information will be submitted from the system of one counterparty, and the accountant of another counterparty will not agree with it and initiate correspondence. Such situations of modern RPA/IPA without a person have not yet been resolved.

Therefore, here we can only talk about smart algorithms and only in combination with a huge number of sensors and mathematics. But part of this responsibility can be shifted to RPA/IPA, but under our "human control," sums up Alexander Kazennov.

Processes should be fast and effective, but the percentage of errors should be minimal, or better without them at all, - Elena Martynova, deputy head of the department, commented on the event.

True, while a previously trained neural network is able to make about 80% of the correct decisions on transactions. Therefore, "living" employees still remain in the working order.

Semantic Data Models for Process Management

Semantic data models today penetrate various corporate systems, for example, form a semantic data layer in large corporate data stores. The enterprise architecture in the Zachman Framework is described using the ontology engine. Ontologies are used to describe processes in ABBYY products. Business processes for the Comindware Business Application Platform are also described using ontologies.

The ontological approach to describing business structures and business processes takes the minds of an increasing number of professionals, - notes Anton Ermakov from Comindware.

According to Rinat Gimranov, head of the information technology department of Surgutneftegas PJSC, this is due to the fact that only ontological Low-code platforms will be able to cope with the flow of a variety of changes taking place in business processes.

It is necessary that real world changes in real time fall into the virtual world, - says Rinat Gimranov. - If we are talking about the data flow, then this is done easily today. For example, with the help of the Internet in real time of things, billions of indicators are collected from trillions of sensors. But if we are talking about changes in the structure, in the process, in the metadata, then today this is done in the overwhelming majority of cases manually - with the help of programmers.

Источник: Почему у low-code нет альтернативы в digital economy. Rinat Gimranov, Surgutneftegas, report at the Low-code conference, "Open Systems," December, 2021

Because the number of these changes is huge, because we are striving to display as many details of the real world as possible in the virtual world. No programmers will be enough to program these changes in seconds. We need platforms that learn to understand the real world in real time.

If we are talking about changes in business processes, then they must take place in the mode of natural dialogue.

If the platform is built in the traditional way - without a semantic engine, then its flexibility is limited by the amount of programming resources that can be invested in the development of the platform.

Ontology as a language that can be spoken by the real world and the virtual world

Source: T. Gavrilova, D. Kudryavtsev, I. Muromtsev, I. Knowledge Engineering. Models and methods: Textbook - St. Petersburg. Publishing house "Lan." 2016. - 324 s silt.

Distributed Business Process Management Intelligence

Among the top 10 trends in data processing and analytics formulated by Gartner in 2020, the trend "Artificial intelligence will become distributed and" responsible "is in first place. It should be noted that the origins of distributed intelligence should be sought in military technologies - for many years, methods have been developed for organizing network-centric (that is, in a single information space of computer networks) collective actions of smart autonomous technical systems. Thus, the long-term comprehensive program of the US Department of Defense Future Combat Systems (FCS) providing for the use of network centric control systems has been implemented since the early 2000s. And for the practical testing of its individual results, the period 2012-2022 is allotted.

The FCS program defines the architecture of promising globally distributed network centric control systems for combat units, their groups, systems and means of support. It is a multi-level system of network-centric integration of stationary and mobile objects of various purposes, equipped with built-in computer intelligence, in a single information-functional control space with their interaction in real time.

Network centric control of combat units within the framework of the FCS program

The key system-forming element of the program is communication facilities and computer networks, which ensure the maintenance of a functionally integrated information field over combat zones by building reliable operation of mobile wireless networks. Autonomous elements are connected to a common network and exchange information. Thus, participants in real time are provided with data on the location of the enemy, coming from remote automatic sensors, reconnaissance vehicles, unmanned aerial vehicles, etc.

Network-centric control systems of this kind are designed to ensure the functioning of a single and ultra-reliable information space, in which complex problems of collecting, accumulating and intelligently processing multi-channel streams of complex structured data are solved in real time. The goal is to form a single dynamic picture of events and ensure superior quality of control of large systems of multi-profile multi-component systems of mobile and stationary objects, which guarantees the achievement of the set goals with minimal losses.

The management system of such groups should have artificial intelligence, means of performing complex calculations. The group management algorithm should determine the conditions and methods for calculating the rational composition of the combat group, solve the problems of situational modeling, targeting and assessing the effectiveness of the task, - say the specialists of the Combat Aviation project complex of the National Research Center Institute named after N.E. Zhukovsky in the article "Artificial intelligence for aviation systems," published in the journal Arsenal of the Fatherland, No. 4, 2021

Advanced Intelligent Aviation System

Source: Artificial Intelligence for Aviation Systems A. Lukashov, O. Panferov, V. Maksimov, A. Bashkirov, Arsenal of the Fatherland, No. 4, 2021

The Intelligent Aviation System (InAS), built on the basis of manned and unmanned aerial vehicles using artificial intelligence technologies, will allow combat operations and multi-component missions characterized by a high degree of uncertainty. At the same time, UAVs from InAS should act not only autonomously, but also collectively, as part of a group. Moreover, each object from the group can perform various tasks in accordance with the plan formed by artificial intelligence in real time.

Each layer/control layer serves as the information provider for the next layer and simultaneously uses the data from the previous layer. Thus, from disparate data on infrastructure, services and assets, a single vision is gathered on how IT the - direction should work in accordance with the need of the business.

Moreover, in this multilayer structure there are at least three layers: strategic, tactical and operational.

In the strategic layer for which the CEO or Chief information officer is responsible, a common set of requirements and rules is defined. At the tactical level, a process is organized to provide effective services: value chains and interactions with other units are described, areas of responsibility and areas of ultimate value formation are determined. Service performers operate at the operational level.

In this structure, it is critically important that many participants are involved in the process of providing services, service managers themselves establish them depending on the peculiarities of the work, - explains Dmitry Rubin. - The process ceases to be linear and does not depend on any particular employee. In other words, it becomes a self-governing system.

This approach was implemented, for example, in the Federal Treasury to organize effective work over 40 thousand. IT employees throughout Russia.

Agent-based systems

Another promising approach to organizing dynamic interactions and information integrations in intelligent systems is multi-agent systems. It assumes a flexible decomposition of all the many problems solved by the system into a number of agent programs that are able to solve some problems at the human level, that is, those problems that require built-in intelligence. The most famous example of an agent approach is Internet bots that simulate human activity on the Internet. In general, agents are able to obtain a task, communicate with other agents, receive missing additional information from them and use it to perform a user-defined task.

General properties of software agents:

Autonomy (can operate independently of the user).
Adaptability (capable of learning while working).
Communicativity (capable of communicating with the user or other agents).
Ability to cooperate (works with other agents to achieve the goal).
Personification (behaves like a person, for example, shows emotions, etc.).
Mobility (can move around the environment).

In the world as a whole and in our country, in particular, the development direction, called the "Agent-Oriented Models" (AOM), is very actively developing. First of all, the focus of attention of such systems is large-scale tasks of socio-economic monitoring and analysis, as well as tasks of this kind in the military sphere.

Thus, the agento-oriented approach turned out to be very successful for the tasks of modeling artificial life. For example, the MANTA system, focused on creating a software environment for the study of social systems, is based on the behavior of an ant colony. Another example is the Tiera genome-level system for modeling the evolution of artificial organisms. The main feature of the Tiera model is that programs are altered by random mutations and permutations, but remain feasible, although they may lose their functional characteristics and utility. This property of variability is used in the Tiera model to control the evolution process: random mutations are introduced that lead to the creation of other individuals that are unable to reproduce, or to offspring that reproduce more often and faster than their ancestors. These examples explain why autonomous agent technologies have become very popular in the field of evolutionary computing, primarily affecting aspects of self-recovery and self-configuration of complex autonomous systems, which consist of a large number of simultaneously functioning software modules.

The emergence of AOM can be considered as another stage in the evolution of the modeling methodology: the transition from mono-models (one model - one algorithm) to multi-models (one model - many independent algorithms). In such a paradigm, AOM is an artificial "society" consisting of interacting independent agents, which allows you to simulate a system as close as possible to reality. In the process of their existence, the created agents are able to analyze the data received from the environment, respond to them and replenish their experience (learn).

One of the largest AOMs - the US National Model, which includes data on the entire US population, was created at the Center on Social and Economic Dynamics at Brookings. It makes it possible to predict the consequences of the spread of diseases of various types. The 2009 version of the US National Model includes 6.5 billion agents, whose activity specification was based on available statistics. It simulated the consequences of the spread of influenza A (H1N1/09) virus across the planet.

RUSSIAN FEDERATION The researchers CEMI RAS MSU launched supercomputer a model at Lomonosov that simulates the development of the socio-economic system of Russia over the next 50 years. This AOM is based on the interaction of 100 million agents, conditionally representing the socio-economic environment of Russia. The behavior of each agent is defined by a set of algorithms that describe its actions interaction with other agents in the real world. Data for modeling were also provided by Federal State Statistics Service the Russian Economic and Health Monitoring Service. population The model is built using the ADEVS product.

The most versatile tool for creating AOM should be considered the AnyLogic modeling system. Models in the AnyLogic project are stored as an XML file containing a tree of parameters necessary for generating code: agent classes, parameters, presentation elements, descriptions of UML diagrams of agent behavior.

Multi-agent systems for different applications

The agent approach to modeling is very versatile and quite convenient for applied specialists due to its clarity. The result was an active study of a variety of mathematical models of agents and multi-agent systems (MACs), concepts and methodologies for multi-agent design and programming, agent programming languages and sufficiently developed tools and platforms for implementing multi-agent applications.

Classification of MAC (multi-agent systems)

Source: Agent-oriented systems: from formal models to industrial applications. A.N. Shvetsov, Vologda State Technical University

At the turn of 2000, a tendency began to appear from purely academic research projects in the field of IAU and AOS to the creation of existing multi-agent applications in various fields, industries information communication, systems state and organizational management.

For example, the Institute for Complex Systems Management Problems of the Russian Academy of Sciences and the Genesis of Knowledge NPK (Samara), together with the Administration of the Samara Region, created a regional management system using multi-agent technologies.

The region management system includes three main subsystems.

IAU of targeted interaction of the population and state authorities in the social sphere. Works on the basis of a social passport card.

The holder of a social passport gets the opportunity from a public Internet kiosk or using the Internet, from any place, to send a request for a benefit or service, as well as describe his cultural needs. The work of IA allows a citizen to receive offers from organizations in real time, taking into account individual preferences and form their own calendar of events.

The subsystem of targeted interaction has a knowledge base of laws, combining more than 40 laws of the federal, regional and municipal levels and storing knowledge of more than 300 benefits and payments and more than 100 related social categories. Based on a request from a person, his agent performs an operational search on the BZ and provides only the information that concerns a specific person.

IAU management department of the social block. Supports Community Managers and Management Process Managers.

The social unit management subsystem, in addition to the legislative BZ, also uses a knowledge base for the treatment of diseases based on global clinical protocols. Agents help to correlate between the diagnosis and the proposed treatment method with the BZ clinical protocols.

Internet portal for resource integration and interagency interaction. Allows citizens and organizations to describe their needs and opportunities; informs users about the best ways to solve their problems and realize their capabilities.

The resource integration portal uses the BZ of social actors and objects and their related needs, which helps to quickly find the opportunity corresponding to the declared need and receive offers from organizations contained in the portal, taking into account the configuration of the personal agent.

The regional management system uses the existing information resources of the departments of the social block, integrating them into one system, the existing databases of the Ministry of Labor and Social Development of the population are connected to it, Samara region database pension fund which store information are about 1.5 million residents of the region. Integration with existing databases data creates the possibility of developing a personal accounting system based on this project.

The use of a multi-agent approach made it possible to reduce the time for paperwork, create a single information space between departments and their subordinate organizations in the social sphere, and ensure transparency in the management and provision of services to citizens.

In the UK, an ARCHOTM electrical substation management system has been developed and implemented in the multi-agent systems (MAC) paradigm, which is used to eliminate identification errors and service restoration of the power grid. It consists of seven agents that are located at various management levels: SCADA, Energy Management Systems (EMS), Substation Automation Systems (SAS). In it, each agent is responsible for its own specific function, for example, identification of an unknown network or an interface of the management system.

The UK also has experience using IAU to control switching equipment in an electric power system. The system includes assistant agents, equipment agents, and distribution agents. Equipment agents represent transformers, buses and transmission lines, and monitor the operational state, service time and use time of equipment elements. Distribution agents represent switches or groups of switches. The system was implemented by the Java language and tested at the level of field modeling.

In addition, a multi-agent system is used to optimize power consumption. It contains agents of devices representing terminal devices of consumers, and service agents who participate in the auction mechanism for the purchase and sale of electricity. Experimentally, it was possible to confirm that the system is able to adapt to price changes within several auction steps. The advantages of the system are that it is capable of interacting with large-scale systems, and can be tuned to individual requirements in a wide range of applications.

In the South-Western State University (Kursk), the IAU was created, designed to build a system of intellectual support for decision-making based on the analysis and classification of chest X-rays (ICRS).

ICMS structure
and

Источник: ИЦРОН (Инновационный Центр развития formations sciences) of Southwestern State University

The core of the MCPC is a multi-agent hybrid image analyzer (MAGAI). MAGAI consists of four intelligent agents that focus on X-ray analysis. The inputs of these agents are X-rays that have been processed in an Image Preprocessing Agent (APOI). The purpose of this agent is to minimize differences in images that are associated with the equipment, the process of obtaining an X-ray image and the constitutional features of the patient.

MAGAI agents include informative feature space formers and decision blocks that are based on a neural network, a fuzzy neural network, and a hybrid neural network. The structures of these neural networks are stored in the knowledge base and can be used by MAGAI agents if necessary.

Researchers from the Vologda State Technical University under the direction of Anatoly Shvetsov, Professor of the Department of Informatics and Information Technologies, works in the field of building universal intelligent control models and decision support systems (SPMS).

Modern CISs (SAP, Baan, PeopleSoft, Oracle, iRenaissance, "1C: Enterprise") mainly automate mass routine operations, which can account for up to 80% of labor costs in corporate systems (CS). Automation of decision-making processes remains the most knowledge-intensive and difficult task, both from a technical and organizational point of view, says Anatoly Shvetsov.

According to the scientist, the specifics of the functioning of the CS increasingly require the use of distributed SPMS (RSPPR), which consist of local SPMS operating in the nodes of the corporate computer network, and jointly solving common problems based on the exchange of information and knowledge carried out both between the SPMS and between the SPMS and other subsystems of the CIS and distributed applications.

Management and Integration Levels in the Enterprise System

Source: Agent-oriented systems: from formal models to industrial applications. A.N. Shvetsov, Vologda State Technical University

To solve such problems, a paradigm of agent-oriented systems arose, using intelligent agents as a high-level abstraction for formalizing and structuring the subject area. Intelligent agents, for example, have a goal-building mechanism, which provides a fundamentally new level of autonomy: such an agent does not necessarily carry out the orders of some other agent or user, but acts, depending on the conditions of the environment, including the goals and intentions of other agents. For example, an agent can assume certain obligations or, conversely, refuse to do some work, motivated by a lack of competence, employment with another task, etc. At the same time, the agent can perform such actions as generating, suppressing and replacing other agents, activating functions (both their own and other agents), activating the activity scenario, memorizing the current state of other agents, etc.

In line with this approach, the tool complex DISIT (Distributed Intellectual System Integrated Toolkit) has been developed, designed to build a MAC. The fundamental difference of this approach is to transfer the focus of development to the stage of conceptual design and assign the task of logical and physical design of models and structures of knowledge and data to tools, emphasizes Anatoly Shvetsov. The DISIT complex is used in the research and training work of VoSTU, in particular, in the tasks of modeling the development of the city.

The MAS-DK instrumental environment, developed by specialists of the group of intelligent systems of the St. Petersburg Institute of Informatics and Automation of the Russian Academy of Sciences under the leadership of V.I. Gorodetsky, is focused on supporting the rapid prototyping of applied IAUs. It supports the life cycle of software systems from conceptual design to the generation and debugging of IA executable code. Class and agent specifications are used to generate the IA program code.

The application specification begins with the definition of domain ontology and interaction protocols. The following describes the agent classes and assigns roles when executing protocols. The next step is to develop variable class models, control models, state machines, behavior script libraries, and external function libraries. The MAC specification process at each of the stages is supported by integrated graphics editors.

This tool environment has been successfully used in the creation of AOS to protect computer networks, distributed training based on classification and intelligent data processing.

But interaction is a significantly more complex process, involving understanding the meaning of transmitted messages and, as a result, the presence of common and shared knowledge. At the modern stage, such knowledge, as a rule, is presented in the form of ontologies, - the author emphasizes.

With such a network approach to decision-making, the main task is to identify needs and opportunities, which implies through negotiations the finding of a resource that satisfies it for each demand. As a result of the search process, a network of connections (relationships) between them is formed, which can be interpreted as a network of needs and opportunities (PO network). In open systems from the external environment, at any time, a new need (or opportunity) may appear, which entails a change in the configuration of the PO network. "

Task transport logistics as a multi-agent network of needs and opportunities

Source: Open multi-agent systems for the operational processing of information in decision-making processes. Petr Skobelev, abstract dissertation for the degree of Doctor of Technical Sciences. Samara, 2003.

In particular, in this paradigm, a multi-agent network was created for the task of transport logistics. The main feature of the system is that instead of the traditional centralized command post, it uses a fully distributed model of direct interaction between order agents: logistics agents (LCs), service stations, roads and secondary logistics facilities.

General structure of the transport logistics system based on a multi-agent approach

This allows you to create the network in question as an open, more flexible and responsive to incoming orders or changes in the state of aircraft and service stations (failure, changes in weather conditions, etc.), more reliable and efficient, providing an individual approach to each order.

AI in niche corporate processes

AI in HR

Employees of the modern HR department are not at all a humanitarian warehouse of a girl with a description of psychological tests in her hands. HR is a data driven activity, that is, based on data processing, says Alexey Korolkov, CEO of WebSoft, whose IT solutions work in the human resources services of many Russian banks. And this is a promising actively growing segment of the IT market, absorbing the most advanced technologies, including artificial intelligence. According to analysts of MarketsandMarkets, the global market for human resources management systems will grow from $16.7 billion in 2019 to 26.5 billion in 2024 with an average annual growth of 9.7%.

Recruitment, personnel document management, distance learning - these segments of the HR service have already become sites where an intense competition unfolds between suppliers of relevant products.

A promising area is Human Capital Management (HCM), a human capital management system. Univertus, which positions itself as an HCM platform, predicts the growth of the management and personnel development systems market in Russia by 40% until 2024, to 5 billion rubles. with an average annual growth of 9%.

At the same time, the focus of attention of WebSoft is a comprehensive representation of the relationship between an employee and a company at all stages of cooperation. In fact, this is a variant of big data analytics that allows you to identify features and anomalies of work processes that affect personnel turnover, future employee performance, and also help to form a personnel reserve.

And the HCM-system Univertus systematizes the processes of management and development of personnel, which is based on performance management. Rather, it can be attributed to the Talent Management Systems class, as it is aimed at managing the development and training of personnel taking into account individual career trajectories, achieving maximum performance.

The foundation of the Univertus system is competency models that describe employees from various points of view: professional, management, corporate values and personal potential. According to Kirill Khramtsov, co-founder and director of methodology at Univertus, in the near future employers will make decisions on hiring an employee using an intelligent recommendation system that works with the candidate's portfolio, formed throughout his professional career and with details up to roles in specific projects and personal features of work in the team.

The T1 Watchman system provides end-to-end monitoring of the efficiency of business processes, which takes into account all the details of processes: from the efficiency of using specific computing equipment at the employee's workplace to the effectiveness of his own work, including monitoring compliance with the deadlines for solving tasks.

The necessary data is collected using software agents that are installed on employees' work computers, and then the collected data is analyzed using a set of performance metrics. These metrics are either technical (loading computing resources, disposing of applications, etc.) or business in nature (information about specific operations, their duration, compliance with regulations, etc.). In general, we are talking about automated analysis of the user experience of employees, and the monitoring data of this user experience serves as the basis for recommendations for optimizing business processes.

Critical elements of such complex HR systems are the construction of models, the development of metrics on the basis of which performance indicators are calculated, etc., as well as the possibility of integration with other corporate information systems: from descriptions of the organizational structure of the enterprise to, for example, the production MES system, if we are talking about the factory workshop and parameters of the employee's implementation of the plan.

Data collection is carried out mainly automatically, and on the basis of all available parameters (Big Data) various AI systems work: prognostic, advisory, explanatory (reasoning recommendations). According to Alexey Korolkov, in the future HR services will be integrated into the usual digital tools that employees use in their work. The simplest example is the remote launch of a training course directly from the business system, if you needed to learn new material. In other words, the personnel service will turn into an invisible service that can be called with just one click.

AI in marketing and sales

A key element of digital marketing is customer behavioral data. They have become a serious asset for B2B companies, since they record in detail how people interact with the products and services offered. Logically, behavioral data is best suited for AI analysis.

For example, Hermann.AI - developed by CleverDATA (part of the LANIT group) - is a platform designed to automate marketing communications and online advertising. The company calls this approach AI-driven marketing and says that the platform fully automates the work of the marketer using all available data sources: sites, CRM, mobile applications, transactional systems, external data, etc. - to build the most personalized communications at all stages of the client's life path.

The platform includes a built-in platform "Data Exchange," which contains data on various segments of external data from more than 9 thousand sources. All of them are available for targeted advertising campaigns.

Omnichannel communications are an important element of the platform. It allows you to optimize communication with customers in a multi-channel environment in accordance with the Customer Journey of each consumer, devices used, as well as online and offline data. You can synchronize all marketing activities with all channels and devices according to a single communication strategy.

Deep machine learning technologies are used for targeting tasks, personal recommendations and optimization of campaigns. The audience is automatically segmented in terms of consumer behavior, previous communications, relevant products, optimal communication channels and time, prices and sizes of discounts and other attributes for conducting personalized communications and advertising campaigns.

Predictive analytics is a tool for predicting the likelihood of loyalty outflows for subsequent segmentation and more accurate targeting.

Automatic personalization of advertising messages in all communication channels is supported based on a deep understanding of the audience. A system of generated feeds for embedding in real-time mode (Dynamic Creative Optimization) at the time of visiting the site, email SMS in/mailings, online advertising has been implemented.

The platform supports the recommendation engine: recommendations are generated automatically based on audience data to determine the most relevant offer.

In addition to supporting the current customer audience, the platform helps to find an Internet audience that is as similar as possible to current customers and continue to work with it: to show an advertising message at the moment when the consumer has a need, tracking marketing signals.

QSOFT in June launched an online sales platform for the farm industry using big data, AI and ML analysis technologies. The comprehensive solution includes a modern marketplace. One of the key modules of the marketplace is the Customer Data Platform (CDP).

As a result, the farm network receives a unique verified knowledge base about each of its customers with the formation of a single customer profile.

Based on the created base, using ML, buyers are segmented by behavior and purchases. This allows you to build interaction in omnichannel mode: site, chats, mobile application, e-mail mailings, push notifications, etc.

The QSOFT CDP platform was trained on a large farm network database. At the same time, the features of each of the regions (different prices, balances, conditions for obtaining goods) were taken into account. Various dependencies of purchasing behavior and preferences were also identified.

Using ML tools, the QSOFT CDP system allows you to customize the sales mechanisms of each pharmacy according to individual customer interaction scenarios. The fact is, says Oleg Demchenko, that each retail outlet has its own daily sales rate, depending on the location, the number of employees, advertising costs, etc. In the CDP system, the achievement of the sales rate is based on control points and depends on the time and day of the week, season and other things. The daily plot of the point is determined on the basis of Big data collected over a large amount of time (at least 1 year).

An interesting detail: in case of non-fulfillment by a specific pharmacy of control indicators and depending on their deviation, one of the scenarios for reaching the daily norm is automatically launched. For example, a sample is made on the best-selling goods and on regular customers who have not bought the goods for more than two weeks. After that, a personalized newsletter is sent to this customer sample with an offer to buy the most frequently purchased drugs by promotion/discount/pre-order. If the application of the scenario does not lead to the planned result, the following relevant scenario for additional samples of buyers and medicines is included. The Big Data-based engine helps the farm network marketing service generate a large number of customer scenarios based on their previous purchases and pre-objective analytics.

The solution allows you to interact more with buyers, increase the number of regular customers, collect data on orders and work with them to increase the average check and the frequency of purchases, "commented on the proposed approach.

The promising development of intellectual support systems for marketing and sales is associated with the trend of social commerce. The study Accenture "Shopping Revolution" showed that the global shopping industry in social networks with a turnover of 492 billion. in the dollars coming years will grow three times faster than traditional e-commerce and will reach $1.2 trillion by 2025. The growth will come mainly from Gen Z social media users and millennials, who Accenture estimates will be responsible for 62% of global social media commerce spending by this time.

Next Overview Material > >
> Browse Home > > >