IMPACTIVE AI, established in July 2021, is an AI demand forecasting company that provides the demand forecasting solution Deepflow based on advanced machine learning and deep learning models.
In the two and a half years since its establishment, the company has applied for and registered 45 domestic and international AI-related patents, and in the case of raw material price prediction, it has achieved an AI prediction accuracy of 98.6% based on an 8-week weekly forecast, saving its clients a cumulative KRW 280 billion in costs. Recently, the company won the Delivery Berlin competition in Germany, proving its technological prowess in the global market.
In this article, we met with Jeong Doo-hee, CEO of IMPACTIVE AI, to talk about the current and future of the demand forecasting AI market.
From the perspective of a user of a company, it seems that you are comparing the demand forecasting function of ERP such as SAP. However, while SAP provides demand forecasting as one of the additional functions within its huge ERP system, we are focusing solely on demand forecasting. We have been continuously accumulating capabilities and developing technical competitiveness in this field, which is why we are showing an advantage in performance.
In particular, we have dealt with various data and solved various problems in the areas of data augmentation and engineering. Existing ERP systems mainly use internal ERP data, which has the limitation of not properly reflecting changes in the market environment. We use more than 6 million cases of vast market data to more accurately predict changes in demand due to changes in the environment.
There are three main reasons. The first is the model aspect. Most AI companies and large IT teams use shared open source models. However, these models often have poor predictive power for data with complex patterns. We are independently developing advanced predictive models to solve these difficult challenges.
The second is the data aspect. We are using a vast amount of data, more than 6 million cases, and this includes various external environment data such as macroeconomic data, trend data, industry data, weather data, and industry event data. For companies that lack data, we are supplementing it with time series augmented data.
Third is customized approach for each client. Among the vast amount of data, the key variables for each client, industry, and industry are all different. We are increasing the accuracy of our predictions by automatically selecting key variables through our proprietary selection model.
The accuracy of demand forecasting is measured by the difference between the actual sales volume and the forecast value. We compare the actual sales volume with the forecast value to precisely analyze the error. For the forecasting of raw material prices such as gold, nickel, and iron ore, we have achieved a forecasting accuracy of up to 98.6% based on an 8-week weekly forecast, which is the highest in the industry.
Currently, we are working on 12 research projects. These include improving the performance of predictive models, solving the problem of data shortage, quantifying market events, and developing alternative models that can overcome the limitations of the latest transformer models. In particular, we are focusing on research to improve the limitations of transformer models in the prediction area.
In both the manufacturing and distribution industries, inventory management is a key challenge. This is because problems caused by excess or shortage of inventory lead to direct costs. In the manufacturing industry, inventory management at production facilities is important, and in the distribution industry, inventory management at storage locations such as warehouses is important. We believe that the value of our solution can be maximized in these areas where inventory costs are an issue.
Currently, manufacturing and distribution are the largest markets. However, these are industries that cover a very wide range. Recently, we have been considering expanding into high-value-added industries, and we are receiving more inquiries from the petrochemical, semiconductor, energy, and steel industries in particular. Each of these industries has its own characteristics and needs, so a customized approach is required for each industry.
In the case of inventory management, the core data required varies by industry, and the types of models applied may also vary accordingly. It is important to identify the core data that is truly important for each industry, and it is also important to consider how the work methods of the field staff and the predicted values presented by the AI model can be helpful in practice. To this end, we are conducting industrial research and interviewing field staff in parallel with technology development.
LLM and predictive AI are technically different areas. LLM is a language model that is specialized in generating something and doing creative work, while we do predictive models. However, the two technologies can be combined to create synergy.
For example, I am working on identifying and predicting market trends through LLM. I analyze and quantify consumer preferences through AI agents, or quickly capture economic changes and situations in the market and reflect them in my predictions.
However, it is not advisable to try to make predictions directly with LLM. This is because LLM is a language model and does not have built-in algorithms for mathematical calculations or time series forecasting. It is important to focus on the areas in which each technology excels and create synergies.
Currently, it has experience in supplying services to the Vietnamese, Indonesian, and American markets, and recently won the Delivery Berlin Marketplace competition in Germany, receiving recognition for its technology. When entering the global market, it is particularly focusing on services that have common needs worldwide, such as raw material price forecasting. This is because such global raw material price forecasting is a service that is needed not only by Korean companies but also by companies in Europe, the United States, and Japan.
First of all, we are currently focusing on upgrading our services and improving their accuracy. It is more important than anything else to improve our services while better understanding and communicating with our customers. We are also preparing to enter overseas markets, and we plan to expand our services to focus on those that are commonly needed in the global market, such as demand forecasting and raw material price forecasting.
However, the most important goal is the growth of our internal members. The growth of employees leads to better services, which in turn leads to the growth of customers. We are actively supporting the growth of our members through apprenticeship training between seniors and juniors, training by inviting external experts, and support for personalized training. We believe this will create a virtuous cycle that leads to the growth of the company.
Through this interview, we were able to identify three key factors that have enabled IMPACTIVE AI to establish itself as a demand forecasting company.
First, it is the focus and expertise in one area: demand forecasting. Unlike the big ERP companies that treat demand forecasting as part of their comprehensive services, IMPACTIVE AI has achieved a high level of performance advantage by focusing solely on this area. This has been made possible by technological differentiation based on expertise, including the development of a unique forecasting model, securing more than 6 million data sets, and selecting industry-specific variables.
The second is a deep understanding of the market and customers. While focusing on the core task of inventory management in the manufacturing and distribution industries, you are providing practical value through a customized approach that takes into account the characteristics of each industry. Rather than simply providing technology, you have developed solutions in a way that understands the work processes of those in charge of the field and improves them.
Finally, we can mention our commitment to continuous innovation and growth. IMPACTIVE AI is constantly taking on new challenges, including technological innovation through 12 research projects, seeking synergy with LLM, and entering the global market.
The foundation of this innovation lies in the growth of our members. We are creating a virtuous cycle of growth by cultivating the capabilities of our organization through apprenticeship-style training between senior and junior members and active investment in education.
IMPACTIVE AI is establishing itself as a leader in the demand forecasting market through three pillars: technology based on expertise, customer-oriented market approach, and continuous pursuit of innovation. It is expected to continue to achieve meaningful results in the global market based on this competitiveness.