Beltel Datanomics spoke at SECR

Anna Plemyashova, Director of Corporate Strategy and Development, BELTEL a presentation on the topic “How we made a product using AI technology for FMCG. From idea to sale”. Anna shared her experience in creating an intellectual product for demand forecasting that takes into account external environmental factors.

“In order to make a decision that has business value, it is not enough to be good at mathematics and to masterly write code. You need to understand the client’s business processes, understand the problem and accurately formulate the task. To grow a product from an AI solution, you need to assess the market and its potential, understand which scenarios in the field of AI will interest customers, determine the price for the product and convince potential customers that this product is worth the requested money. Our company managed to make a number of solutions for FMCG production using machine learning algorithms, but, unfortunately, these solutions did not scale well, because they were focused on a specific customer and were built on the basis of a historical dataset of a specific production. Teaming up with industry experts in the food industry, we were able to formalize a universal task for this industry. The solution was based on the already developed algorithm for demand forecasting by machine learning methods, but the main feature of the solution is the imposition of additional sales data in retail outlets, which are marked by factors that affect demand (availability of schools, transport stops, train stations, etc.). With the help of this product, the manufacturer’s representative receive a convenient tool for working with stores — a mobile application through which they can enter data on current balances, the solution automatically generates a recommended order that the forecast model builds. The key feature of this algorithm is precisely that it takes into account the potential of the outlet and, thus, allows you to increase sales by recommending the introduction of a new product into the store or redistribution of items depending on the parameters of a particular outlet. We thought over the system of reports and visualization for different levels of users and built another predictive model for top management with an extended forecast horizon (up to a year). The purpose of the report: to share their own experience of creating a product based on AI. As from the “raw” ideas, the concept of a product was formed that is valuable for business. What are the difficulties in building AI solutions. What traps should be avoided, and what is necessary to pay attention as a guide to action”.

You can find presentations here: