Application of control theory to auction networks is important since it offers computationally efficient solutions to otherwise intractable problems. It is also of great significance due to the widespread use of auctions for resource allocation, e.g. in online advertising. But controlling agents in many large auction networks is challenging because of an uncertain and discontinuous plant. In this paper we utilize bid randomization to make the plant effectively continuous and present novel theoretical results for how to estimate the plant gain given a model uncertainty of the competitive landscape.