The appearance of suppliers on the electricity market is a demanding techno-economic process, primarily due to the high uncertainty of consumer consumption. The paper gives a brief overview of the methodology for predicting consumption throughout history, all the way to modern artificial intelligence algorithms. The key parameters and challenges for forecasting consumption (temperature, wind, seasonality, days of the week, etc.) and the differences in the consumption curves of characteristic groups of electricity consumers are described. The methodology for predicting the electricity consumption of a specific customer portfolio supplied by a supplier is described according to the rules prescribed by HEP-ODS d.o.o.. The LSTM model for predicting electricity consumption in this paper belongs to the group of advanced neural networks, and was developed using the Python programming language. The model predicts the consumption of electricity distribution and the consumption of all consumers in the distribution system equipped with an interval meter, after which the methodology of consumption forecasting for a specific customer portfolio is applied. The indicators of the accuracy of forecasting the electricity consumption for this model are the mean absolute percentage error and the mean absolute error. After calculating the electricity consumption of a specific customer portfolio, the calculation of balancing energy costs incurred by the deviation of the projected consumption from the realized electricity consumption is described.