Roman Styatyugin, an expert in product development and business development within the IT sphere, shares insights on the various tasks that can be solved with Generative AI, and how this emerging technology is transforming industries and shaping the future of business operations.
The excitement around generative artificial intelligence (GenAI) has driven various industries to adopt solutions based on this technology. According to a study by venture capital firm Menlo Ventures, enterprise spending on AI has grown over sixfold in 2024, reaching $13.8 billion. Additionally, 72% of CEOs anticipate greater adoption of GenAI tools in the near future.
Today, GenAI is much more than a tool for creating text and illustrations, interpreting data, and generating reports. It is increasingly being used to enhance communication with audiences, drive decision-making in key business processes, and even detect dangerous diseases at an early stage. Roman Styatyugin, an expert in product and business development within the IT sector, shared insights on how GenAI is helping organizations across various industries today. He has successfully launched numerous cutting-edge products that are not only widely used across various sectors but have also received prestigious industry awards.
Roman has over 25 years of experience working on B2B projects across a wide range of industries. He spent much of his career developing core banking systems before launching a startup known for its data-driven products in digital marketing. Currently, he oversees a team of ML and IT engineers. Over the past 10 years, Roman has built his team to develop application products in data analytics and machine learning. Under Roman's leadership they have successfully incorporated generative AI into their decision-making processes.
Today, companies from many industries are optimizing their processes with the help of machine learning. An important service, “GeoCursor”, was developed to assist retailers, banks, pharmacy chains, restaurants, and other businesses in selecting locations for new points of sale, forecasting revenue, and evaluating the competitive landscape. GeoCursor's machine learning models take into account more than 500 parameters, including anonymized audience characteristics, local infrastructure, pedestrian and vehicle traffic, and purchasing activity. Based on these factors, the system analyzes and forecasts sales data, turnover, and average purchase size.
Another product, "Developer," won the CNews AWARDS in the category AI Project of the Year in the Real Estate Market. The jury recognized their service, which solves a range of challenges for developers, from apartment design to determining optimal pricing per square meter based on sales targets. Additionally, their ML product for the energy sector, which predicts and reduces non-technical power losses, won the Al Russia Awards in the Revenue Growth category.
Among the latest solutions is a geo-analytical service "TelecomRadar" for the telecom industry. With its help, telecom operators can assess which districts and settlements have problems with coverage and communication quality, forecast the development of the situation and plan infrastructure modernization, as well as compare themselves with competitors and assess the prospects for expansion into new locations.
A key focus of Roman's team is ML services for the financial sector, which optimize risk management processes and build client interaction. In fact, the product "Rating" was recognized at The Retail Finance 2024 Awards, where it won the Best IT Solution category.
In addition to data analytics services, Roman has recently been focusing on developing GenAI-based business solutions. These solutions are being successfully implemented across various sectors. Their efforts span several directions, including the creation of AI assistants designed to assist the support team in processing the flow of user requests. These assistants help classify inquiries, find relevant information, and prepare responses for consultation.
An AI-powered marketing solution is actively being developed to automatically generate personalized marketing materials, tailoring product information for different audience segments and formats—such as email newsletters, banners, and push notifications. Given his extensive experience, Roman notes that this type of solution is in high demand in today’s market because it enables the creation of large volumes of personalized content while automating routine operations. This approach helps businesses increase marketing efforts without expanding the teams of designers and copywriters.
"This tool allows marketers to automatically generate product descriptions and marketing messages that will resonate with different customer groups. To date, we have launched several projects, including with a renowned coffee producer, a major bank, and a retail chain, all of which are personalizing their marketing communications using our GenAI solution."
Another high-demand area in the market is enterprise search and retrieval. A solution is being developed using Large Language Models (LLM) and Retrieval-Augmented Generation (RAG) architecture that enables a semantic search of internal corporate documents, reports, and archives. This solution allows users to quickly locate relevant data and facts that correspond to the meaning of their queries, generating clear answers with summaries of the found information and links to sources.
There are many application scenarios for using this solution. For example, in equipment monitoring and diagnostics, employees can quickly access information about incidents and their resolutions by analyzing documentation, instructions, and data from similar situations stored in the knowledge base. Without such tools, only experienced employees who have specialized in these tasks and contributed to building the knowledge base would be able to perform the task correctly. Transferring this expertise to new employees is a time-consuming and poorly structured process without the aid of these advanced tools.
As the use of AI in practical tasks grows, there is a growing need for technologies that ensure data confidentiality. For example, a solution was developed for joint AI model training using data from multiple companies, without the need to share the actual datasets. It leverages federated learning (FL), a method that allows organizations to collaborate on data-driven projects while protecting sensitive information. In federated learning, rather than exchanging raw data, only the processed results, specifically, the model weights, are shared.
"Combining the data accumulated by companies through their day-to-day operations can create significant value. For example, FMCG manufacturers often lack a complete view of their consumers, as this information is typically held by retailers. FL allows for the training of AI models using this data without the need to share sensitive information. This enables manufacturers to accurately identify customer segments while keeping proprietary data private. FL not only has the potential to drive commercial projects but can also support socially impactful initiatives. For instance, AI solutions have great potential in fields like early-stage diagnosis of heart disease and cancer. Training these models requires access to large volumes of medical research data, which is stored in medical centers and is highly sensitive. By using FL, healthcare organizations can collaborate on training models, enhancing patient outcomes by receiving effective treatment.”
IBM estimates that FL can reduce the risk of data breaches by 50% while simultaneously increasing model accuracy by 20-40% by using more data in training. Currently, the solution based on Federated Learning is being tested in collaboration with several partners across the retail, FMCG, and financial sectors.
In 2025, the demand for GenAI-based solutions has grown significantly. According to Roman, this trend is evident in the increasing number of requests from potential customers and the expanding range of services emerging in the market. Roman actively participates as an expert at international hackathons, offers guidance on the implementation of AI technologies, and observes the rise of several innovative solutions leveraging GenAI.
"AI is becoming increasingly democratized—it’s no longer necessary to be an expert in machine learning and data analysis to start using these tools. LLMs now handle many tasks through user-friendly interfaces, and their capabilities are growing tremendously. In enterprise projects, where models must be integrated into specialized workflows, companies are developing frameworks to implement RAG architectures and AI-agents. These advancements are accompanied by a growing array of integration components, allowing LLMs to be tailored to the specifics of enterprise processes and seamlessly connect with other systems."
Implementing LLMs requires companies to invest heavily in the infrastructure. However, as Roman notes, there is a growing demand to optimize the size of models without sacrificing quality. For specific tasks, models are becoming more specialized and less resource-intensive. This trend suggests that over time, the cost of implementing AI solutions will decrease, requiring less investment from companies.
Roman is confident that demand for GenAI will continue to grow. Over the last years, the number of cases on the market has increased significantly, and companies have become convinced that these tools can truly enhance their efficiency. Many now realize that AI is an assistant—one that is integrating into all areas of life.
"Speaking about the future directions of GenAI development, we can confidently say that the development of foundational models will continue. These models will become increasingly intelligent and proficient at handling a wide range of tasks, including exercises related to mathematics, logic, and programming," says Roman Styatyugin. “In terms of business applications, we can expect models to become more specialized and optimized to solve specific application challenges, without relying on hundreds of billions of parameters. Today, LLM implementation teams are actively exploring the most efficient ways to utilize these models, and I see that this work has significantly scaled up. Another important area of development for LLMs will be developing their autonomy and creating agents capable of autonomously executing chains of actions and interacting with the environment to perform necessary tasks.