Machine learning (ML) procedures are used to reduce energy consumption in production. We consider a three-tier management system of production energy consumption with ML. It is headed by the Manager who controls the Brigadier in charge of energy consumption. In turn, the Brigadier evaluates the Employee, who directly controls energy consumption in production. The Brigadier (and, to an even greater extent, the Manager) does not know exactly the actual consumption of energy required in production. Thus, the Employee can distort energy consumption in such a way to get incentive from the Brigadier. Similarly, the Brigadier can distort energy consumption to get incentive from the Manager. The three-tier mechanism for reducing energy consumption in production is proposed. Such mechanism includes ML procedures for the Manager and the Brigadier, as well as the Brigadier and the Employee evaluation procedures. It has been proven that this mechanism exploits random opportunities to reduce the energy consumption, which are unknown to higher-tier managers. Such mechanism in maintenance of freight wagons.