This article is devoted to the issue of forecasting the duration of projects in an it company specializing in automation of technological and business processes. The initial data for the study are information about real projects over 10 years. The processed data set with inclusion of additional correlating characteristics is considered. This data does not have pronounced seasonal characteristics and an explicit trend sufficient for project planning, so the task of predicting the duration of work in this case is of great relevance. The study examined time series prediction models for data of a similar nature. Machine learning models were experimentally used to predict the studied data set. To improve the selection of characteristics, modern methods of optimization and search for configuration parameters were used. The experimental part of the study investigated the results of predictions obtained by the models LSTM, Auto ARIMA, Linear regression, K-averages, ARIMA, SARIMA, Prophet. The results of each model construction were analyzed on the basis of which the models were modified to achieve the necessary criteria for the task. Comparison of the results of forecasting models on the dataset under consideration revealed that LSTM neural network models are most suitable for solving the problem. The LSTM model showed an accuracy ~ 61 times that of the Prophet model based on standard error. Linear regression and K-averages also showed a satisfactory result. The information presented in this paper can be useful for forecasting modelers, data analysts and IT specialists, engineers in the field of automation and process control.