A large corporation often has branches and factories located in different regions. Its management is not always able to keep track of the stochastic dynamic situation in these regions, and is forced to act in the face of uncertainty. This paper examines a hierarchical stochastic discrete model of real-time expenditure control in such a corporation. At its upper tier, there is a top manager who is responsible for all expenditures. At the middle tier, there is a regional manager responsible for the expenditures of his branch, including factory expenditures, as well as overheads. At the bottom tier is the director of the factory. To manage subordinate objects in the face of uncertainty, managers can use various artificial intelligence, machine learning, and adaptation procedures. However, lower-tier managers may have interests of their own that may not align with those of the corporation as a whole. This is reflected in the model by introducing their target functions. So lower-tier managers can use the opportunities available to them to manipulate costs at uncertainty in order to increase their target functions. Based on this model, sufficient conditions are found for the synthesis of hierarchical expenditure control of a corporation, ensuring the use of stochastic possibilities of reducing its expenditure. According to these conditions, it is enough for the top manager to establish the gradation control of regional expenditures including procedures of supervised learning and self-learning. This control encourages the regional manager, first, to maintain control of factory expenditures, and, second, to minimize regional overhead costs. Such control includes adaptive planning and incentive procedures, which encourages the director to minimize factory expenditures. The application of developed control of corporation expenditures is illustrated by the example of freight car overhaul in JSC Carriage Repair Company.